2018
High-Performance Computing Center
Stuttgart
ANNUAL REPORT
ENERGY
HEALTH
CLIMATE
MOBILITY
PHILOSOPHY
2018
HLRS ANNUAL REPORT The High-Performance Computing Center Stuttgart (HLRS) was established in 1996 as the first German national high-performance computing (HPC) center. As a research institution affiliated with the University of Stuttgart and a founding member of the Gauss Centre for Supercomputing, HLRS provides HPC services to academic users and industry. We operate leading-edge HPC systems, provide training in HPC programming and simulation, and conduct research to address key problems facing the future of supercomputing. Among our areas of expertise are parallel programming, numerical methods for HPC, visualization, grid and cloud computing concepts, data analytics, and artificial intelligence. Our system users conduct research across a wide range of scientific disciplines, with an emphasis on computational engineering and applied science.
Director‘s Welcome Grußwort
2 /3
HLRS ANNUAL REPORT 2018
Welcome to the HLRS 2018 Annual Report. We are
Wir freuen uns, mit Ihnen auf die zahlreichen Erfolge der
pleased to present this look back at our activities over
letzten zwölf Monate am HLRS zurückzublicken.
the past 12 months, which were marked by many suc-
An oberster Stelle der wichtigsten Ereignisse steht
cesses.
sicherlich die Beschaffung unseres neuen Höchstleis-
At the top of our list of achievements was certainly our
tungsrechners. Nach einer intensiven, einjährigen Aus-
procurement of a new high-performance computer.
schreibungsphase, an der führende Hardware-Anbieter
After an intense, yearlong competition among leading
teilnahmen, unterschrieb das HLRS im November 2018
hardware vendors, HLRS signed a contract with Hewlett
einen Vertrag mit Hewlett Packard Enterprise (HPE), um
Packard Enterprise (HPE) in November to build Hawk,
„Hawk“ — unseren Höchstleistungsrechner der nächs-
our forthcoming next-generation supercomputer. As
ten Generation — in Betrieb zu nehmen. Als Teil dieser
part of this collaboration, HLRS and HPE will also work
Zusammenarbeit werden das HLRS und HPE gemein-
together on improving services, optimizing codes, and
sam Services ausbauen, Codes optimieren und neue
developing new types of high-performance computing
Anwendungen für Maschinelles Lernen und Künstliche
(HPC) applications for machine learning and artificial
Intelligenz vorantreiben. Wir freuen uns darauf, unse-
intelligence. In the coming years we look forward to
ren Nutzern in den kommenden Jahren deutlich leis-
being able to offer our users a much more powerful
tungsfähigere Ressourcen für das Hoch- und Höchst-
supercomputing resource and will be well positioned to
leistungsrechnen (HPC) anbieten zu können und damit
lead as HPC technologies continue to advance.
unsere führende Position für die Weiterentwicklung
HLRS also successfully participated in the launch of
zukünftiger HPC-Technologien auszubauen.
several new EU-funded projects, including new Euro-
Das HLRS konnte seine Arbeit in zahlreichen neuen,
pean Centres of Excellence (CoE’s) in high-performance
EU-geförderten Projekten beginnen — darunter vier
computing that will form the backbone of the European
neue europäische Exzellenzzentren (CoEs) für HPC-An-
HPC strategy. As a member of multiple new CoE’s cre-
wendungen, die das Rückgrat der europäischen
ated in 2018 — EXCELLERAT, HIDALGO, CHEESE, and
HPC-Strategie bilden sollen. Als Mitglied von EXCEL-
POP2 — HLRS is involved in four of the nine European
LERAT, HIDALGO, CHEESE und POP2 ist das HLRS in
centers. As project leader of the new European Centre
vier der neun in 2018 gestarteten europäischen Exzel-
of Excellence for Engineering Applications (EXCEL-
lenzzentren vertreten. Als koordinierender Partner des
LERAT), HLRS will strengthen its position as Europe’s
neuen European Center of Excellence for Engineering
leading center for simulation in engineering, a critical
Applications (EXCELLERAT), wird das HLRS seine Posi-
tool for high-tech innovation. Our support for the com-
tion als führendes europäisches Zentrum für Simula-
putational engineering and applied science communi-
tion in den Ingenieurwissenschaften stärken. Unsere
ties is the focus of this annual report’s spotlight article.
Unterstützung für die computergestützte Ingenieur-
Also important in 2018 was the positive evaluation
und angewandten Wissenschaften ist ein Schwerpunkt
of several key research projects that HLRS leads or
dieses Jahresberichts.
participates in as a major partner. In September the
occasion to talk with us for this annual report about the
Ein weiteres Ereignis in 2018 war die positive Bewertung
und Forschung arbeiten wird, um technischen Heraus-
Deutsche Forschungsgemeinschaft (DFG, or German
economic and scientific benefits of networking the Ger-
einiger wichtiger Forschungsprojekte, die das HLRS lei-
forderungen aus den Bereichen Medien und Kunst zu
Research Foundation) once again recognized the Uni-
man automotive industry.
tet oder daran als bedeutender Partner teilnimmt. Die
begegnen. Das MSC ist der neueste Bestandteil unse-
versity of Stuttgart’s Cluster of Excellence in Simula-
In terms of key performance indicators HLRS can
Deutsche Forschungsgemeinschaft hat zum Beispiel
rer industriellen Strategie, die mit dem Automotive
tion Technology (SimTech) as a German national center
proudly look back on a very successful year. Our third-
im September das Exzellenzcluster SimTech der Uni-
Simulation Center Stuttgart (asc(s) begann. In diesem
of excellence for its proposal to expand our focus on
party funding saw an increase over previous years, our
versität Stuttgart wieder als deutsches Exzellenzzen-
Jahr feierte das asc(s sein zehnjähriges Bestehen und
data-integrated simulation science. As the prominent
income from industrial HPC users was again at a high
trum anerkannt, das seinen Fokus auf daten-integrierte
demonstrierte somit, wie nachhaltig diese Strategie
facility for high-performance computing at the Univer-
level, and the number of people benefiting from our
Simulationswissenschaft legen wird. Als herausra-
sein kann. Wir danken asc(s Geschäftsführer Alexander
sity, HLRS looks forward to the coming seven years, in
training and education activities remained strong.
gende Einrichtung für das Höchstleistungsrechnen an
Walser dafür, dass er für diesen Jahresbericht mit uns
which we will work closely with our colleagues here to
Users of our HPC systems also had a very productive
der Universität Stuttgart, freut sich das HLRS auf die
über die Vorteile des Networking innerhalb der Auto-
push the limits of using big data in simulation.
year. This annual report showcases some examples of
nächsten sieben Jahre, in denen wir in enger Zusam-
mobilindustrie sprach.
HLRS recognizes that the future of high-performance
our users’ research, including a new sound prediction
menarbeit mit unseren Kollegen die Grenzen der Simu-
In Bezug auf unsere Leistungskennzahlen kann das
computing lies not just in new technologies, but also
model that could help reduce machinery noise, appli-
lation unter Benutzung von Big Data erweitern werden.
HLRS stolz auf ein erfolgreiches Jahr zurückblicken.
in addressing humanity’s greatest challenges. Con-
cations of HPC and machine learning to make power
Für uns liegt die Zukunft des Höchstleistungsrechnens
Unsere Drittmitteleinnahmen stiegen über die letz-
sidering important relationships among supercom-
plants more efficient, and a computational approach
nicht nur in der Entwicklung von neuen Technologien,
ten Jahre hinweg, unser Einkommen aus industrieller
puting, society, and politics we created a new Socio-
that is helping to point the way toward new semicon-
sondern auch darin, die größten Herausforderungen
HPC-Nutzung war wieder auf einem hohen Niveau und
political Advisory Board, which held its first meeting
ductor technologies. Such applications show how sim-
der Menschheit zu adressieren. Unter Berücksichtigung
die Anzahl der Personen, die von unserem Weiterbil-
this year. The gathering identified many topics that
ulation is making important contributions in areas that
wichtiger Zusammenhänge zwischen dem Höchst-
dungsprogramm und unseren Vorlesungen profitieren,
will feed back into our research and enable HLRS to
are critical for our wellbeing.
leistungsrechnen, der Politik und der Gesellschaft
blieb hoch.
remain focused on society’s needs and requirements.
With this annual report we express our heartfelt thanks
haben wir einen neuen Gesellschaftspolitischen Beirat
Auch die Benutzer unseres HPC-Systems hatten ein
Throughout the year we were also glad to welcome
to our supporters and funders, who have made our suc-
geschaffen, der 2018 zum ersten Mal zusammengetrof-
sehr produktives Jahr. Dieser Jahresbericht verdeutlicht
prominent politicians from Germany and Europe to
cesses in 2018 possible. At the same time, we look for-
fen ist. Die Sitzung hat mehrere Themen identifiziert,
anhand einiger Beispiele aus der Forschung unserer
HLRS to discuss the future of high-performance com-
ward to tackling new challenges in 2019.
die unsere Forschung voranbringen können und es
Nutzer, wie wichtig Simulation für unser Wohlergehen
dem HLRS ermöglichen werden, sich stärker als bis-
ist. Dazu gehören ein Geräusch-Vorhersage-Modell,
her auf die Bedürfnisse und Anforderungen der Gesell-
das helfen könnte, den Lärm von Maschinengeräu-
HLRS’s focus on HPC training grew in strength this year
schaft zu konzentrieren. Dieses Jahr haben wir uns
schen zu reduzieren; eine Machine Learning-Anwen-
with the completion of the first course offered by the
auch darüber gefreut, viele prominente Politikerinnen
dung, die Kraftwerke effizienter machen könnte; sowie
Supercomputing-Akademie. Its “blended learning” cur-
und Politiker aus Deutschland und Europa am HLRS
einen Berechnungsansatz, der den Weg zu einer neuen
riculum is designed to address the needs of HPC users
willkommen zu heißen, um die Zukunft des Höchstleis-
Halbleitertechnologie ebnet.
puting and how its potential could be maximized to With best regards,
benefit society.
in industry, and complements HLRS’s existing state-
Prof. Dr.-Ing. Dr. h.c. Dr. h.c. Prof. E.h. Michael M. Resch
tungsrechnens und dessen Nutzen für die Gesellschaft
Mit diesem Jahresbericht bringen wir auch unseren
of-the-art training program. Trainers and trainees were
Director, HLRS
zu diskutieren.
Dank an unsere Unterstützer und Förderer zum Aus-
overwhelmingly positive about the Supercomputing-
Der Schwerpunkt des HLRS auf die HPC-Weiterbildung
druck, die uns diese Erfolge in 2018 ermöglicht haben.
Akademie’s first test-run, and we look forward to
verstärkte sich in diesem Jahr durch den Abschluss
Gleichzeitig freuen wir uns darauf, die neuen Herausfor-
watching this innovative program grow in the coming
des ersten Moduls der Supercomputing-Akademie. Ihr
derungen 2019 in Angriff zu nehmen.
years.
Kursplan im „Blended Learning“-Format entspricht den
Another highlight of 2018 was the inauguration of the
Bedürfnissen der HPC-Nutzer aus der Industrie und
Media Solution Center Baden-Württemberg (MSC),
ergänzt das wegweisende Schulungsprogramm des
through which HLRS will work closely with partners
HLRS. Nach den vielen positiven Reaktionen der Aus-
from industry and academia to find solutions to chal-
bildenden und Teilnehmenden auf den ersten Testlauf
lenges facing the fields of media and the arts. The MSC
der Supercomputing-Akademie freuen wir uns darauf,
is the latest product of our industrial strategy, which
dieses innovative Programm in den nächsten Jahren
began with the Automotive Simulation Center Stuttgart
wachsen zu sehen.
(asc(s). This year the asc(s celebrated its 10th anniver-
Ein weiteres Highlight von 2018 war die Gründung des
sary, showing just how sustainable our strategy can be.
Media Solution Centers Baden-Württemberg (MSC), in
We thank asc(s director Alexander Walser for using the
dem das HLRS eng mit unseren Partnern aus Industrie
4 /5
HLRS ANNUAL REPORT 2018
8
Spotlight
9
Engineering the Future
14
News & Highlights
15
News in Brief
22
New European Centers of Excellence Funded
23
HLRS Visualization Supports Stuttgart 21 Construction
24
Supercomputing-Akademie Completes First Course
25
Sociopolitical Advisory Board to Help Guide HLRS Vision
26
SimTech Named National Center of Excellence
27
Media Solution Center to Promote Innovation in Film and Digital Art
28
EXCELLERAT Will Bring HPC Applications to Engineering Industry
29
Building a Sustainable HPC Infrastructure
30 Interview: Networking the German Auto Industry 33
Student Dives into Data to Predict Train Delays
36
Enhanced User Support Improves Performance of HPC Systems
38
EuxDat: Taking Agriculture into the Cloud
39
PhD Graduates 2018
IMPRINT
40 Looking Inside Simulation‘s ”Black Box“
6 /7
HLRS ANNUAL REPORT 2018
34
HLRS by the Numbers
42
User Research
43
Supercomputer Enables Sound Prediction Model for Noise Control
46
Using Computational Chemistry to Investigate New Semiconductor Technologies
48
Simulation and Machine Learning Could Make Power Plants More Efficient
50 Selected User Publications
56
About HLRS
57
Inside Our Computing Room
59
User Profile
60 Third-Party Funded Research Projects 66
HPC Training Courses in 2018
68
Workshops and Conferences 2018
69 Structure 70
Divisions and Departments
Engineering the Future
HLRS has decades of experience supporting computational engineering for academic and industrial research. With recent expansions in its infrastructure and services, it is now laying a foundation that will enable engineers and applied
SPOTLIGHT
© Mathis Bode, Institute for Combustion Technology, RWTH Aachen
scientists to take full advantage of the next generation of faster supercomputers.
8 /9
HLRS ANNUAL REPORT 2018
Located in the capital of the state of Baden-Württem-
industrial research and development — for example, in
berg, the High-Performance Computing Center Stutt-
the automotive, energy, and aerospace industries. In
gart (HLRS) sits at the epicenter of one of Germany’s
fact, approximately 10% of HLRS computing hours are
most dynamic regions for high-tech engineering. The
dedicated to industrial users, a figure that stands out as
region is not only home to prominent international con-
particularly noteworthy in Germany’s academic super-
cerns such as Daimler, Bosch, and Porsche, but also
computing landscape.
to a large and thriving ecosystem of small and medi-
Bridging the academic and industrial worlds in this way
um-sized companies with a long tradition of manufac-
has enabled HLRS to develop expertise in applying
turing high-quality mechanical, electronic, and chemi-
high-performance computing (HPC) to many kinds of
cal products. Together, these communities have been a
real-world computational engineering problems. This
major engine of Germany’s economic prosperity.
knowledge has, in turn, guided its strategy for growth.
This regional strength in engineering has also played an
HLRS’s goal of providing resources that optimally serve
important role in shaping HLRS’s evolution. Based on
the needs of computational engineers has been a key
the campus of the University of Stuttgart — itself a cen-
driver of investments in its new computing infrastruc-
ter for research and higher education with an empha-
ture, professional training offerings, and in-house com-
sis on engineering and the applied sciences — HLRS
puter science research. As HLRS lays the groundwork
has for decades maintained a state-of-the-art infra-
for ever-larger supercomputers in the future, the center
structure for computational simulation, visualization,
is also focused on providing the tools and services the
and data analytics. Such tools have become essential
engineering community will need to take full advantage
for creating models of natural phenomena that are so
of these resources as they become available.
complex or that occur at such small scales that they are difficult to study in any other way. In engineering, the
HPC improves automotive manufacturing
design of many precision products for which Germany
For Stefan Hildenbrand at chemicals manufacturer
is renowned has become unimaginable without such
Pfinder KG, HLRS has become an important partner,
computational models.
enabling the company to develop new services for cli-
HLRS has long provided essential supercomputing
ents in the automotive industry.
resources for scientists in disciplines such as com-
Pfinder supplies wax-based anticorrosive coatings that
putational fluid dynamics, physics, chemistry, and
are typically applied by assembly line robots during car
materials science, among other fields. This includes
manufacture. As team leader for digital engineering,
not only academic researchers, but also scientists in
Hildenbrand performs computational simulations to
Sample simulation of the application of wax to a sill. The colors indicate the thickness of coverage.
advance applications in energy, climate, mobility, and
Writing software for large-scale parallel computing sys-
health. The 5,000-node Hawk system will have a theo-
tems like Hazel Hen or Hawk is much more complex
retical peak performance of 24 petaFLOPs.
than running the same algorithm on a desktop com-
An important feature of Hawk is that its technical spec-
puter or a small computing cluster. Indeed, developing
ifications were chosen with the specific needs of com-
and implementing effective and efficient codes for HPC
help optimize this process. Spe-
putational engineers in mind. Hawk’s architecture is
systems requires specialized knowledge, and will be
cifically, he uses OpenFOAM, a
based on chip manufacturer AMD’s next-generation
especially important as supercomputers approach the
widely used software package for
EPYC processor, code-named Rome. HLRS picked the
exascale, the next major plateau in computing power
computational fluid dynamics, to predict
Rome processors because they utilize a memory sub-
over that of current petascale systems. Optimizing
how the coatings will behave when injected by a nozzle,
Keeping up with changes in HPC technologies would
system that makes them particularly well suited for
codes for such extremely parallelized systems will be
the paths the droplets will follow through the air, and
be impossible for a small company.
simulation applications in fields such as computational
necessary to accelerate individual computational sci-
how they will form films on surfaces. These models are
Having access to this kind of computing power became
fluid dynamics, molecular dynamics, and other research
entists’ research and to make the operation of super-
used to optimize the amount of fluid that needs to be
particularly important for Hildenbrand in 2018, when
areas in which many of HLRS’s users are engaged. In
computers more efficient — this will, in turn, help save
sprayed for a specific application, to identify the best
for the first time his team simulated the application of
addition, the AMD EPYC processors complement the
energy and make HPC resources available to the largest
nozzle layout, and to ensure they are fully protective
Pfinder coatings not just for one component, but for an
use of competing processors at the other two centers
number of potential users.
and durable.
entire automobile simultaneously. This jump in scale
in the Gauss Centre for Supercomputing — the alliance
For many years, HLRS has recognized that the future
Pfinder has a small parallel computing system in-house,
required a leap in computing power that was far beyond
of Germany’s three national supercomputing centers.
of high-performance computing lies in developing and
but when complex questions arise or when results need
the reach of its own system. He anticipates continuing
The choice supports GCS’s goal of offering its users a
increasing access to state-of-the-art codes capable of
to be delivered quickly, Hildenbrand turns to HLRS.
to utilize the HLRS supercomputer in the future, as this
diverse set of computing architectures.
scaling to ever-larger systems. For these reasons, the
“When we are deep in a project for a client that requires
ability to model entire systems could offer a new mar-
“We are excited that Hawk constitutes a sizable
center has undertaken several projects aimed at prepar-
a large-scale simulation,” Hildenbrand explained, “we
ket opportunity for his firm to provide complete coating
increase in the performance of our flagship supercom-
ing computational engineers for the future.
can’t wait four weeks to run the computation on our own
application solutions.
puting system,” said Prof. Dr. Michael M. Resch, Director of HLRS. “But the real winners will be our user commu-
Bringing industrial engineering up to scale
nity of computational engineers in academic research
In 2018, for example, HLRS became the lead coordi-
we have here. Instead of waiting for up to 4 weeks, this
Hawk to offer new capabilities for industrial production
and industry who will benefit from the ability to run
nating center of a new Horizon 2020-funded research
can deliver a result in as little as 3 to 4 days.”
Users of HLRS’s HPC systems will soon have even
much more complex simulations.”
program called EXCELLERAT (European Centre of
Being able to call on HLRS also has other advantages
more power at their fingertips. In November 2018, HLRS
system. HLRS makes it possible for us to quickly run a job on many hundreds of cores as opposed to on the 16
Excellence for Engineering Applications). A collabora-
over installing its own larger HPC system. Because
and Hewlett Packard Enterprise announced a joint col-
Getting to the next level
Pfinder’s computational workload fluctuates widely, it
laboration to build a next-generation supercomputer
There is no doubt that larger supercomputers enable
EXCELLERAT will provide the HPC and engineering
would not make economic sense to install a large sys-
that will be 3.5 times faster than Hazel Hen, HLRS’s
larger simulations and increase the number of core
communities access to scalable applications and
tem of its own, only to have it sit idle for multiple months
current flagship system. The upcoming supercomputer,
hours available for the research community. At the
facilitate their availability through technology transfer.
of the year. Moreover, Hildenbrand can be confident
called Hawk, will be the fastest in the world for indus-
same time, however, it has become clear that build-
(See page 28 to learn more about EXCELLERAT.)
that HLRS uses reliable, state-of-the-art hardware that
trial production, powering computational engineering
ing hardware alone is not enough to meet the growing
Recognizing the need for such resources and expertise,
continually grows in power as HPC systems evolve.
and research across science and industrial fields to
needs of computational engineering.
companies including Porsche, Ansys, and Festo have
10 / 11
HLRS ANNUAL REPORT 2018
tive network including 13 partners from across Europe,
Representatives of the University of Stuttgart and Hewlett Packard Enterprise celebrated the contract signing for the new Hawk supercomputer.
expressed support for EXCELLERAT. Aircraft manufac-
In July, the Supercomputing-Akademie — which is sup-
turer Airbus has also been an early participant in the
ported by the Baden-Württemberg Ministry for Social
project, as EXCELLERAT will enable academic knowl-
Affairs and Integration through the European Social
edge to be connected with industrial needs.
Fund, and by the Baden-Württemberg Ministry of Sci-
In addition to EXCELLERAT, HLRS participates in proj-
ence, Research, and Art — completed its first 15-week
ects focused on identifying and addressing opportuni-
course on parallel programming. In 2019, it will begin
ties for supercomputing in specific industries. For more
offering a second course, focusing on simulation. In
than 10 years it has been a key partner in the Automo-
the coming years, the curriculum will grow to include
tive Solution Center Stuttgart (see page 30), an alliance
additional modules on visualization; performance
of companies that use simulation in automotive devel-
optimization; cluster, cloud, and high-performance
opment and design. In 2018, it also joined the Hoch-
computing; ecology and economics of supercomput-
schule der Medien und the Center for Art and Media in
ing; and data management — topics that are of great
Karlsruhe (ZKM) in founding the Media Solution Cen-
interest to system administrators, programmers, and
ter, an alliance of organizations interested in explor-
HPC software users in industry.
ing opportunities for high-performance computing in
The Supercomputing-Akademie is just one compo-
the media arts. (see page 27). In coming years, HLRS
nent of an ongoing dialogue between HLRS and com-
intends to implement this solution center model to
putational engineers in industry. Working with the
focus on increasing access to high-performance com-
nonprofit organization SICOS-BW, the center has for
puting for health research and other areas.
the last two years also organized the Industrial HPC User Roundtable (iHURT), a full-day event designed to
Supercomputing-Akademie trains HPC experts in industry
promote information exchange in industrial R&D. This
Providing HPC training has long been a key activity for
important challenges that HPC users in industry face,
HLRS. In 2018, however, the center opened a new con-
and will continue to guide its development of new infra-
tinuing education program designed with the needs of
structure and services in the coming years.
dialogue also enables HLRS to understand the most
Aron Precht (VP Sales, DACH & Russia, Hewlett Packard Enterprise)
Heiko Meyer (General Manager und Vice President, Enterprise Group, Hewlett-Packard GmbH) Jan Gerken (Chancellor, University of Stuttgart) Michael M. Resch (Director, HLRS)
computational engineers in industry in mind. To make these training activities as accessible as pos-
Putting all the pieces together
Installing the next-generation Hawk supercomputer
sible, the new program, called the Supercomputing-
In the coming decade, simulation, visualization, data
in 2019 will be one important step toward this goal.
Akademie, uses a “blended learning” format. This
analytics, and machine learning will remain critical for
Additionally, HLRS’s ambitious training program and
means that only a small part of the training actually
research and development, and will play important
its outreach and training programs focused on indus-
takes place in a classroom. Approximately 80% of a
roles in addressing some of our greatest challenges.
try are important parts of this effort. Through such
participant’s time spent learning can be done online,
By identifying and focusing on the needs of engineers
activities, HLRS is positioned to be a key partner for
making it easy for someone with a full-time job to join in
and applied scientists, HLRS aims to provide the wid-
innovation in Baden-Württemberg, in Germany, in
the course alongside his or her normal responsibilities.
est possible access to such tools.
Europe, and beyond.
12 / 13
HLRS ANNUAL REPORT 2018
(CW)
Günther Oettinger
HLRS Director Michael Resch
NEWS HIGHLIGHTS
Commissioner Oettinger Visits HLRS
14 / 15
HLRS ANNUAL REPORT 2018
As a longtime supporter of HLRS in his previous role as minister president of the state of Baden-Württemberg, European Commissioner for Budget and Human Resources Günther Oettinger is acquainted with the needs and challenges facing high-performance computing. On May 17 Oettinger and HLRS Director Michael Resch discussed the importance of a European strategy for high-performance computing capable of providing European scientists access to state-of-the-art technology. One topic of discussion was how European HPC will be supported through the EuroHPC Joint Undertaking (EuroHPC), which plans to fund two pre-exayears. Both Oettinger and Resch expressed a need to
HLRS Supports Crime Investigation with 3D Visualization
make high-performance computing available and pro-
An October 8, 2018, episode of the ARD television pro-
ductive for both scientific applications and industrial
gram Kriminalreport focused on how 3D modelling and
research.
simulation can help solve complex crimes. The episode
flop systems and two exaflop systems in the coming
(LB)
detailed methods the State Office of Criminal Investigations, Baden-Württemberg (LKA) uses in analyzing 3D images of crime scenes, as well as HLRS’s recent advances in developing such approaches. HLRS Visualization Department leader Uwe Wössner explained that investigators can use visualization environments to accurately study bullet trajectories or conduct autopsies, among other aspects of crime scene investigation. Such methods can provide critical forensic evidence that would be hard to gather using other methods. In recent years, HLRS has been collaborating with the LKA to develop new methods for 3D-digital reconstruction of crime scenes to study them even more effectively.
(EG)
Golden Spike Awards Recognize Outstanding Research Each year, the Results and Review Workshop offers an opportunity for HLRS supercomputer users to discuss their research and best practices in HPC. On October 4-5, the 21st annual workshop featured 41 user projects in 22 scientific talks and 19 posters. The event showcased research from computational fluid dynamics, climate research, computer science, chemistry, physics, and biology. At the workshop HLRS also presented Golden Spike Awards to representatives of three outstanding projects: Mathis Bode of RWTH Aachen (“Towards clean propulsion with synthetic fuels”), Travis Jones of the Fritz Haber Institute of the Max Planck Society (“Sulfur in ethylene epoxidation on sil-
Girls’ Day 2018
ver”), and Christoph Wenzel of the University of Stutt-
Girls’ Day is an annual event that takes place across
gart (“DNS of compressible turbulent boundary layers
Germany to encourage girls to pursue careers in sci-
with adverse pressure gradients”).
ence, engineering, and information technology. As
HPC User Forum Considers State of European Supercomputing
the meeting featured presentations by Leonardo Flores (European Commission) on the EuroHPC Joint Under-
part of the 2018 festival, ten students between the
Organized by HPC industry consultants Hyperion
taking; by Hyperion’s Earl Joseph and Steve Conway
ages of ten and twelve visited HLRS to learn about
Research, the 2018 HPC User Forum took place at HLRS
on trends in the HPC industry; by leading HPC hard-
the world of high-performance computing. Visitors
on October 1-2. It offered an intimate venue for promoting
ware manufacturers about technical challenges they
received a hands-on introduction to computing hard-
interactions among HPC users and technology suppli-
face; by industrial HPC users focusing on autonomous
ware, a tour of the HLRS computing room, and training
ers, and provided insights into how supercomputers and
driving; and by SICOS-BW, the Fortissimo Project, and
in using the programming language Scratch. On their
their uses are evolving. In addition to updates from HLRS
the EXCELLERAT project looking at unique challenges
last stop, they hopped into a flight simulator in HLRS’s
and its partners in the Gauss Centre for Supercomputing,
faced by HPC users in industry.
3D visualization room to paraglide virtually above the
(CW)
(CW)
Black Forest. After learning to “fly,” each girl obtained an HLRS pilot’s license. Girls’ Day offered the students
Stuttgart and Wuhan Sign Collaboration Agreement
enjoyable experiences that would raise their interest in
On April 3, HLRS took its next step in strengthening
would take home good memories.
technology; at the end of the day all agreed that they (AL)
its scientific cooperation in Asia. In a short ceremony in Wuhan, China, Supercomputing Center of Wuhan University Director Dengyi Zhang and HLRS Director
Illustris Project Featured on Postage Stamp
project released five simultaneous papers document-
Michael Resch signed an agreement pledging collabo-
In December the Deutsche Post announced the release
ing results of the first two IllustrisTNG simulations. The
ration for the next three years. HLRS and the Supercom-
of a new postage stamp design featuring the Illustris
project is a two-time winner of the Golden Spike Award,
puting Center at Wuhan University plan to exchange
project. The international scientific collaboration, led
an annual prize recognizing outstanding research using
scientists and to focus on key research topics in
by Volker Springel at the Heidelberg Institute for The-
HLRS computing resources.
high-performance computing. Both sides will also share
oretical Studies in partnership with researchers at the
experience in installing large-scale computing systems.
Massachusetts Institute of Technology, Harvard Uni-
In addition to its cooperation with Wuhan University,
versity, and the Max Planck Institute for Astrophysics,
HLRS maintains agreements with many other HPC cen-
has created the world’s largest cosmological simulation
ters in Asia, Europe, Russia, and the United States with
of galaxy formation. Currently, the Illustris team is working with HLRS to run its next-generation simulations
the goal of promoting the exchange of HPC knowledge and expertise.
(CW)
16 / 17
Dengyi Zhang
HLRS ANNUAL REPORT 2018
Michael M. Resch
on Hazel Hen, a project called IllustrisTNG. In 2018 the
(CW)
Artist Visualizes Air Quality Problems in Stuttgart
optimal platform for users whose computation relies
For the Drehmoment Produktionskunst-Festival, orga-
on millions of small data files. Because the cloud-
nized by the KulturRegion Stuttgart, artist Michael Saup
based platform is connected to Baden-Württemberg’s
collaborated with HLRS to make the city’s notorious air
BelWü high-speed digital network, it also offers the
pollution visible. His virtual reality project, called Staub
possibility for users to access their data directly from
(Dust), interpreted and visualized sensor measure-
their home institutions, speeding their research. In
ments of particulate emissions, integrating them with
addition, the scalability and affordability of the new
open-source street imagery and real-time data about
platform will permit users to save data on the system
traffic conditions. Visitors to the installation could wear
© FSE/Hikisch Gergely
for the duration of their project allocations, rather than 60 days as required on Lustre. Moving suitable proj-
ment. “It is inspirational to work with people who use
New Cloud-Based File System Offers Users Greater Flexibility
similar tools but have a completely different sense of
To enhance its user service, HLRS implemented a new
solution but also reduces bottlenecks on the Lustre
what to do with them,” Saup said. “Exchanging ideas
Quobyte cloud-based file system. The system comple-
system so that it can be dedicated to projects for which
with people working in computing ... can lead to new
ments HLRS’s existing Lustre file system, offering an
it is best suited.
a VR headset to explore this otherwise hidden environ-
ects to the Quobyte system not only offers a cheaper
(CW)
narratives about the future and what it could look like.” The installation was shown in St. Maria’s Church in
Stuttgart Wins Formula SAE Michigan for Third Consecutive Year
Stuttgart for two weeks in October.
(CW)
Minister of Hungary Visits to Discuss HPC in Europe, Industry On October 19, Hungarian Minister for Innovation and
of Stuttgart won the 2018 Formula SAE Michigan, the
Technology Prof. Dr. Laszlo Palkovics and Hungarian
unofficial world cup of the international Formula Stu-
Foreign Trade Attaché Dr. David Bencsik visited HLRS.
dent network. It was the third consecutive year that a
Following an introduction to HLRS research and indus-
team from the University of Stuttgart won the contest.
trial activities — focusing on collaboration between
In addition to constructing a racecar and testing it on
HLRS and the automotive engineering industry, par-
the track, team members presented it to experts from
ticularly with respect to research on autonomous driv-
the motorsport, automotive, and aerospace industries.
ing vehicles — Minister Palkovics and HLRS Director
Teams are judged on the car’s performance, design,
Michael Resch discussed the current high-performance
construction, cost, and their sales presentations. HLRS
computing landscape in Europe, including the EuroHPC
has been an official supporter of the University of Stutt-
joint undertaking and the Partnership for Advanced
Joseph Schuchart Wins Best Paper Award at IWOMP 2018
gart team since 2016, sponsoring computing hours for
Computing in Europe. HLRS has existing collaborations
At the International Workshop on OpenMP in Barce-
with the University of Pécs and Hungarian Academy of
lona, HLRS scientist Joseph Schuchart was recognized
Science, and both sides agreed to continue searching
for his investigation of taskyield, an important piece
for new avenues for cooperation in autonomous driving
of software for scheduling parallel computing tasks.
and high-performance computing.
Schuchart examined several potential implementations
computational fluid dynamics simulations to improve the car’s aerodynamics. “The support from HLRS is one of the cornerstones for developing our race cars,” said team member Johannes Burgbacher.
© Michael Saup
The Formula Student Racing Team of the University
(AL)
(EG)
Joseph Schuchart
of taskyield, comparing their correctness and performance. Because it can be difficult to know how task-
Third Annual German-Russian Workshop
of Applied Mathematics, Russian Academy of Sci-
yield prioritizes tasks, he and his colleagues presented
On April 23-26, investigators from academic insti-
ences) and Prof. Dr.Sci. Andrei Klemeshev (Rector,
a test that can detect which variant of taskyield is used
tutes across Russia and Germany met in Kalinin-
Immanuel Kant Baltic Federal University), the packed
in different OpenMP compilers and runtimes. “Our
grad to forge closer ties between the two research
three-day conference explored a wide range of topics
motivation is to help developers and users by informing
communities and identify opportunities for potential
related to HPC and its applications. Talks focused on
them about the weak spots of the taskyield feature and
shared research projects. The meeting was the third
scientific challenges in fields such as computational
providing a tool that offers insights into the inner work-
in a series of annual workshops organized to promote
fluid dynamics, cosmology, and structural mechanics,
ings of OpenMP implementations,” says Schuchart. In
international dialogue. Following welcoming remarks
and on cutting-edge methods in high-performance
by Prof. Dr.Sci. Boris Chetverushkin (Keldysh Institute
computing.
18 / 19
HLRS ANNUAL REPORT 2018
(CW)
the future he intends to explore other types of applicaMichael Resch
Laszlo Palkovics
tions and add to the research on OpenMP tasks.
(AL)
Run for Digitalization On July 27, digitalization experts ran around the Uni-
New Book: High Performance Computing in Science and Engineering ‘17
versity of Stuttgart’s Vaihingen campus — literally.
Including findings from researchers using HLRS
Held with special guests Thomas Strobl, Baden-Würt-
high-performance computing systems in 2017, this
temberg’s Minister of the Interior, Digitalization and
book presents the state-of-the-art in supercomputer
Migration, and ultramarathon runner Jürgen Mennel,
simulation. The reports cover all fields of computational
the event, titled “Digitalisierung: Läuft,” raised aware-
science and engineering, ranging from computational
ness of Baden-Württemberg as a location of digital
fluid dynamics to computational physics, and from
expertise. During a press conference, HLRS Director
chemistry to computer science, with a special emphasis
HLRS Opens its Doors on Tag der Wissenschaft
Michael Resch emphasized that the state has the ideal
on industrially relevant applications. Presenting find-
On June 30 HLRS took part in the Tag der Wissenschaft
foundation for a successful digitalization strategy,
ings made possible by HLRS’s Hazel Hen supercom-
(Science Day) at the University of Stuttgart. Welcoming
pointing out that HLRS’s successful collaborations
puter, this volume covers the main methods in high-per-
more than 300 visitors, the event offered the general
with municipalities and German industry could be fur-
formance computing
public an opportunity to experience high-performance
ther developed through collaboration among politics,
and
computing and to learn about its use in research and
industry, and the scientific community. Strobl praised
that deliver a high
The spring of 2018 arrived at HLRS with more color
technology development. Visitors toured the HLRS
Baden-Württemberg’s progress in digitalization and
level
sustained
than in the past. Led by the HLRS Sustainability in HPC
computer room and visited the CAVE, a 3D virtual real-
encouraged ongoing persistence in the effort.
applications of
HLRS Blooms
performance. Scien-
Centers project, employees volunteered in late 2017 to
ity facility. There, they enjoyed the thrills of piloting a
tists and engineers
plant beds of flower bulbs on the HLRS grounds. Their
driving simulator through city streets or flying a para-
will find accomplish-
efforts paid off in April, when fields of crocuses and
gliding simulator high above a virtual landscape. Staff
ments in achieving
narzissuses burst through the earth. The activity was
also demonstrated visualization applications for city
the best performance
organized with the twin goals of strengthening biodi-
planning, and HLRS Director Michael Resch presented
for production codes
versity on the HLRS grounds and promoting feelings
of particular interest.
of wellbeing among the staff. In this way, it supported
The book includes a
the general mission of the HLRS sustainability project,
wealth of color illus-
which aims through a wide array of activities to simul-
trations and tables of
taneously address environmental, economic, and social
results.
dimensions of sustainability.
derwerke der Technik und des Geistes.” The talk provided a lively introduction to supercomputing as well as its cultural and philosophical context.
© courtesy of KISTI
a lecture titled “Simulation auf Supercomputern: Wun-
(LB)
(CW)
(CW)
Sustainability Days
HLRS and KISTI Identify Collaboration Opportunities
This year‘s Nachhaltigkeitstage (Sustainability Days)
Meeting in Daejeon, South Korea, HLRS Director
2nd Industrial HPC User Roundtable
at the University of Stuttgart took place in June as
Michael Resch and Korean Institute of Science and
Organized by HLRS and SICOS-BW, the second annual
including from Pfinder (supplier for the automobile
part of N!-Tage, a program focusing on sustainability
Technology Information (KISTI) President Dr. Choi
Industrial HPC User Roundtable (iHURT) focused on
industry), Putzmeister (cement pumping technology),
efforts across the state of Baden-Württemberg. Bring-
Heeyoon identified several collaboration opportunities.
the diverse challenges that industrial users of HPC
Bosch, Porsche, Stihl, and BASF. Lectures by HPC users
ing together sustainability initiatives at the University
One specific topic involves developing new tools for
face, including complex programming codes, new
provided insights into a variety of industrial HPC user
— including HLRS’s sustainability in supercomputing
digital product design and fabrication. By combining
technologies, licensing models, user support, and the
scenarios, covering topics including methods for sim-
program — the event promoted networking and idea
HEMOS-Fluid (a software platform developed at KISTI
organization of workflows, among others. Among the
ulating fire protection systems, the use of HPC in the
exchange. Over the course of two days, it featured talks
for fluid analysis) with COVISE (a software environment
more than 30 participants were representatives from
development of new cement pumps, and the impor-
about sustainability challenges and progress the univer-
developed at HLRS for simulation post-processing and
companies spanning a wide assortment of industries,
tance of HPC for the chemical industry.
sity is making, with a focus on the challenges of energy
visualization), the centers see opportunities to develop
consumption and cooling in high-performance comput-
a powerful new approach for visualizing simulation
ing. “With the Nachhaltigkeitstage we want to demon-
data. The centers anticipate that pooling their institutes’
strate that sustainable action has to be an overarching
expertise and strengthening collaboration on this topic
goal that is based on a common conviction and will ben-
could benefit the competitiveness of small- and medi-
efit all parties involved in the long run,” says Sabine Eger,
um-scale enterprises (SMEs) that use simulation in
a member of the HLRS sustainability team.
product design and development.
(AL)
(CW)
(CW)
Steffen Poppitz (Bosch) Stefan Hildenbrand (Pfinder KG) Matthias Sitte (BASF)
20 / 21
HLRS ANNUAL REPORT 2018
Susanne Kilian (hhpberlin)
Christian Simander (Putzmeister Engineering)
Markus Klietmann (SICOS-BW GmbH)
New European Centres of Excellence Funded
HLRS Visualization Supports Stuttgart 21 Construction
Four new COEs will address important social and industrial challenges.
3D virtual reality models created at HLRS helped engineers to successfully build a complex new kind of reinforced concrete column for the new Stuttgart train station.
In 2018 the European Commission approved all propos-
HiDALGO will combine data analysis, machine learning,
als for “Centres of Excellence on HPC” (CoE’s) contain-
and HPC to address global challenges in finance, health
In 2018 construction began on the most architectonically
staff created an interactive digital model of one of the
ing an HLRS contribution, with HLRS participating in
care, migration, and energy. In this unique consortium,
distinctive elements of the new Stuttgart central train
columns for display in the CAVE, a walk-in virtual reality
four out of nine funded EU CoE’s in this call. HLRS will
the computational and social sciences will join forces,
station. The so-called “Kelchstütze” are a set of mas-
environment.
receive three-year funding totalling €7 million. The proj-
with HLRS providing HPC expertise and its infrastruc-
sive, chalice-shaped columns that will cradle windows
Meeting at HLRS, representatives of the Deutsche Bahn
ects started operation in late 2018.
ture for high-performance data analysis.
to bring natural light into the subterranean station. At a
and construction managers at Ed. Züblin AG could then
HLRS will play the largest role in EXCELLERAT (Euro-
HLRS is also part of FocusCoE, a collaboration support
height of over 12 meters and a diameter of approximately
explore and interact with the visualization to develop a
pean Center of Excellence for Engineering Applica-
action that will bring all nine CoEs together to facilitate
32 meters, the elegantly curving forms of reinforced con-
strategy for erecting the column most efficiently and at
tions), acting as project coordinator thanks to its experi-
knowledge exchange. HLRS will contribute to training
crete promise to become a landmark in the city.
the highest possible structural integrity. Once the first
ence with industrial engineering partners. EXCELLERAT
activities aimed at identifying and solving shared prob-
For construction engineers working on the project —
column was poured onsite, the HLRS team also created
will support industrial HPC users by improving simula-
lems among the CoEs.
called Stuttgart 21 — the columns’ complex geometry
3D scans of the physical structure to compare the vir-
tion technologies for engineering applications, mak-
created unique challenges. Although CAD drawings of
tual model to the actual column and determine whether
ing the development and testing of high-tech products
the structures existed, the curving volumes and need to
it had been erected to specifications.
safer and more efficient. The goal of EXCELLERAT is to
precisely arrange an enormous number of steel bars to
According to Uwe Wössner, who heads the HLRS
support industry in scaling such applications to (pre)-
reinforce the concrete proved more challenging than is
Visualization Department, “The effort was extremely
exascale HPC systems.
the case with more conventional pillars.
productive, as the participants had an easier time visu-
To better plan the structures’ fabrication, engineers at
alizing the complex geometry, identified problems in
the Deutsche Bahn called upon HLRS to develop a 3D
the original design, and were able to develop effective
visualization. Using CAD and point cloud data, HLRS
solutions.”
HLRS is contributing to four new European Commission COE’s.
(LB)
POP 2 will support scientific users and software developers by helping them more efficiently parallelize and, in turn, improve the performance of their codes. Researchers who optimize their codes with HPC experts ultimately make better use of European supercomputers and improve their individual productivity. The CheeSE project will use HPC to predict natural disasters quicker and more reliably in order to reduce
A 3D visualization of one of the Stuttgart 21 “Kelchstütze,” in the HLRS CAVE.
response time.
22 / 23
HLRS ANNUAL REPORT 2018
(CW)
Supercomputing-Akademie Completes First Course
Sociopolitical Advisory Board to Help Guide HLRS Vision
Using a “blended learning” approach, the new training program will provide engineers and computer scientists with
A new multidisciplinary committee will provide counsel on how computer simulation can benefit society and help
essential HPC skills.
address its key challenges. in partnership with SICOS-BW, which promotes access
three-month training module offered by the Supercom-
to high-performance computing for small and medi-
puting-Akademie. Participants included engineers and
um-sized enterprises (SMEs). The Supercomputing-
computer scientists interested in developing their skills
Akademie aims to provide continuting education pro-
for working in high-performance computing environ-
gram that is attractive for both large and small compa-
ments.
nies.
The course, which focused on code development
The Supercomputing-Akademie courses are conceived
for parallel computing systems, included interactive
using a “blended learning” format, meaning that only
online content, animations, programming exercises,
a small part of the training takes place in a classroom.
and explanatory videos. Participants also met with one
Approximately 80% of a participant’s course time is
another and with instructors in virtual online meetings.
spent online, reducing the need to travel to participate.
The course covered the architecture of parallel systems
Courses are held in the German language, focusing on
and programming models such as MPI and OpenMP,
the needs of the country’s precision technology com-
and taught participants how to use programming librar-
munity.
From predicting weather and discovering new medi-
of society, representing socially relevant disciplines
ies and key parallelization concepts efficiently.
The Supercomputing-Akademie’s training program
cines, to enabling the more efficient design of cars and
including nursing, architecture, design, art, education,
With funding from the European Union and the State
continues to develop with future modules to cover
traffic flow, simulation indirectly benefits society in
and journalism.
of Baden-Württemberg, HLRS has been developing
topics such as simulation; visualization; performance
many ways. However, it is also often driven by scientific
“Simulation can make social phenomena more acces-
the Supercomputing-Akademie in collaboration with
optimization; cluster, cloud and HPC; ecology and eco-
or economic interests, while knowledge about it can be
sible to decision-makers and affected people,” said
the University of Freiburg and University of Ulm, and
nomics of supercomputing; and data management.
limited to experts in a particular field. This reality can
advisory board chairman Prof. Ortwin Renn, Scientific
lead to limitations in vision about how simulation could
Director at the Institute for Advanced Sustainability
best benefit society.
Studies in Potsdam. “Against the background of this
“There is no question that computer simulations pro-
progress, however, it must be ensured that ethical val-
vide added value, but they must not be a privilege of
ues are not violated and that social preferences are
elites,” says HLRS Director Michael Resch. For this
taken into account.”
reason, he and Andreas Kaminski, head of the HLRS
Discussions at the inaugural meeting focused on topics
Department of Philosophy of Computer Simulation,
in which simulation can have an impact, such as urban
convened the inaugural meeting of a new sociopoliti-
development, social inequality, and populism, as well as
cal advisory board on April 9. “We want to ensure that
questions regarding the methodology of simulations. In
a wider cross-section of society benefits from the use
the future the board will gather and refine such thematic
of simulation,” Resch explained. The sociopolitical advi-
and methodologic ideas, with the goal of developing a
sory board includes 13 members from different areas
pilot proposal for a research project.
24 / 25
HLRS ANNUAL REPORT 2018
(CW)
© Julian Rettig
Attendees at the first HLRS Sociopolitical Advisory Board meeting.
On July 17, fourteen participants completed the first
(LB)
SimTech Named National Center of Excellence
Media Solution Center to Promote Innovation in Film and Digital Art
A University of Stuttgart research consortium focused on data-integrated simulation technologies was awarded a
A new association will facilitate the integration of applications of high-performance computing and simulation into
prestigious seven-year grant.
media production.
Since 2007 the Excellenzcluster SimTech (SimTech
to see it once again recognized as a national center of
In October HLRS, together with the Center for Art
organize scientific continuing education activities that
Center of Excellence) has nurtured a multidisciplinary
excellence,” said Prof. Resch. “Our collaborations with
and Media Karlsruhe (ZKM) and the Hochschule der
train people working in media to use high-performance
community at the University of Stuttgart focused on the
scientific investigators will not only support exciting
Medien Stuttgart (HdM) founded a new association
computing more effectively. This includes focusing on
field of simulation. The cluster includes developers of
new discoveries, but also enable HLRS to continue to
called the Media Solution Center Baden-Württemberg
the use and development of software for high-perfor-
models, methods, and other simulation technologies as
improve its capabilities.”
lenges in media production where high-performance
The MSC is open to membership of other individuals
research in molecular and particle physics, mechanics,
computing and simulation could help, and to accel-
and companies working in the media industry — includ-
dynamic systems, and numerical mathematics.
erate the development of technologies that address
ing animation studios, visual effects specialists, and
Recognizing the University of Stuttgart’s role as
these needs.
arts organizations — as well as leaders of scientific
a national leader in simulation technology and
The MSC will undertake collaborative research and
and technical research groups in Baden-Württemberg
applications, the Deutsche Forschungsgemeinschaft
development projects aimed at solving problems that
whose work has relevance for media production. Ulti-
media producers face. It will also facilitate sharing of
mately, the MSC aims to facilitate a precompetitive
the knowledge and technologies that result from these
exchange of knowledge and information among all
collaborations and organize events focusing on themes
organizations that choose to participate.
that are important to the media industry. One key set
Prof. Dr. Bernd Eberhardt of the Hochschule der
of application areas in which the MSC will be engaged
Medien was named chairman of the association. HLRS
includes animation, simulation, and visual effects
Director Prof. Dr. Michael Resch will serve as deputy
(VFX). In coordination with the HdM, HLRS will also
chairman.
This highly selective program supports key research centers that are advancing Germany’s national strategy for scientific excellence. The new grant will enable SimTech to grow in new directions. For the next seven years, it will focus on increasing capacities for data-driven simulation. Such applications are important in fields in which large amounts of data are produced using sensors, experiments, simulations, and other methods. High-performance computing could enable simulation based on such datasets and so improving supercomputing infrastructure to enable efficient and effective high-performance data analytics is one of the key problems facing the field. For this reason, HLRS is an important partner in the SimTech cluster, with HLRS Director Michael Resch participating as a principal investigator for high-performance computing systems. “HLRS is well networked within SimTech and so we are delighted
26 / 27
HLRS ANNUAL REPORT 2018
(CW)
HLRS was recognized in 2017 with an HPC Innovation Award for its work with M.A.R.K. 13 and Studio 100 Media, who used high-performance computing to accelerate production of the animation film Die Biene Maja (Maya the Bee).
a national center of excellence in its latest funding cycle.
Perfusion MRI is a promising method for supporting therapy for multiple sclerosis. In one project, SimTech researchers conduct detailed small-scale simulations of how MR contrast agents spread in the brain. The method could potentially improve interpretation of images of MS-lesions.
announced on September 27 its selection of SimTech as
(CW)
© Studio 100 Media. Studio B Animation
mance computing environments.
© University of Stuttgart, Max Kovalenko
(MSC). The MSC’s mission is to identify technical chal-
well as computational scientists who use simulation for
Building a Sustainable HPC Infrastructure
A new European Union Centre of Excellence coordinated by HLRS will support industrial HPC users in adopting more
A two-day meeting at HLRS focused on the challenges of designing environmentally responsible facilities for super-
powerful hardware and software technologies.
computing.
Participants at the kick-off meeting of the EXCELLERAT Project.
EXCELLERAT Will Bring HPC Applications to Engineering Industry
HLRS launched EXCELLERAT (excellerat.eu), a new
industry to take advantage of the opportunities that the
The march toward faster, more powerful supercom-
planning can make HPC more sustainable, while at the
European Union Centre of Excellence (CoE) for engineer-
latest HPC technologies offer.
puters shows few signs of slowing down. Although
same time highlighting some important challenges and
ing applications. With roughly €8 million in EU funding,
EXCELLERAT will work with consortium partners who
such growth is good for science and technology devel-
problems to be addressed.
the center includes 13 partners in 7 European countries.
have developed important codes for academic appli-
opment, it also leads to increased physical needs for
The goal of any design for a supercomputing center is
Its mission is to accelerate technology transfer of lead-
cations in engineering fields such as aerospace, auto-
supercomputing centers — building spaces able to
to have the newest technology installed when the cen-
ing-edge HPC developments to the engineering sector.
motive, combustion, and fluid dynamics. To facilitate
accommodate larger machines, cooling systems that
ter opens. But considering the fact that erecting a brick
EXCELLERAT will facilitate the further development of
the codes’ integration into real-life industrial applica-
efficiently manage greater heat production, and energy
and mortar structure can take 5-7 years, the best tech-
important codes for high-tech engineering, maximizing
tions, these partners will work closely with end users
to keep everything running. How can the need for more
nologies often only become available after a project is
their scalability to increasingly large computing archi-
outside the consortium. This will ultimately lead to fast
resources be managed in a sustainable way?
underway. As the meeting revealed, this fact requires
tectures and supporting the technology transfer that
feedback-cycles in all areas of the HPC engineering
On October 23-24, IT specialists and research infra-
close communication among stakeholders and a plan-
will enable their uptake within the industrial environ-
application lifecycle, from consultation on method-
structure experts from the German-speaking HPC
ning approach that accounts for future expansion or
ment. These activities will support engineers through
ology and code implementation to data transfer and
community met at HLRS to discuss holistic strategies
other potential requirements that could arise when new
the entire HPC engineering application lifecycle, includ-
code optimization. End users will benefit by gaining
for addressing such challenges. The event — the sec-
technologies materialize. In addition, close cooperation
ing data pre-processing, code optimization, application
first-hand access to the project results. This concept of
ond annual meeting organized by HLRS on the topic
with funding agencies and implementing better metrics
execution, and post-processing. In addition, EXCEL-
a one-stop-shop for all services is unique in the area of
of sustainability in supercomputing centers — revealed
for tracking progress toward sustainability are import-
LERAT will provide training that prepares engineers in
industrial HPC.
how sensors, new cooling technologies, and creative
ant parts of the puzzle.
28 / 29
HLRS ANNUAL REPORT 2018
(LB)
(CW)
Networking the German Auto Industry: An Interview with Alexander Walser
The projects we work on are very diverse in terms of the number of computing cores used or of the parallelization necessary to achieve research goals. By including HLRS in project planning, we also learn about potential
asc(s Director Alexander F. Walser
problems and how to address them. Another important question for OEMs and small businesses is how to design a workflow that connects the workstation under your desk with the HLRS cluster system. In the past we have built workflows in which data and results are automatically transferred between classical workstations and HLRS using only the necCompetition in the automobile industry is intense, but the Automotive Simulation Center Stuttgart has showed that
essary software resources. In practice this means that
cooperation can offer important advantages.
engineers don’t just push everything onto the HLRS supercomputer, but rather they can integrate it into their
Car manufacturer Porsche has long used computer sim-
of interested parties necessary to get projects off the
ulation resources at HLRS to develop better vehicles. In
ground, in terms of expertise or financing.
the early 2000s, however, it became clear that exist-
It is also important to our mandate to integrate small
ing commercial software packages were not keeping
and medium-sized enterprises (SMEs), as well as scien-
pace with the constantly doubling performance of new
tific institutions. SMEs sometimes develop very innova-
supercomputing hardware. To accelerate the develop-
tive software packages, but if the interfaces aren’t built
Ten years ago no one was talking about self-driving
ment of more efficient and effective simulation soft-
correctly or if the workflow doesn’t fit, industry might
vehicles. Electric cars were a theme that was in the air,
ware, Porsche joined forces in 2008 with HLRS, Daim-
not be able to use them. Getting OEMs and SMEs to
but not something that was as prominent as it is today.
ler, Opel, and other companies — with support of the
work together reduces this risk. Building these relation-
Now everything must develop much, much faster.
Baden-Württemberg Ministry for Science, Research,
ships also incorporates industry’s needs into software
Virtual crash tests — which are necessary for safe and
and the Arts — to found the Automotive Simulation Cen-
development early and makes it possible to design
efficient development — are also becoming increas-
ter Stuttgart, also called asc(s.
solutions that meet the needs of multiple OEMs at the
ingly important. Countless test scenarios need to be
Since then the asc(s has built a network of large and
same time.
conducted, raising the question of how HPC and driving
small companies from across the automobile indus-
Ten years ago the foundational idea of the asc(s was
simulation can be better combined. In the medium-term,
try, with HLRS playing a central role. Recently, we
very innovative. Today, it fits contemporary needs per-
such tests will have to be conducted on a supercomputer.
spoke with asc(s Director Alexander F. Walser about
fectly and in the future it will be indispensable for driv-
Technological trends related to artificial intelligence
the association’s activities, HLRS’s contributions to
ing innovation.
are also gaining in importance. This will result not only
What are the most important topics in simulation for the future of automobile design and development?
in the use of the newest AI-technologies in driving sys-
them, and why simulation is growing in importance What role does HLRS play in this network?
for the field.
workflows.
tems, but also in asking how we can use AI to lessen
High-performance computing is a key focus at the
process. How can we apply AI in designing simulation
asc(s, a fact that distinguishes our organization from
models? How can we extract more knowledge from the
The asc(s is a nonprofit association that brings car
other interest groups in the automotive industry. It is
simulation results that now exist?
manufacturers, software developers, hardware manu-
only through our work with HLRS that we can maintain
Because of these new trends, the horizontal networking of
facturers, engineering service providers, start-ups, and
this strong focus on HPC.
automobile manufacturers that the asc)s promotes is very
scientific organizations together to steer the future of
Cooperation with HLRS enables us to have access to
important, particularly as new key players enter the mar-
virtual automobile development. Just as an OEM (orig-
the newest hardware architecture. HLRS is always a step
ket that have never been integrated into the supply chain.
inal equipment manufacturer) can buy a screwdriver
ahead of what’s available in industry, where the update
to build a car, we try to ensure that manufacturers
cycles are longer and the cluster sizes are smaller. Using
What plans does the asc(s have to make the auto-
have access to the simulation methods they need. To
HLRS resources, we can show relatively early that new
motive industry better and to prepare for the future of
make this possible, the asc(s gathers the critical mass
methods will work on future cluster systems.
simulation?
30 / 31
HLRS ANNUAL REPORT 2018
© Benjamin Stollenberg
the burden on engineers during the virtual development How do you describe the asc(s?
Student Dives into Data to Predict Train Delays
As a participant in the HLRS program Simulated Worlds, high school student Niklas Knöll used machine learning to investigate a common problem.
Simulation is becoming ever more important for the
Smartphone apps from the Deutsche Bahn and Stutt-
for solving big data problems on supercomputers. He
gart’s local transit networks can show delays of trains
developed models and trained them using machine
and buses in real-time. But wouldn’t it be nice to know
learning algorithms to predict train departures to the
earlier whether there’s time to stop at the bakery for a
minute. In 8 out of 10 cases his approach could clearly
pretzel?
predict whether a city train would leave at least three
As a high school student living in Stuttgart, Niklas
minutes late.
Knöll is very familiar with this problem. In 2018 he was
From a strictly statistical point of view, this result was
one of eight scholars who participated in Simulated
not fully satisfying because the database Niklas used
Worlds, an initiative of the Baden-Württemberg Minis-
was too small for a big data problem. But Simulated
try of Science, Research, and the Arts. Over the school
Worlds’ aspiration is not to produce scientific break-
motivate students to study virtual automobile develop-
year, students receive €1,000 to conduct a research
throughs. Most importantly, it provides the participat-
auto industry. This presents an important question:
ment and to select the best program of study.
project at HLRS, receiving supervision from HLRS
ing scholars with advanced technical competencies
what qualifications will tomorrow’s simulation engineer
The need for experts in simulation is constantly grow-
employees. The aim is to prepare young people for the
and problem-solving skills. These skills were on dis-
or even an engineer in another field need?
ing. By promoting young investigators and helping
digital realm.
play at an awards ceremony at HLRS on July 4, when
At the asc(s we have been asking ourselves how we
them to quickly become integrated into the develop-
At HLRS students were encouraged to analyze big
Niklas and the other participating students presented
could support our member organizations by helping to
ment process, we can not only offer advantages for
datasets. Niklas accessed public data from the Deut-
their projects and results. According to Simulated
ensure that university students gain the right knowl-
industry but also strengthen Baden-Württemberg as a
sche Bahn and learned to analyze it using the pro-
Worlds project coordinator Doris Lindner, “Everyone
edge during their studies. This includes engaging in
center for research. We expect that this kind of cooper-
gramming language Python and the programming
was impressed by how thoroughly the students had
dialogue with universities, as well as finding ways to
ation will grow stronger in the future.
framework Apache Spark, tools that are often used
studied their subject.”
32 / 33
HLRS ANNUAL REPORT 2018
(CW)
(LB)
System Usage
HLRS by the Numbers
Education and Training
1.28
Visiting Researchers
Staff
155
BILLION
107
36
41
63
25 Teilnehmer 997 Participants
129 Course-Days
CORE HOURS PRODUCED USER PROJECTS INDUSTRIAL CUSTOMERS USER PUBLICATIONS
23
CONTINUING EDUCATION COURSES
12
SCIENTIFIC WORKSHOPS 451 Participants 24 Days
22
UNIVERSITY LECTURES 2,733 Participants 36 SWS
78
SCIENTISTS
49
NONSCIENTISTS
Visitors at HLRS
RESEARCH ASSISTANTS
2,142
5
Third-Party Funds Talks
8,136,13 8
44
€
52 %
Staff Publications
20 2 34 / 35
PAPERS IN JOURNALS, BOOKS, AND CONFERENCE PROCEEDINGS
33 %
11 % 3 %
BOOKS
HLRS ANNUAL REPORT 2018
1 %
FEDERAL GOV. EU STATE GOV. DFG OTHER
Providing personalized guidance to HPC users results in faster codes and more efficient use of computing resources. Since 2007 the Gauss Centre for Supercomputing
from the help of HPC experts in selecting the best tools
(GCS) — the network of Germany’s national supercom-
and optimizing their performance. This can make a big
puting centers that includes HLRS, the Jülich Super-
difference, making it possible to ask new, more com-
computing Centre, and the Leibniz Supercomputing
putationally demanding kinds of scientific questions.
Centre — has coordinated the installation and oper-
Since 2016 HLRS has held a biannual optimization
ation of world-class systems for high-performance
workshop in which Hazel Hen users meet with sup-
computing. In 2018 it expanded in an important new
port staff to identify opportunities for improving their
direction, implementing enhanced user support ser-
codes’ performance. These meetings have been very
vices that will empower scientists to use these sys-
productive, resulting in improvements in runtime and
tems more efficiently.
data transfer speeds; in some cases, codes can end up
“Every time you help individual users to utilize the
running twice as fast, or faster.
machine more efficiently you enhance the capabilities
With grants provided by the Baden-Württemberg Min-
of the computer as a whole,” says Thomas Bönisch,
istry of Science, Research, and Art, HLRS added new
leader of the HLRS user management team. “By ensur-
staff positions in 2018, each of which is dedicated to
ing that applications are optimized for the computing
supporting specific user groups all year round. In this
hardware they are run on, we can get a lot more science
way, deeper collaborative relationships can develop
out of the same computing resources.”
in which experts in scientific domains and experts
Enhanced user support involves more than helping
in high-performance computing regularly exchange
users port codes or transfer data. Instead, scientists
ideas, working together to continually improve com-
receive hands-on, personalized advice for improving
putational tools.
their codes’ performance on HPC resources. This can
This enhanced user support initiative has also made
occur at several levels: selecting and optimizing the
it possible to improve coordination of user support
best algorithm for a particular problem and supercom-
activities between HLRS and its GCS partner institu-
puter, improving its ability to scale up to parallel com-
tions. Representatives of the three centers now meet
puting networks, structuring it to take advantage of
regularly to share knowledge about their accomplish-
unique characteristics of the specific processors that
ments in improving computational efficiency. This
make up the system, and eliminating bottlenecks in
dialogue also enables the three centers to address
the movement of large datasets.
the needs of users who compute at multiple GCS sites
While some computational scientists develop their
in a more coordinated way. In this way, the initiative
own computational tools for analyzing a scientific
promises to raise the performance of users’ codes
problem, others rely on community-built codes or
across Germany’s entire national supercomputing
open-source software. In either case, they can benefit
infrastructure.
36 / 37
HLRS ANNUAL REPORT 2018
(CW)
Attendees at an HLRS optimization workshop receive guidance on improving their codes.
Enhanced User Support Improves Performance of HPC Systems
EUXDAT: Taking Agriculture into the Cloud
PhD Graduates 2018
High-performance data analytics holds enormous potential to improve efficiency in food production.
Three students completed their doctoral studies at HLRS in 2018. Read more about their dissertation topics below.
In the past, farmers consulted books like The Farmer’s
datasets is needed, users could access high-perfor-
Björn Bliese
Almanac to know which crops to plant, when to plant
mance computing resources at HLRS through the web.
An Augmented Reality System Suitable for Geometric Analysis in Product
them, and when to harvest them. As in other industries
Having access to such a product could, for example,
Development
being transformed by the Internet of Things, however,
enable farmers to identify the optimal time to fertilize
Augmented reality offers the potential to perform geometric analyses between
digital tools have initiated new thinking about the future
their crops to produce the best harvest.
physical and virtual models in product development and production planning,
of agriculture. High-resolution weather and climate
Atos, a European provider of business technology,
although today it is rarely used in practice. Bliese’s thesis defines basic require-
models and sensors for measuring soil moisture and
invited HLRS to join the project because of its exper-
ments for a usable AR system for geometric analyses and improves upon the
composition, for example, offer rich sources of data that
tise in high-performance computing, cloud comput-
chosen system components. It provides a process-driven user interface to guide
could make farming more productive. The challenge
ing, and big data analytics. According to Dr. Michael
a user through an AR-assisted investigation. The developed AR system is finally
now is to design systems to integrate, interpret, and
Gienger, who heads the cloud computing research
evaluated within a usability study.
apply this information.
group at HLRS, “For us, this is the best of all possible
As part of a multidiscplinary European Union-funded
projects, because it combines many key areas of exper-
Marius Feilhauer
innovation project called EUXDAT (European e-Infra-
tise at HLRS, while offering an opportunity to test new
Simulation-Supported Safeguarding of Driver Assistance Systems
structure for Extreme Data Analytics in Sustainable
approaches for integrating cloud and HPC technologies
In theory, ensuring that an autonomous vehicle drives safely would require pro-
Development), researchers at HLRS are developing a
in a field that could really benefit from it.”
gramming it to recognize all possible situations, including various traffic or light-
(CW)
software platform to address this need. The goal is to
ing conditions. This is impossible, however. In his thesis, Feilhauer proposes an
create an off-the-shelf product hosted in the cloud that
expandable catalog of traffic scenarios that would grow as it incorporates lessons
could make big data analytics available to anyone in
Learn more about how data analytics and cloud com-
learned based on real or virtual driving test data. Driving scenarios could be inves-
agriculture. In cases where analysis of extremely large
puting can support agriculture at www.euxdat.eu.
tigated in virtual test drives on computers, with the results being fed back into the catalog. Such a resource would improve iteratively, benefiting the training of machine learning applications for autonomous driving.
Matthias Nachtmann Model-Centric Task Debugging at Scale With the increasing complexity of HPC hardware and software, there is a need for tools to support computer scientists and engineers in efficiently porting and optimizing their computational codes regardless of what system they are using. In addition, researchers want to be able to debug their applications within their respective programming models. In his thesis, Nachtmann focuses on further developing this “model-centric” debugging approach, spearheading the redevelopment of the backend of the HLRS-developed TEMANEJO graphical debugging tool.
© Shutterstock/ESB Professional
38 / 39
HLRS ANNUAL REPORT 2018
Looking Inside Simulation‘s “Black Box”
Nicole J. Saam
Johannes Lenhard
Visiting scholars in the HLRS Department of Philosophy of Computer Simulation work toward a clearer understanding
The opacity of simulation
from individual to individual? As defined now, opacity
of computational research and its limitations
Perhaps the most interesting philosophical problem in
is a relational concept. We would love to have a way to
computer simulation is the result of what philosophers
measure this.”
Simulation and other applications of high-perfor-
that a conceptual framework was needed for discuss-
call its “opacity.” Often, a computer simulation behaves
Saam also anticipates a practical use for such a tool.
mance computing have become indispensable tools for
ing the deeper implications of this trend, “For philoso-
in unexpected ways. In many cases it is impossible
“If one could know in advance which kinds of simula-
research in the basic and applied sciences. Employing
phers of science,” Lenhard says, “one important prob-
to determine why, because it is impossible to mecha-
tion models typically suffer from specific problems due
HPC has had many positive impacts, helping to strive
lem is what it means for a scientist to know something.
nistically observe how an algorithm functions. “This
to opacity,” she suggests, “one might be able to iden-
for a deeper understanding of our world and enabling
In the case of computer simulation, is the knowledge it
turns the idea of mathematical modeling on its head,”
tify and avoid typical problems early in the process of
the development of new technologies. As our reliance
produces different from that produced by older meth-
Lenhard says, “because we used to think that mathe-
implementing a model.” What might at first seem to be
on computer simulation grows, however, many wonder
ods? If so, how? As a historian, one can also ask how
matical models should be able to make the causes of a
a theoretical issue could thus enable tangible improve-
how it is changing science, technology, politics, and
the conditions in which computer simulation is used
behavior crystal clear.”
ments in how simulation is done.
society, and how we should react.
have changed over the years. This gives us a better
During her residency at HLRS, Nicole J. Saam has been
As a supercomputing center, HLRS is unusual in that
understanding of how simulation is used today.” In his
developing a systematic method for defining, catego-
The value of interdisciplinary dialogue
for many years it has prioritized discussion about such
research Lenhard has written about features that distin-
rizing, and measuring such opacity. Although her proj-
Bridging the conceptual divides between scientists and
issues. Through its Department of Philosophy of Sim-
guish today’s computer simulation from earlier mathe-
ect is still evolving, such a model might address factors
researchers in the humanities and social sciences is not
ulation, HLRS promotes interdisciplinary dialogue
matical modelling.
such as instability in the numerical model itself, meth-
easy, but Saam’s project suggests why it is important to
among simulation scientists, philosophers, historians
One example is the change in experiments using sim-
ods to simplify a model to reduce computational effort,
make the effort. “Sometimes it can take years or even
of science, sociologists, and other scholars whose
ulation. Supercomputers enable scientists to easily
and the number of different research teams involved in
decades to get scientists and researchers from diverse
insights enable reflection on the nature of simulation.
adjust a mathematical model’s parameters so that it
developing a model, each of whom only understands a
disciplines such as natural sciences, engineering sci-
In 2018, the Department of Philosophy of Simulation
approaches observed data. Deep learning algorithms
small piece of the project. Such criteria address the fact
ences, humanities, or the social sciences to have a pro-
hosted two visiting scholars — PD Dr. Johannes Lenhard
using neural networks are perhaps the best exam-
that opacity in computer simulation can have technical,
ductive conversation,” Saam remarks. “The opportunity
(Philosophy Department, University of Bielefeld) and
ples of this approach; by adjusting parameters, scien-
mathematical, and social origins.
for me to work directly with simulation scientists here at
Prof. Dr. Nicole J. Saam (Institute for Sociology, Fried-
tists gain the ability to model almost any behavior. But
After defining the dimensions of opacity, Saam has
HLRS is very unusual.”
rich-Alexander-Universität Erlangen-Nürnberg). Their
although such a capability can be useful, Lenhard asks,
developed a scientific framework and questionnaire.
Lenhard agrees. “I find it wonderful that HLRS has a
research highlights some key questions facing the phi-
is it still science?
Her team has started conducting interviews with
group whose job it is to talk with interested scientists
losophy of simulation today, and their experiences at
“The idea that general laws can reveal order in chaos is
groups of scientists who run their simulations at HLRS
about their work, and to promote greater self-reflection
HLRS show the benefits of promoting multidisciplinary
historically connected to the idea that these laws can be
in order to gain a more precise understanding of opacity
about what they are doing,” he observes. “It is some-
discourse.
formulated mathematically,” Lenhard says. “But newer
in all of its different forms.
thing that the scientific community and society in general need to do more often.”
kinds of computer modelling have nothing to do with
“Opacity can look quite different from the perspec-
Simulation yesterday and today
finding such general laws. Today, we hope to describe
tive of a principal investigator in comparison with that
Johannes Lenhard initially trained as a mathematician,
and manage things in a predictable way, even if we don’t
of a graduate student,” Saam points out. “Because of
but in 2001 became captivated by philosophical and
have a law for it.” In this sense, computer simulation is
each’s degree of experience, their perceptions of opac-
historical questions related to the growing use of com-
different from some earlier scientific approaches, but
ity can be different. Is opacity an objective condition of
puting across many scientific disciplines. He realized
offers completely new options.
simulation, or is it a subjective experience that varies
40 / 41
HLRS ANNUAL REPORT 2018
(CW)
USER RESEARCH 42 / 43
HLRS ANNUAL REPORT 2018
Supercomputer Enables Sound Prediction Model for Noise Control
Combining principles from computational fluid dynamics and acoustics, researchers at the TU Berlin have developed a model that could simplify the process of designing noise cancelling structures for airplanes, ships, and ventilation systems. Noise-cancelling headphones have become a popular
In a paper published in the journal Acta Mechanica,
accessory for frequent flyers. By analyzing the back-
Lewin Stein and Jörn Sesterhenn of the TU Berlin
ground frequencies produced by an airplane in flight
describe a new analytical model for sound predic-
and generating an “anti-noise” sound wave that is per-
tion that could make the design of Helmholtz cavities
fectly out of phase, such headphones eliminate dis-
cheaper and more efficient. The development of the
turbing background sounds. Although the headphones
model was facilitated by a dataset produced using
can’t do anything about the cramped seating, they can
direct numerical simulation at the High-Performance
make watching a film or listening to music in flight
Computing Center Stuttgart (HLRS).
nearly as enjoyable as at home.
The analytical model can predict, in a way that is more
To minimize the disturbing noise caused by loud
generally applicable than before, a potential Helm-
machines like cars, ships, and airplanes, acoustic engi-
holtz cavity’s sound spectrum as turbulent air flows
neers use many strategies. One technology, called a
over it. The authors suggest that such a tool could
Helmholtz cavity, is based on a similar concept to that
potentially be used to tune Helmholtz cavities to can-
used in noise-cancelling headphones. Here, engineers
cel out or to avoid any frequency of interest.
build a resonating box that opens to a slit on one side. As air passes over the slit, the box vibrates like a church
Simulation approaches all the scales of nature
organ pipe, producing a tone. By adjusting the size and
When moving air passes over the slit of a Helmholtz
shape of the cavity and its slit, acoustic engineers can
cavity, its flow becomes disrupted and turbulence is
tune it to produce a specific tone that — like the head-
enhanced. Vortices typically arise, detaching from the
phones — cancels a dominant, irritating sound pro-
slit’s upstream edge. Together they form a sheet of
duced by machinery.
vortices that covers the slit and can interact with the
Historically, the process of tuning a Helmholtz cavity was
acoustic vibrations being generated inside the cavity.
a brute force undertaking involving costly and time-con-
The result is a frequency-dependent damping or exci-
suming trial and error. Engineers had no other choice but
tation of the acoustic wave as air passes through this
to physically build and test many different geometries
vortex sheet.
experimentally to find an optimal shape for a specific
In the past it was difficult to study such interactions
application, especially in an environment of turbulent flow.
and their effects numerically without making crude
Today, however, high-performance computing offers
approximations. For the first time, Stein’s simulation
the potential to undertake such tests virtually, making
realistically integrates turbulent and acoustic phe-
the design process faster and easier.
nomena of a Helmholtz cavity excited by a turbulent
30 milliseconds of physical time. Each run of the
predict acoustic proper-
numerical model on HLRS’s Hazel Hen supercom-
ties of other Helmholtz
puter required approximately four 24-hour days, using
cavity geometries and
some 40,000 computing cores. Whereas a physical
airflow
experiment is spatially limited and can only track a few
comparing the extrapo-
physically relevant parameters, each individual DNS
lated model results with
run provides a 20-terabyte dataset that documents all
experimental data pro-
flow variables at all time steps and spaces within the
vided by Joachim Golliard
mesh, delivering a rich resource that can be explored
at the Centre de Transfert
conditions.
By
de Technologie du Mans in France, Stein found that the model did so with great accuracy. The model reported in the paper is optimized for low speed airflows and for low frequencies, such as those found in ventilation systems. It is also designed to be modular so that a cavity that includes complex materials like foam instead of a hard wall can be investigated as well. Stein anticipates that gaining more computing time and access to faster supercomputers would enable him to numerically predict a wider range of © Lewin Stein, TU Berlin
potential resonator shapes and flow conditions. Having recently completed his PhD and now working as a postdoc at the Institute of Fluid Dynamics and Technical Acoustics in the group of Prof. Sesterhenn (TU Berlin), Stein foresees some attractive opportunities to cooperate with industrial partners and possibly to
flow passing over its slit. At an unprecedented reso-
equations — to get as close as possible to the actual
in detail. Stein says that running the simulation over
apply his model in real-life situations. “Although I stud-
lution, it makes it possible to track the flow–acoustic
phenomenon in nature while using as little approxima-
this time period provided a good compromise between
ied theoretical physics,” he explains, “it is fulfilling to
interaction and its implications for the cavity’s reso-
tion as necessary,” Stein says. “Our DNS enabled us to
being able to set up a reliable database and getting
work on problems that reach beyond pure academic
nance.
gain new insights that weren’t there before.”
results in a practical amount of time.
research and can be applied in industry, where people
This achievement is possible using a method called
Stein’s direct numerical simulation divides the sys-
direct numerical simulation (DNS), which describes a gas
tem into a mesh of approximately 1 billion grid points
Moving toward a general sound prediction model
This latest paper is an opportunity to prove the utility
or liquid at a fundamental level. “I’m using the most com-
and simulates more than 100 thousand time steps,
Once the details of the acoustic model were devel-
and applicability of our work. It’s a great moment after
plex form of fluid equations — called the Navier–Stokes
in order to fully resolve the system dynamics for just
oped, the next challenge was to confirm that it could
years of working on a PhD.”
44 / 45
HLRS ANNUAL REPORT 2018
can potentially profit from what you’ve accomplished.
(CW)
Using Computational Chemistry to Investigate New Semiconductor Technologies
molecules develops over time, at the level of atoms and
“To do this,” Tonner explains, “we really need to under-
electrons and at time scales of picoseconds (one pico-
stand how the interfaces between silicon and these
second is one trillionth of a second).
organic compounds look and behave. The reaction
“Increasing computing power has made it possible for
between these two material classes needs to proceed
computational chemistry and quantum chemistry to
in a very controlled manner so that the interface is as
describe real molecular systems. Just 15-20 years ago,
perfect as possible. With computational chemistry we
people could only look at small molecules and had to
can look at the elemental details of these interactions
make rather strong approximations,” Tonner explains.
and processes.”
“The computational chemistry and solid state theory
For example, to cover a slab of silicon, liquid precursor
communities have now solved the problem of paralleliz-
molecules for the constituent atoms of gallium arsenide
ing their codes to operate efficiently on HPC systems.
are placed in a bubbler, where they are then brought into
University of Marburg researchers are exploring how functionalizing silicon with other compounds could make light-
As supercomputers get bigger, we anticipate being able
the gas phase; following a chemical process gallium and
based signaling in semiconductors feasible.
to develop increasingly realistic models for experimen-
arsenide atoms attach to the silicon, forming a GaAs film.
tal systems in materials science.”
How atoms are arranged when they adsorb to a surface
As new methods have become available for under-
Bringing computation to chemistry
standing and manipulating matter at its most funda-
Atoms bond together to form molecules and com-
Toward light-based semiconductors
how this takes place is an open question. Previously, it
mental levels, researchers in the interdisciplinary field
pounds when they approach one another and then
One area in which Tonner is currently using computa-
had been suggested that repulsive relationships among
of materials science have successfully synthesized
trade or share electrons orbiting around their nuclei.
tional chemistry is to study ways to improve silicon for
atoms is the most important factor in “steering” atoms
new kinds of materials. Often the goal is to design
The specific atoms involved, the physical shapes that
use in new kinds of semiconductors. This problem has
into place when they adsorb on a surface. By using DFT
materials that incorporate properties that can be use-
the molecules take, their energetic properties, and how
gained urgency in recent years, as it has become clear
and looking at intriguing features of how electrons are
ful for performing specific functions. Such materials
they interact with other nearby molecules are all prop-
that the microelectronics industry is reaching the limits
distributed, Tonner determined that the ability of atoms
can, for example, be more chemically stable or resis-
erties that give a compound its unique properties. Such
of its ability to improve semiconductors using silicon
to steer other atoms into place on the surface can also
tant to physical breakage, have advantageous electro-
characteristics can determine whether compounds are
alone.
result from attractive dispersive interactions.
magnetic characteristics, or react in predictable ways
likely to remain stable, or whether stresses such as
Tonner and experimental colleagues have been investi-
Gaining a better understanding of these fundamental
to specific environmental conditions.
changes in temperature or pressure could affect their
gating how functionalizing silicon with compounds such
interactions should help designers of optically active
Dr. Ralf Tonner and his research group at the Uni-
reactivity.
as gallium arsenide (GaAs) could enable the design of
semiconductors to improve adsorption of the precur-
versity of Marburg are addressing the challenge of
Tonner uses a computational approach called density
new kinds of semiconductors. This research posits that
sor molecules onto silicon. This, in turn, would make it
designing functional materials in an unusual way — by
functional theory (DFT) to explore such characteris-
such new materials would make it possible to use light
possible to combine light signal conduction with sili-
applying approaches from computational chemistry.
tics at the quantum scale; that is, at the scale where
instead of electrons for signal transport, supporting the
con based microelectonics, bringing together the best
Using computing resources at HLRS, Tonner models
Newtonian mechanics becomes replaced by the much
development of improved electronic devices.
of both worlds in optical and electronic conduction. (CW)
phenomena that happen at the atomic and subatomic
stranger world of quantum mechanics. DFT uses infor-
scale to understand how factors such as molecular
mation about variations in the density of electrons
structure, electronic properties, chemical bonding,
within a molecule — a quantity that can also be exper-
and interactions among atoms affect a material’s
imentally measured using x-ray diffraction — to derive
behavior.
the energy of the system. This, in turn, enables the
Tonner and his group have highlighted the ability of
researchers to infer interactions among nuclei as well
computational chemistry and high-performance com-
as between electrons and nuclei, factors that are critical
puting to reveal interesting phenomena that occur
to understanding chemical bonds and reactions.
between organic molecules and surfaces, and have
DFT can provide useful, though static, information
demonstrated how these interactions can be under-
about the energy profiles of the compounds they study.
stood with respect to the molecular and solid state
To gain a better understanding of how systems of mol-
world. The knowledge they have gained could help
ecules actually behave when interacting with a surface,
design patterned surfaces, a goal of scientists working
Tonner’s group also uses high-performance comput-
on the next generation of more powerful, more effi-
ing at HLRS to perform molecular dynamics simula-
cient semiconductors.
tions. Here, the scientists look at how the system of
rL
ab
46 / 47
HLRS ANNUAL REPORT 2018
Artist’s rendering of organic molecules adsorbing on a silicon surface.
ne
© Aaron Beller
©
n To
is determined by chemical bonding, though precisely
whether carbon dioxide could replace water in power generation. In conventional steam power plants, residual water is
the high-resolution fluid dynamics simulations they
separated from power-generating steam. This process
required.
limits efficiency, and in early generation power plants
© IKE, University of Stuttgart
Using high-performance computing and data-driven machine learning, University of Stuttgart researchers investigate
The simulation shows the structure and the (red) high and (blue) low speed streaks of the fluid during a cooling process. The researchers observed a major difference in turbulence between downward flowing (left) and upward flowing (right) supercritical carbon dioxide.
Simulation and Machine Learning Could Make Power Plants More Efficient
could be volatile, leading to explosions. In the 1920s,
The heat of the moment
Mark Benson realized that the risk could be reduced
Water is often used in power generation and heat trans-
do so, the team turned to a method called direct numeri-
Critical next steps
and power plants could be more efficient by bringing
fer because it is easily accessible, well-understood on
cal simulation, which is very computationally expensive
To date, the team has been using OpenFOAM, a com-
water to a supercritical state — when a fluid exists as
a chemical level, and predictable under a wide range
and requires HPC resources.
munity code, for its DNS simulations. While Open-
both a liquid and gas at the same time. Though it was
of temperature and pressure conditions. That said, in
FOAM is a well-established code for a variety of fluid
too expensive to use in practice during his day, the
order to become supercriticial. water must be heated
Neural networks for commercial computers
Benson Boiler offered the world its first glimpse of
to 374°C and placed under extremely high pressure.
Using the stress and heat transfer data coming from its
wanted to use a higher-fidelity code. The researchers
supercritical power generation.
Further, when a material enters its critical state, even
high-fidelity DNS simulations, the team worked with
are working with a team from University of Stuttgart’s
Almost a century later, researchers at the University
slight changes to temperature or pressure can have a
SIT’s Dr. Wanli Chang to train a deep neural network
Institute of Aerodynamics and Gas Dynamics (IAG) to
of Stuttgart’s Institute of Nuclear Technology and
large impact. For instance, supercritical water does not
(DNN), a machine learning algorithm modeled roughly
use its FLEXI code, which offers higher accuracy and
Energy Systems (IKE) and Institute of Aerospace Ther-
transfer heat as efficiently as it does in a purely liquid
after networks of neurons in the brain.
can accommodate a wider range of conditions.
modynamics (ITLR) are revisiting Benson’s concepts
state, and the extreme heat needed to reach supercriti-
Using Hazel Hen, the team ran 35 DNS simulations,
Pandey also mentioned he is using a method called
to explore how it can improve safety and efficiency in
cal levels can lead to degradation of piping and, in turn,
each focused on one specific operational condition,
implicit LES in addition to the DNS simulations. While
modern power plants. Using high-performance comput-
catastrophic accidents.
and then used the generated dataset to train the DNN.
implicit LES simulations do not have quite the same
ing, the researchers are developing tools that can make
Although using carbon dioxide (CO2) to make power
The team uses inlet temperature and pressure, heat
high resolution as the team’s DNS simulations, it does
supercritical heat transfer more viable.
plants cleaner may sound like an oxymoron, Pan-
flux, pipe diameter, and heat energy of the fluid as
allow them to run simulations with higher Reynolds
Sandeep Pandey, a PhD candidate at IKE, and Dr.-Ing.
dey and his colleagues are investigating its use as an
inputs, and generates the pipe’s wall temperature and
numbers, meaning it can account for a wider range of
Xu Chu of ITLR are leading the computational aspects
alternative. The common molecule offers a number of
wall shear stress as output. Eighty percent of the data
turbulence conditions.
of this research. In cooperation with computer science
advantages, chief among them being that it reaches
generated in the DNS simulations is randomly selected
The team wants to continue to enhance its database
researchers at the Singapore Institute of Technology
supercriticality at just over 31°C, making it far more effi-
to train the DNN, while researchers use the other 20
in order to further improve its DNN tool. Further, it is
(SIT), they are employing machine learning techniques
cient than water. In addition, supercritical CO2 (sCO2)
percent of data for simultaneous, but separate, vali-
collaborating with IKE experimentalists to conduct pre-
informed by high-fidelity simulations. They are also
needs far less space and can be compressed with far
dation.
liminary experiments and to build a model supercritical
developing a tool that can be easily used on commer-
less effort than subcritical water. This, in turn, means
After the team felt confident with the agreement, they
power plant in order to test the agreement between
cial computers.
that it requires a smaller power plant.
used the data to start creating a tool for commercial
experiment and theory. The ultimate prize will be if the
To achieve these goals, the team needed to run com-
In order to replace water with carbon dioxide, though,
use. Using the outputs from the team’s recent work as
team is able to provide an accurate, easy-to-use, and
putationally intensive direct numerical simulations
engineers need to understand its properties on a fun-
a guide, they were able to use their DNN to simulate the
computationally efficient tool that helps engineers and
(DNS), which is only possible using high-performance
damental level, including how the fluid’s turbulence
heat energy of specific operational conditions in 5.4
power plant administrators generate power safer and
computing. HLRS’s Hazel Hen supercomputer enabled
transfers heat and in turn, interacts with machinery. To
milliseconds on a standard laptop computer.
more efficiently.
48 / 49
HLRS ANNUAL REPORT 2018
dynamics simulations, Pandey indicated that the team
(EG)
Selected User Publications
Ahmad N, Farooqi MN, Unat D. 2018. Load balancing for parallel multiphase flow simulation. Sci Programming-Neth.
Coppola E, Sobolowski S, Pichelli E, et al. 2018. A first-of-its-kind multi-model convection permitting ensemble for
2018:6387049.
investigating convective phenomena over Europe and the Mediterranean. Climate Dynamics. DOI:10.1007/s00382018-4521-8.
Albers M, Meysonnat PS, Schröder W. 2018. Closed-loop control of turbulent boundary layers using spanwise traveling surface waves. Proc Appl Math Mech. 18:201800336.
Cormier M, Caboni M, Lutz T, et al. 2018. Numerical analysis of unsteady aerodynamics of floating offshore wind turbines. J Phys Conf Ser. 1037:072048.
Alon OE, Cederbaum LS. 2018. Attractive Bose-Einstein condensates in anharmonic traps: Accurate numerical treatment and the intriguing physics of the variance. Chem Phys. 515:287-298.
Donnert J, Jang H, Mendygral P, et al. 2018. Towards exascale simulations of the ICM dynamo with WENO-Wombat. Galaxies. 6(4):104.
Babb J, McLaughlin BM. 2018. Radiative charge transfer between the helium ion and argon. Astrophys J. 860(2). Dumbser M, Guercilena F, Köppel S, et al. 2018. Conformal and covariant Z4 formulation of the Einstein equaBangga G, Lutz T, Krämer E. 2018. Energy assessment of two vertical axis wind turbines in side-by-side arrangement.
tions: Strongly hyperbolic first-order reduction and solution with discontinuous Galerkin schemes. Phys Rev D.
J Renew Sustain Energy. 10(3):10.1063/1.5028199.
97(8):084053.
Biastoch A, Sein D, Durgadoo JV, et al. 2018. Simulating the Agulhas system in global ocean models — nesting vs.
Forrey R, Babb J, Stancil P, McLaughlin B. 2018. Rate constants for the formation of CS by radiative association. Mon
multi-resolution unstructured meshes. Ocean Model. 121:117-131.
Not R Astron Soc. 479(4):4727–4734.
Bordia P, Alet F, Hosur P. 2018. Out-of-time-ordered measurements as a probe of quantum dynamics. Phys Rev A.
Fröhlich K, Schneiders L, Meinke M, Schröder W. 2018. Validation of Lagrangian two-way coupled point-particle mod-
97:030103(R).
els in large-eddy simulations. Flow Turbul Combust. 101(2):1-25.
Börner PC, Kinyanjui MK, Björkman T, et al. 2018. Observation of charge density waves in free-standing 1T-TaSe2
Galindo-Lopez S, Salehi F, Cleary MJ, et al. 2018. A stochastic multiple mapping conditioning computational model in
monolayers by transmission electron microscopy. Appl Phys Lett. 113:173103.
OpenFOAM for turbulent combustion. Comput Fluids. 172:410-425.
Borsanyi S, Fodor Z, Guenther JN, et al. 2018. Higher order fluctuations and correlations of conserved charges from
Goela S, Cross G, Stukowski A, et al. 2018. Designing nanoindentation simulation studies by appropriate indenter
lattice QCD. J High Energy Phys. 2018:205.
choices: Case study on single crystal tungsten. Comput Mater Sci. 152:196-210.
Bose S, Chakravarti K, Rezzolla L, et al. 2018. Neutron-star radius from a population of binary neutron star mergers.
Harutyunyan A, Nathanail A, Rezzolla L, Sedrakian A. 2018. Electrical resistivity and Hall effect in binary neutron-star
Phys Rev Lett. 120:031102.
mergers. Euro Phys J A. 54:191.
Bradley C, Emamy N, Ertl T, et al. 2018. Enabling detailed, biophysics-based skeletal muscle models on HPC systems.
Hötzer J, Reiter A, Hierl H, et al. 2018. The parallel multi-physics phase-field framework PACE3D. J Comp Sci. 26:1-12.
Front Physiol. 9:816. Jarniven R, Vainio R, Palmroth M, et al. 2018. Ion acceleration by flux transfer events in the terrestrial magnetosheath. Chu X, Chang W, Pandey S, et al. 2018. A computationally light data-driven approach for heat transfer and hydraulic characteristics modeling of supercritical fluids: From DNS to DNN. Int J Heat Mass Tran. 123:629-636.
50 / 51
HLRS ANNUAL REPORT 2018
Geophys Res Lett. 45(4):1723-1731.
Juusola L, Pfau-Kempf Y, Ganse U, et al. 2018. A source mechanism for magnetotail current sheet flapping. Ann
Liu X, Huang G, Hu K, et al. 2018. Sharing of Na+ by three −COO– groups at deprotonated carboxyl-terminated self-
Geophys. 36:1027–1035.
assembled monolayer-charged aqueous interface. J Phys Chem-US. 122(16):9111–9116.
Kaltenbach C, Laurien E. 2018. CFD simulation of spray cooling in the model containment THAI. Nucl Eng Design.
Lode AUJ, Diorico FS, Wu E, et al. 2018. Many-body physics in two-component Bose-Einstein condensates in a cavity:
328:359-371.
fragmented superradiance and polarization. New J Phys. 20:055006.
Kaltenbach C, Laurien E. 2018. CFD simulation of aerosol particle removal by water spray in the model containment
Luitz DJ, Lazarides A, Bar Lev Y. 2018. Periodic and quasiperiodic revivals in periodically driven interacting quantum
THAI. J Aerosol Sci. 120:62-81.
systems. Phys Rev B. 97:020303(R).
Klyushin A, Jones T, Lunkenbein T, et al. 2018. Strong metal support interaction as a key factor of Au activation in CO
Marquard P, Smirnov AV, Smirnov VA, Steinhauser M. 2018. Four-loop wave function renormalization in QCD and
oxidation. ChemCatChem. 10(18):3985-3989.
QED. Phys Rev D. 97:054032.
Knodel MM, Nägel A, Reiter S, et al. 2018. Quantitative analysis of hepatitis C NS5A viral protein dynamics on the ER
Nicholson CW, Lücke A, Schmidt WG, et al. 2018. Beyond the molecular movie: dynamics of bands and bonds during
surface. Viruses. 10(1):28.
a photoinduced phase transition. Science. 362(6416):821-825.
Kohns M, Horsch M, Hasse H. 2018. Partial molar volume of NaCl and CsCl in mixtures of water and methanol by
Pandey S, Chu X, Laurien E, Weigand B. 2018 Buoyancy induced turbulence modulation in pipe flow at supercritical
experiment and molecular simulation. Fluid Phase Equilibria. 458:30-39.
pressure under cooling conditions. Phys Fluids. 30(6):065105.
Kokh DB, Amaral M, Bomke J, et al. 2018. Estimation of drug-target residence times by τ random acceleration molec-
Pásztor A. 2018. Recent results on small μB QCD from the lattice. J Phys Conf Ser. 1024:012026.
ular dynamics simulations. J Chem Theory Comput. 14(7):3859–3869. Pecher L, Schmidt S, Tonner R. 2018. Dispersion-mediated steering of organic adsorbates on a precovered silicon Köster A, Thol M, Vrabec J. 2018. Molecular models for the hydrogen age: hydrogen, nitrogen, oxygen, argon, and
surface. Beilstein J Org Chem. 14:2715-2721.
water. J Chem Eng. 63(2):305-320. Pecher L, Tonner R. 2018. Computational analysis of the competitive bonding and reactivity pattern of a bufunctional Kühne M, Börnert F, Fecher S, et al. 2018. Reversible superdense ordering of lithium between two graphene sheets.
cyclooctyne on Si(001). Theor Chem Accounts. 137:48.
Nature. 564:234-239. Pfeiffer M. 2018. Particle-based fluid dynamics: comparison of different Bhatnagar-Gross-Krook models and the Lesnicki D, Sulpizi M. 2018. A microscopic interpretation of pump−probe vibrational spectroscopy using ab initio
direct simulation Monte Carlo method for hypersonic flows. Phys Fluids. 30:106106.
molecular dynamics. J Phys Chem B. 122(25):6604–6609. Pietracaprina F, Macé N, Luitz D, Alet F. 2018. Shift-invert diagonalization of large many-body localizing spin chains. Letzgus P, Lutz T, Krämer E. 2018. Detached eddy simulations of the local atmospheric flow field within a forested
SciPost Phys. 5:045.
wind energy test site located in complex terrain. J Phys Conf Ser. 1037:072043. Pindzola MS, Loch SF. 2019. Proton impact excitation of the Na atom. J Phys B-Atom Mol Opt. 52:025202.
52 / 53
HLRS ANNUAL REPORT 2018
Pogorelov A, Schneiders L, Meinke M, Schröder W. 2018. An adaptive Cartesian mesh based method to simulate tur-
Tsoutsanis P, Antoniadis AF, Jenkins KW. 2018. Improvement of the computational performance of a parallel unstruc-
bulent flows of multiple rotating surfaces. Flow Turbulence Combust. 100(1):19-38.
tured WENO finite volume CFD code for implicit large eddy simulation. Comput Fluids. 173:157-170.
Pogorelov A, Meinke M, Schröder W. 2018. Large-eddy simulation of the unsteady full 3D rim seal flow in a one-stage
Wang B, Kronenburg A, Tufano GL, Stein OT. 2018. Fully resolved DNS of droplet array combustion in turbulent con-
axial-flow turbine. Flow Turbulence Combust. DOI:10.1007/s10494-018-9956-9.
vective flows and modelling for mixing fields in inter-droplet space. Combust Flame. 189:347-366.
Rodriguez-Gomey V, Snyder GF, Lotz JM, et al. 2018. The optical morphologies of galaxies in the IllustrisTNG simula-
Weeber R, Kreissl P, Holm C. 2018. Studying the field-controlled change of shape and elasticity of magnetic gels using
tion: a comparison to Pan-STARRS observations. Mon Not R Astron Soc. 483(3):4140-4159.
particle-based simulations. Arch Appl Mech. 1.
Roy S, Bar Lev Y, Luitz DJ. 2018 Anomalous thermalization and transport in disordered interacting Floquet systems.
Weir DJ. 2018. Gravitational waves from a first order electroweak phase transition: a brief review. Philos Trans A Math
Phys Rev B. 98:060201(R).
Phys Eng Sci. 376(2114).
Sanna S, Lichtenstein T, Mamiyev Z, et al. 2018. How one-dimensional are atomic gold chains on a substrate? J Phys
Wyrwich R, Jones TE, Günther S, et al. 2018. LEED-I⁄(V) structure analysis of the (7x√3)rect S04 phase on Ag(111):
Chem C. 122:25580-25588.
precursor to the active species of the Ag-catalyzed ehylene epoxidation. J Phys Chem C 122:26990-27004.
Saveleva VA, Wang L, Teschner D, et al. 2018. Operando evidence for a universal oxygen evolution mechanism on
Zhu Q, Xu D, Gaspari M, et al. 2018. Formation of a Malin 1 analogue in IllustrisTNG by stimulated accretion. Mon Not
thermal and electrochemical iridium oxides. J Phys Chem Lett. 9(11):3154–3160.
R Astron Soc. 480(1):L18-L22.
Schäferlein U, Keßler M, Krämer E. 2018. Aeroelastic simulation of the tail shake phenomenon. J Am Helicopter Soc. 63(3):1-17. Shterenlikht A, Margetts L, Cebamanos L. 2018. Modelling fracture in heterogeneous materials on HPC systems using a hybrid MPI/Fortran coarray multi-scale CAFE framework. Adv Eng Softw. 125:155-166. Steinmetz P, Hötzer J, Kellner M, et al. 2018. Study of pattern selection in 3D phase-field simulations during the directional solidification of ternary eutectic Al-Ag-Cu. Comput Mater Sci. 148:131-140. Tavelli M, Dumbster M. 2018. Arbitrary high order accurate space-time discontinuous Galerkin finite element schemes on staggered unstructured meshes for linear elasticity. J Comput Phys. 366:386-414. Tsoutsanis P. 2018. Extended bounds limiter for high-order finite-volume schemes on unstructured meshes. J Comp Phys. 263:69-94.
54 / 55
HLRS ANNUAL REPORT 2018
ABOUT HLRS
Inside Our Computing Room
56 / 57
HLRS ANNUAL REPORT 2018
Cray XC40 Hazel Hen Hazel Hen is at the heart of HLRS’s HPC system infrastructure. With a peak performance of 7.42 Petaflops, it is one of the most powerful HPC systems in the world (position 30 in the TOP500, November 2018). Hazel Hen entered operation in October 2015, is based on the Intel Haswell Processor and the Cray Aries network technologies, and is designed for sustained application performance and high scalability.
CPU
Intel® Xeon CPU E5-2680 v3 12 core @ 2.5 GHz
Number of nodes / cores
7,712 / 185,088
Peak performance
7.42 PFLOPS
Memory
128 GB/node
Disk storage
15 PB
Cray Urika-GX Increasingly, projects running on the Cray XC40 Hazel Hen generate large amounts of data. To provide a powerful tool for analyzing such results, HLRS installed a specialized Data Analytics Platform in December 2016. This research project gives users the ability to adapt data analytics methods for engineering applications.
Optimized software for
Spark Hadoop CGE (CrayGraph Engine)
Number of nodes
48 + 16
Cooperation with academic and industrial partners
Daimler, Porsche, Sicos BW, among others
User Profile In 2018 the Gauss Centre for Supercomputing approved 10 new large-scale projects for Hazel Hen (each project requiring more than 35 million core hours within one year), for a total of 766 million core-hours. The Partnership for Advanced Computing in Europe (PRACE) also approved 4 international simulation projects for HLRS, for a total of 113 million core-hours. In total, 129 projects were active on Hazel Hen in 2018 with 1.28 billion core-hours used.
System Usage by Scientific Discipline
NEC SX-ACE The NEC SX-ACE is a vector computer optimized for applications demanding vector operations and high memory bandwidth.
Electrical Engineering 0.73 % Computer Science 0.77 % Other 1.21 % Bioinformatics 1.39 %
CPU
NEC Vector CPU, 4 cores @ 1.0 GHz
Number of nodes / cores
64 / 256
Peak performance
~16 TFLOPS
Memory
4 TB
Memory BW per node
220 GB/s (single core), 256 GB/s (4 cores)
Interconnect
NEC IXS
Materials Science 0.73 % Structural Mechanics 0.01 %
Transportation and Climate 1.77 % Chemistry 2.18 % Solid State Physics 2.21 % Reacting Flows 3.2 %
CFD 53.69 % Physics 32.12 %
NEC Cluster System Usage by State This standard PC cluster was installed in spring 2009. To meet increasing demands for compute resources, its Lower Saxony 0.65 %
configuration has been constantly adapted. The current configuration is as follows:
Berlin 1.14 %
Node type
Intel Xeon E5-2670 (SandyBridge)
124
Hessen 1.64 %
Bavaria 0.34 % Hamburg 0.29 %
Federal Research Center 4.17 %
Node type
Intel Xeon E5-2660 v3 @ 2.6 GHz (Haswell)
88
Node type
Intel Xeon E5-2680 v3 @ 2.5 GHz (Haswell)
360
Memory per node
32 / 64 / 128 / 256 GB
Interconnect
Infiniband QDR/FDR
Rheinland-Palatinate 4.49 %
North Rhine-Westphalia 45.4 %
Baden-Württemberg 41.87 %
58 / 59
HLRS ANNUAL REPORT 2018
Third-Party Funded Research Projects
In addition to providing supercomputing resources for scientists and engineers in academia and industry, HLRS
Leadning European HPC centers December 2018 - November 2021
conducts its own funded research on important topics relevant for high-performance computing. These activities,
➜ In ChEESE, leading European HPC centers, academia, hardware developers, alongside SMEs, industry and
many of which are conducted in collaboration with investigators at other institutes, are designed to address key
public governance bodies such as civil protection are working together to prepare European flagship codes for
challenges and opportunities in the field. The following is a list of funded projects that operated in 2018.
upcoming pre-exascale and exascale supercomputing systems to tackle global challenges in the domain of
EU
solid earth. For more information about our current projects, visit www.hlrs.de/about-us/research/current-projects/ DIPL-ING
April 2017 - March 2019
BMBF
➜ The project is researching solutions for efficiently managing the high amounts of data emerging from engineer-
ing education programs at the University of Stuttgart.
Project
Duration
Funded by
BEAM-ME
December 2015 - November 2018
BMBF
EOPEN
November 2017 - October 2020
EU
➜ BEAM-ME is exploiting the potential of parallel and high-performance computing using distributed memory for
➜ EOPEN is tackling technical barriers that result from massive streams of Earth observation data and seeks to
high-resolution optimization models in energy system analyses.
ensure that methods for data harmonization, standardization, fusion, and exchange are scalable.
bw Naha 2
January 2017 - December 2019
MWK
➜ This project supports implementation of an energy management system (ISO 50001) and an environmental man-
EuroLab-4-HPC 2
May 2018 - April 2020
EU
➜ EuroLab-4-HPC aims to establish a European Research Center of Excellence for HPC systems.
agement system (EMAS), which will reduce consumption, improve environmental performance, and contribute to EUXDAT
the realization of HLRS’s sustainability strategy.
November 2017 - October 2020
EU
➜ EUXDAT provides a platform that unites HPC and cloud infrastructures to manage and process high amounts of
BW Stiftung II
October 2016 - September 2019
MWK
heterogeneous data. Its focus is to support sustainable development in agriculture.
➜ BW Stiftung supports universities and other nonprofit research institutions in Baden-Württemberg in using HLRS
ExaFLOW
computers and advises them on issues related to optimization.
October 2015 - September 2018
EU
➜ ExaFLOW is addressing key algorithmic challenges in computational fluid dynamics that will need to be solved
bwHPC-S5
July 2018 - December 2020
MWK
➜ BwHPC coordinates support for HPC users in Baden-Württemberg and the implementation of related measures
to enable simulation at exascale. It is guided by use cases of industrial relevance and will provide open-source pilot implementations.
and activities, including data intensive computing and large scale scientific data management. Exasolvers bwVisu II
August 2014 - October 2020
MWK
➜ bwVisu is developing a service for remote visualization of scientific data, ensuring high scalability through cloud
May 2016 - April 2019
DFG
➜ The exascale computers of the future are characterized by extreme parallelism. Exasolvers is combining crucial
aspects of extreme scale solving, developing methods that scale perfectly and have optimal complexity.
technologies. EXCELLERAT CATALYST
October 2016 - September 2019
MWK
December 2018 - November 2021
EU
➜ EXCELLERAT’s goal is to facilitate the development of important codes for high-tech engineering, including
➜ CATALYST researches methods for analyzing modelling and simulation data with the goal of implementing a
maximizing their scalability to ever larger computing architectures and supporting the technology transfer that will
framework that unites HPC and data analytics.
enable their uptake within the industrial environment.
60 / 61
HLRS ANNUAL REPORT 2018
EXPERTISE
March 2017 - February 2020
EU
MoeWe
July 2016 - December 2020
ESF, MWK
➜ EXPERTISE is a European training network (ETN) for the next generation of mechanical and computer science
➜ To address the long-term demand for supercomputing experts, particularly in industry, MoeWe has been devel-
engineers. Its objective is to develop advanced tools for analyzing fluid dynamics in large-scale models of turbine
oping a modular, flexible training program called the Supercomputing-Akademie.
components and to eventually enable the virtual testing of an entire machine. MontBlanc 3 FocusCoE
December 2018 - November 2021
EU
➜ FocusCoE supports EU-funded Centres of Excellence to more effectively exploit the tangible benefits of extreme
October 2015 - September 2018
EU
➜ MontBlanc 3 aims to design a new type of computer architecture capable of setting future HPC standards. The
approach is based on energy efficient solutions used in embedded and mobile devices.
scale applications for addressing scientific, industrial or societal challenges by creating a platform for the CoEs to MWK CoE: Automotive
coordinate strategic collaboration and outreach.
March 2016 - June 2018
MWK
Simulation Exzellenzcluster 2 FORTISSIMO 2
November 2015 - December 2018
EU
➜ This project is establishing and strengthening concepts for using simulation and HPC in the automotive industry.
➜ FORTISSIMO 2 supports small and medium-sized enterprises (SMEs) in accessing simulation tools on super-
The center of excellence will be developed with the support of international networks and the analysis of funding
computers, promoting an expansion of their business and improvements in their competitiveness.
opportunities.
HiDALGO
OpenForecast
December 2018 - November 2021
EU
September 2019 - August 2020
EU
➜ HiDALGO enables the assessment of Global Challenges problem statements by enabling highly accurate simu-
➜ The overall goal of OpenForecast is to deliver a novel generic service of high quality and efficiency for the Public
lations, data analytics, artificial intelligence and data visualization, but also by providing knowledge on how to inte-
Open Data Digital Service Infrastructure. This generic service combines public open data sources and high-perfor-
grate the various workflows and the corresponding data.
mance computing (HPC) to establish a new generation of services.
HPC-Europa 3
May 2017 - March 2020
EU
OSCCAR
Juni 2018 - Mai 2021.
EU
➜ HPC-Europa 3 fosters transnational cooperation among EU scientists (especially junior researchers) who work
➜ The EU Horizon 2020 research project OSCCAR — Future Occupant Safety for Crashes in Cars — is developing
on HPC-related topics such as applications, tools, and middleware.
a novel, simulation-based approach to safeguard occupants involved in future vehicle accidents.
HyForPV
October 2018 - September 2021
BMWi
PetaGCS
January 2010 - December 2019
BMBF / MWK
➜ The overall aim of HyForPV is to develop and operationalize new prediction products for the system integration
➜ PetaGCS has been supporting the procurement and operation of next-generation supercomputers at HLRS from
of photovoltaics (PV) into the energy market and smart grids by delivering highly detailed simulations of PV power
2011 to 2019. Acquisitions are coordinated by the Gauss Centre for Supercomputing.
output with very high resolution in space and time. PHANTOM InHPC-DE
November 2017 - September 2021
BMBF
December 2015 - November 2018
EU
➜ This project is addressing challenges in the development of energy-efficient parallel infrastructures in domains
➜ This collaboration aims to lay the groundwork for a standardized and distributed HPC ecosystem that integrates
such as the Internet of Things and high-performance computing, using acceleratable hardware such as GPU and
Germany’s three Tier-1 supercomputing centers. It provides funding for 100 Gbit networking and opportunities for
CPU.
high-speed data management and visualization.
62 / 63
HLRS ANNUAL REPORT 2018
Pop-CoE
October 2015 - November 2018
EU
TranSim
January 2016 - December 2018
MWK
➜ POP-CoE assesses the performance of computing applications, identifying issues affecting code performance
➜ The goal of the project Transforming Society – Transforming Simulation is to explore how computer simulation
as well as the best ways to address them.
is transforming science and the worlds of work, knowledge, and values.
PRACE
May 2017 - April 2019
EU
VirMuDeKo
März 2018 - Februar 2019
MWK
➜ PRACE supports high-impact scientific discovery and engineering R&D to enhance European competitiveness
➜ The goal of this project is to develop a pilot and demonstration framework through which virtual reality and aug-
for the benefit of society.
mented reality can support scientific collaboration in spatially distributed projects.
Reallabor Stadt:quartiere 4.0
January 2016 - December 2018
MWK
Visdral
May 2016 - December 2018
BMWi
➜ Reallabor is exploring new methods and technologies that support participatory town planning and sustainable
➜ Visdral is testing the use of 3D laser scanning and simulation technologies to virtually document and investigate
development, resulting in the development of a digital twin and a planning and decision support tool.
traffic accidents.
SimTech
December 2018 - May 2022
DFG
➜ SimTech aims at developing simulation technologies towards an integrative system science based on data inte-
grated modelling. Within the project, HLRS supports the development of efficient methods for uncertainty quantification and management. Simulated Worlds
February 2011 - June 2018
MWK
➜ Despite its importance in R&D, simulation is not widely understood. Simulated Worlds aims to bridge this gap by
making it accessible to students, the next generation of scientists. SiVeGCS
January 2017 - December 2025
BMBF / MWK
➜ Coordinates and ensures the availability of HPC resources of the Gauss Centre for Supercomputing, address-
ing issues related to funding, operation, training, and user support across Germany’s national HPC infrastructure.
Funder Abbreviations BMBF
Federal Ministry of Education and Research
➜ Smart-DASH is continuing development of the C++ template library DASH, which offers distributed data struc-
BMWi
Federal Ministry for Economic Affairs and Energy
tures with flexible data partitioning schemes and a set of parallel algorithms.
DFG
German Research Foundation
ESF
European Social Fund
EU
European Union
MWK
Baden-Württemberg Ministry for Science, Research, and Art
Smart-DASH
TalPas
May 2016 - April 2019
DFG
January 2017 - December 2019
BMBF
➜ TalPas is developing a self-optimizing, task-based approach to high-performance particle simulations.
64 / 65
HLRS ANNUAL REPORT 2018
HPC Training Courses in 2018
HLRS offered 41 courses in 2018, providing continuing professional education on a wide range of topics relevant for high-performance computing. The courses took place over 129 course-days in Stuttgart and at other locations in Germany and internationally. Approximately 1,000 trainees participated in these activities. For a current listing of upcoming courses, please visit www.hlrs.de/training/.
Date
Location
Topic
Jan 18
Garching
Introduction to Hybrid Programming in HPC (MPI+X) *
Jul 10-13
Stuttgart
Advanced C++ with Focus on Software Engineering
Feb 5-7
Paderborn
Parallel Programming with MPI and OpenMP
Aug 20-23
Zürich
Parallel Programming with MPI and OpenMP
Feb 5-7
Amsterdam
Parallel Programming with MPI and OpenMP
Sep 10-11
Innsbruck
Parallel Programming with MPI and OpenMP
Feb 12-16
Dresden
Parallel Programming & Parallel Tools
Sep 10-14
Garching
Iterative Linear Solvers and Parallelization
Feb 19-23
Siegen
Computational Fluid Dynamics
Sep 10-14
Stuttgart
Computational Fluid Dynamics
Mar 5-9
Stuttgart
CFD with OpenFOAM®
Sep 12-14
Innsbruck
Parallel Programming with MPI and OpenMP
Mar 12-13
Stuttgart
OpenMP + OpenACC GPU Directives for Parallel Accelerated Super
Sep 19
Stuttgart
Deep Learning Workshop
computers *
Sep 24-28
Siegen
CFD with OpenFOAM®
Mar 19-23
Stuttgart
Iterative Linear Solvers and Parallelization
Oct 15-19
Stuttgart
Parallel Programming and Advanced Topics in Parallel Programming *
Apr 9-12
Mainz
Parallel Programming with MPI and OpenMP
Oct 25-26
Stuttgart
Scientific Visualization
Apr 9-13
Stuttgart
Fortran for Scientific Computing *
Nov 5-6
Vienna
Parallel Programming with MPI and OpenMP
Apr 11-13
Innsbruck
Parallel Programming with MPI and OpenMP
Nov 5-9
Stuttgart
Optimization of Scaling and Node-Level Performance on Hazel Hen
Apr 23-26
Stuttgart
Optimization of Scaling and Node-Level Performance on Hazel Hen
Nov 7-9
Vienna
Parallel Programming with MPI and OpenMP
Apr 26-27
Vienna
Parallel Programming with MPI and OpenMP
Nov 19-22
Stuttgart
Advanced C++ with Focus on Software Engineering
May 7-8
Stuttgart
Scientific Visualization
Nov 26-28
Jülich
Advanced Parallel Programming with MPI and OpenMP
May 14-17
Stuttgart
Advanced C++ with Focus on Software Engineering
Nov 27 + 30
Heverlee
Parallel Programming with MPI and OpenMP
May 15-17
Vienna
Parallel Programming with MPI and OpenMP
Dec 3-7
Stuttgart
Fortran for Scientific Computing
Jun 6-7
Vienna
Introduction to Hybrid Programming in HPC (MPI+X)
Dec 13
Heverlee
Parallel Programming with MPI and OpenMP
Jun 12-13
Stuttgart
Fortran Modernization Workshop
Jun 14-15
Stuttgart
Node-Level Performance Engineering *
Jun 19
Stuttgart
Introduction to Hybrid Programming in HPC (MPI+X)
Jun 20-21
Stuttgart
Cluster Workshop
Jul 2-3
Stuttgart
Concepts of GASPI and Interoperability with Other Communication APIs *
Jul 5-6
Stuttgart
Introduction to UPC and Co-Array Fortran *
66 / 67
HLRS ANNUAL REPORT 2018
*
PRACE courses: HLRS is a member of the Gauss Centre for Supercomputing (GCS). GCS is one of ten PRACE Training Centres in the EU. The marked courses are in part sponsored by PRACE and are part of the PRACE course program.
Workshops and Conferences 2018
Date
Location
Partners
Topic
Mar 9-10
Erlangen
FAU Institute for Sociology
Simulation in the Social Sciences and the Sociology of Simulation
Apr 16-18
Stuttgart
17th HLRS/hww Workshop on Scalable Global
Structure
Institut für Höchstleistungs rechnen
Director Prof. Dr.-Ing. Dr. h.c. Dr. h.c. Prof. E.h. Michael M. Resch Secretary, Assistant
Parallel File Systems Apr 23-26
Kaliningrad
Russian Academy of
Third Annual German-Russian Workshop
Manager
Sciences, Keldysh Institute June 7-8
Stuttgart
Sep 17
Stuttgart
ZIH, TU Dresden
12th International Parallel Tools Workshop
Oct 1-2
Stuttgart
Hyperion Research
HPC User Forum
Oct 4-5
Stutgart
Oct 9-10
Stuttgart
Nachhaltigkeitstage (Sustainability Days)
Dr. Bastian Koller
High-Performance Computing in Science & Engineering: 21st Results and Review Workshop Tohoku University
28th Workshop on Sustained Simulation Performance
Oct 23-24
Stuttgart
Nov 26-27
Stuttgart
Nov 28-30
Stuttgart
Dec 11
Stuttgart
Workshop on Sustainable HPC Infrastructure HPC-Europa3, SICOS-BW
1st HPC-Europa Workshop for Small and Medium
Software & Systems
Applications & Visualization
Administration & Information
Staff Unit
Thomas Beisel
Dr. Bastian Koller
Agnes Lampke
Dr. Norbert Conrad
HPCN Production
Visualization
Administration Staff, Finances, Projects
Infrastructure
Numerical Methods & Libraries
Scalable Progamming Models & Tools
Communications & Industrial Trainings
Project & User Management, Accounting
Service Management & Business Processes
Enterprises Science and Art of Simulation Workshop 2018 SICOS-BW
2nd Industrial HPC User Roundtable
BelWü
Parallel Computing, Training & Application Services
68 / 69
HLRS ANNUAL REPORT 2018
Philosophy of Simulation
Divisions and Departments
Administration and Information
of insights. The department has expertise in tools such
Responsible for the operation of all platforms in the
on parallel programming, computational fluid dynam-
➜ Leader: Agnes Lampke
as virtual reality, augmented reality, and has designed a
compute server infrastructure. This department also
ics, performance optimization, scientific visualization,
method for integrating processing steps spread across
operates the network infrastructure necessary for HPC
programming languages for scientific computing and
multiple hardware platforms into a seamless distributed
system function and is responsible for security on net-
data in HPC. The department also organizes the review
software environment.
works and provided platforms.
process for simulation projects running at the national
Numerical Methods and Libraries Leader: Dr.-Ing. Ralf Schneider
provision for industrial clients. Additionally, it pro-
ning, controlling and bookkeeping, financial project
Scalable Programming Models and Tools Leader: Dr. José Gracia
management and project controlling, legal issues,
Conducts research into parallel programming models
Provides numerical libraries and compilers for HLRS
researchers in structural mechanics and chemistry.
human resources development, personnel adminis-
and into tools to assist development of parallel appli-
computing platforms. The department has expertise
tration, procurement and inventory, and event support.
cations in HPC. Currently the focus is on transparent
in implementing algorithms on different processors
global address spaces with background data transfers,
and HPC environments, including vectorization based
Communications and Industrial Trainings Leader: Dr. Jutta Oexle
task-parallelism based on distributed data-dependen-
on the architecture of modern computers. Department
cies, collective off-loading of I/O operations, and par-
members also conduct research related to the simula-
Supervises and executes HLRS’s communication to the
allel debugging. As a service to HLRS users, the group
tion of blood flow and bone fracture in the human body,
general public and the media. It is the central point of
also maintains part of the software stack related to pro-
and are responsible for training courses focused on
Philosophy of Science and Technology of Computer Simulation Leader: Dr. Andreas Kaminski
contact for all questions regarding the center and its sci-
gramming models, debugging, and performance anal-
programming languages and numerical methods that
Examines both how computer simulation changes sci-
entific work, and promotes new findings, achievements,
ysis tools.
are important for HPC.
ence and technology development and how society and
Project and User Management, Accounting Leader: Dr.-Ing. Thomas Bönisch
standing of knowledge and how we justify scientific
and workshops for the industrial and service sectors,
Service Management and Business Processes Leader: Michael Gienger
expanding interest in and accessibility of HPC technol-
Works on the development and operation of dynamic
Responsible for user management and accounting,
tainties about the future? And how do we deal with the
ogies and solutions beyond its traditional community
and scalable cloud computing services, particularly in a
including creating and maintaining web interfaces
uncertainties of simulation itself?
of scientific users.
business context. The group conducts research focus-
necessary for (federal) project management and data
ing on performance and availability monitoring, elastic
availability for users. The department also conducts
workflow management, and energy-efficient operation
activities related to the European supercomputing infra-
Infrastructure Leader: Marcel Brodbeck
for federated cloud environments. It also works on issues
structure (PRACE) and data management. This involves
Responsible for planning and operating facilities and
related to the establishment of high-performance com-
operating and continually developing high-perfor-
infrastructure at HLRS. This division ensures reliable
puting clouds, particularly for data intensive applications.
mance storage systems as well as conceiving new
and efficient operation of the HLRS high-performance
strategies for data management for users and projects
computing systems, provides a comfortable working
working in the field of data analytics.
environment for HLRS staff, and fosters all aspects of
Administration Leader: Agnes Lampke Manages issues related to the day-to-day operation of HLRS. Areas of responsibility include financial plan-
supercomputing center and participates in service
Applications and Visualization ➜ Leader: Dr. Bastian Koller
Visualization Leader: Dr.-Ing. Uwe Wössner Supports engineers and scientists in the visual analysis of data produced by simulations on high-perfor-
Software and Systems ➜ Leader: Thomas Beisel
mance computers. By providing technologies capable of immersing users in visual representations of their data, the department enables users to interact directly with it, reducing analysis time and enabling new kinds
70 / 71
Staff Units: Related Research
politics react to it: Does simulation change our under-
and other news from around the center. In addition, the department designs and offers training courses
vides installation and software support for academic
High-Performance Computing Network – Production (HPCN Production) Leader: Thomas Beisel
HLRS ANNUAL REPORT 2018
results? How can simulation help to overcome uncer-
energy efficient HPC operation. It is also responsible
Parallel Computing, Training and Application Services Leader: Dr. Rolf Rabenseifner Organizes HLRS’s academic continuing education program in high-performance computing, with emphases
for HLRS’s sustainability program, which encourages and supports the entire HLRS staff in acting according to principles of sustainability.
© 2019
Director, HLRS: Prof. Dr.-Ing. Dr. h.c. Dr. h.c. Prof. E.h. Michael M. Resch
High-Performance Computing Center Stuttgart (HLRS)
Leader, Department of Communications and Industrial Training: Dr. Jutta Oexle
University of Stuttgart Nobelstrasse 19 | 70569 Stuttgart | Germany
Editor: Christopher M. Williams
phone ++49 (0)711 685 - 87269
Contributing Writers: Lena Bühler (LB), Eric Gedenk (EG),
fax
Amelie Liebgott (AL), Christopher M. Williams (CW)
++49 (0)711 685 - 87209
Translations: Samantha Siegert
email
[email protected] web www.hlrs.de
Production Manager: F. Rainer Klank Photography and Images: Unless otherwise indicated, all images property of HLRS. Printing: Nino Druck GmbH Design: Zimmermann Visuelle Kommunikation www.zimmermann-online.info
This magazine is printed on paper that has been certified by FSC, the EU Ecolabel, and the Blue Angel Ecolabel.
72
HLRS ANNUAL REPORT 2018
Stuttgart
This magazine is printed on paper that has been certified by FSC, the EU Ecolabel, and the Blue Angel Ecolabel.
© Mathis Bode, Institute for Combustion Technology, RWTH Aachen
High-Performance Computing Center