Promoting a Joint EU-BR Digital Future - High Performance Computing -

1. Alba Cristina Magalhaes Alves de Melo
Professor at the University of Brasilia, Brasil
CNPq Researcher - Fellowship PQ 1D
Cloudscape, July 25th 2017
Promoting a Joint EU-BR Digital Future
High Performance Computing

2. Why High Performance Computing?
SCIENCE
ITION
ULAR
HEMISTRY
ULAR ENERGY
UCTION
OHYDRATE
BOLISM
ETABOLISM
GRATIVE
BOLISM
EIN SYTHESIS
n Synthesis Overview
Acids
Acid Absorption
Acid Catabolism
n Synthesis
n Structure
n Folding and
dation
ranslational Modification
f DNA Mutations
Metabolism
TICS
HEMICAL LAB
NIQUES
emistry Folding and
Degradation
Medbullets
Team
Folding
Overview
required for a protein to
achieve a proper tertiary
protein structure
involves heat shock proteins
(Hsp)
essential for normal
protein folding
some function as
chaperones and some
function as chaperonins
the more mutated a protein, the more help it needs from
chaperones
if a protein is not folding properly, a chaperone may send it
directly for degradation
clinical relevance
cystic fibrosis
pathogenesis
3 nucleotide deletion on
chromosome 7
ΔF508 mutation in
chloride channel (CFTR)
↓ stability of the protein
and ↑ folding time
instead of insertion into
the plasma membrane
the protein is degraded in
the Golgi apparatus
↓ chloride conductance
results in ↓ Na+
and Cl
reabsorption in sweat
glands
presentation
↓ water content of mucus
which results in a thick
mucus that cannot be
cleared
respiratory
infections
nasal polyps
malabsorption
meconium
ileus
biliary
1 of 2 images
Overview image of protein folding.
Protein folding – drug design, new materials
S&TR October/November 2013Seismic Simulations
intensity was highest some distance north of the city because of the
amount of slip and the local geology.
SPECFEM3D_GLOBE, a code that accounts for Earth’s
spherical geometry and three-dimensional variations in
seismic properties and topography, simulated ground motion
and seismic-wave propagation during the 1906 earthquake
on a larger scale. Researchers ran two calculations for the
Academy project using this code—one for surface motion and
one for seismic waves penetrating deep into Earth’s interior.
In the resulting visualization, filmmakers superimposed
mantle temperature data provided by Simmons onto a slice of
the interior wave simulation. Simmons used the Livermore-
developed Earth model LLNL_G3D, which is an image of a
large-scale geologic structure generated by analyzing the arrival
times of millions of seismic waves passing through Earth. (See
S&TR, March 2009, pp. 4–12.)
Displaying mantle temperature variations in the film
helps illustrate another important point. Rodgers explains,
“Earthquakes and other kinds of geologic activity are primarily
caused by the motion of plates near the surface as they are pulled
and pushed by mantle convection. These motions in the mantle
are the result of heat trying to escape from Earth.” (See the
figure on p. 17.)
run simulations customized to our needs that show the 1906
San Francisco earthquake at many levels of detail and scales,
with a continuous camera move from a local street view to
a global level.” A single shot early in the film, presenting
the science behind the 1906 earthquake, integrates high-
resolution photography, six simulations, and a historically
accurate city model based on photographs, insurance maps, and
personal accounts.
Supercomputer-Powered Seismology
Seismic-wave research is a core competency at Livermore.
While most often associated with earthquakes, the research has
many other applications of national interest, such as nuclear
explosion monitoring, explosion forensics, energy exploration,
and seismic acoustics. When studying seismic waves,
Livermore scientists rely on two programs that run on powerful
supercomputers—the Wave Propagation Program (WPP) for
local- to regional-scale simulations and SPECFEM3D_GLOBE for
global-scale simulations.
WPP uses three-dimensional models of the local geology and
topography and the structure of the fault. Says Petersson, the
program’s creator, “WPP can work with any material model, of
any size, as long as it fits on the computer.” Because the speed
(left) A dome show visualization (courtesy of the California Academy of Sciences) of a hypothetical magnitude 7.0 earthquake along the Hayward Fault is
derived from (right) a Livermore simulation of horizontal ground motion generated by Michael Loomis and Rich Cook. The Hayward Fault represents the
highest hazard of all San Francisco Bay Area faults.
Walnut Creek
Livermore
San Jose
Hayward
Oakland
San Francisco
Berkeley
Point Reyes
Earthquake simulation, climate
*from llnl.gov*from medbullets.com
NATIONAL CENTER FOR
SUPERCOMPUTING APPLICATIONS
U N I V E R S I T Y O F I L L I N O I S A T U R B A N A - C H A M P A I G N
News
SCIENTISTS PIONEER USE OF DEEP LEARNING FOR
REAL-TIME GRAVITATIONAL WAVE DISCOVERY
01.25.18 -
Scientists at the National Center for Supercomputing Applications (NCSA), located at the University of
Illinois at Urbana-Champaign, have pioneered the use of GPU-accelerated deep learning for rapid
detection and characterization of gravitational waves. This new approach will enable astronomers to
study gravitational waves using minimal computational resources, reducing time to discovery and
increasing the scientific reach of gravitational wave astrophysics. This innovative research was
recently published in Physics Letters B.
Combining deep learning algorithms, numerical relativity simulations of black hole mergers—obtained
with the Einstein Toolkit run on the Blue Waters supercomputer—and data from the LIGO Open
Science Center, NCSA Gravity Group researchers Daniel George and Eliu Huerta produced Deep
Filtering, an end-to-end time-series signal processing method. Deep Filtering achieves similar
sensitivities and lower errors compared to established gravitational wave detection algorithms, while
being far more computationally efficient and more resilient to noise anomalies. The method allows
faster than real-time processing of gravitational waves in LIGO's raw data, and also enables new
HOME NEWS SCIENTISTS PIONEER USE OF DEEP…
Blue Waters numerical relativity
simulation of two colliding black
holes with the open source,
numerical relativity software, the
Einstein Toolkit. Authors: R. Haas
and E. Huerta (NCSA/University of
Illinois); Visualization: R. Haas.
RELATED STORIES
I mpr ove your H PC skills through
Astrophysics – black holes, galaxies
*from ncsa.illinois.edu
Quantum computing simulations
*from fz-juelich.de

3. HPC History
ROAD TO EXASCALE
Metric: Flops (Floating Point Operations per Second)
1970
250 MFlops
Cray-1
Cray
1985
2.5 GFlops
CM-2
Connection
Machine
1997
1 TFlop
ASCI Red
Intel
2008
1 PFlop
RoadRunner
IBM
2011
10 PFlops
KComputer
Fujitsu
2016
93 PFlops
TaiHuLight
Sunway
2018
122 PFlops
Summit
IBM
China and USA have announced ExaFlop machines for 2021
More likely in 2023…

4. Red Hat reaches the Summit – a new t
Network World
Red Hat reaches the Summit – a new top scientific supercomputer
Visit Save View saved
700 × 394 -­ Images may be subject to copyright. Find out more
Related images:
Get help -­ Send feedback2,282,544 cores (GPU, CPU)
122 PFlops (LINPACK)
Oak Ridge NL, USA
HPC – Leading Countries
China and USA have more than 65% of the supercomputing capacity
Top 1 – Summit (USA) Top 2 – TaiHuLight (China)
10,649,600 cores (CPU)
93 PFlops (LINPACK)
NSC, China

5. HPC – USA
• Includes TACC (Stampede):
top 15
• Staff 24/24
• Training
• Privacy, Data protection
• Active Monitoring
XSEDE NSF Founded organization to share advanced
digital services (supercomputers, storage)
*from www.xsede.org

6. HPC – EU
• Top 13: 253,600 cores, 12.3 PFlops, Eni (O&G), Italy
• Top 14: 561,408 cores (CPU+MIC), 11.96 PFlops, CEA, France
• Top 18: 312,936 cores (CPU+MIC), 8.43PFlops, Cineca, Italy
• Top 22: 156,216 cores, 6.47 PFlops, BSC, Spain
• Top 23: 114,480 cores, 6.17 PFlops, FZJ, Germany
EU countries have many PFlops supercomputers
but none at the top 10

7. HPC – Europe
• Includes Piz Daint (top 6),
Marconi (top 18),
Marenostrum (top22),
Juwels (top 23)
• Staff 24/24
• Training
• Privacy, Data protection
PRACE International non-profit organization for
HPC services provided by Spain, France, Italy,
Germany and Switzerland
*from www.prace-ri.eu

8. HPC – Brazil
• Includes Santos Dumont
(18,144 + GPU/MIC),
June/2017: top 472 (10,162
cores, 0.65 PFlops)
• All remaining 8 participants
do not have more than
1,000 cores
SINAPAD Brazilian government initiative to provide
HPC services to the academic community,
government and enterprises
*from www.prace-ri.eu
(https://commons.wikimedia.org/wiki/File%3ABrazil_Political_Map.svg)
Notícias
BRAZIL
(HTTP://BRASIL.GOV.BR)
Information access Barra GovBr
Obs: only one entry at the current top500:
Top 199, marked as “cloud provider”: 48,000 cores

9. Cooperation EU-Brazil – Past and
Present
• In the last 10 years (FP7 and H2020), EU and Brazil joined
efforts to develop advanced digital services (coordinated calls):
– Areas: IoT, Cloud, Networking, Health, Energy, etc
– 362 collaborative projects, Euro 737.6 Million (EU and Brazil)
– Many success histories
• EU-FET (Future and Emerging Technology) calls:
– Coordinated call on HPC (2018-2020) with Brazil and Mexico
– Brazil areas: (a) e-health and drug design and (b) energy

10. HPC – EU-Brazil - Future
• HPC budget EU: EuroHPC-JU (Euro 486 million –
2019-2026 – goal: top 3 in 2023 - http://eurohpc.eu/)
• HPC budget Brazil: ?
• We hope that the highly successful EU-Brazil
collaboration continues in the foreseeable future
– This cooperation is strategic and highly important
– HPC: specific areas vs open calls

11. There will always be problems
that a supercomputer is unable
to solve because human
imagination tends to infinityNCE
N
R
STRY
R ENERGY
ION
DRATE
SM
BOLISM
IVE
SM
SYTHESIS
nthesis Overview
s
Absorption
Catabolism
nthesis
ucture
ding and
n
ational Modification
A Mutations
abolism
S
stry
0 0 0 8 1Protein
Folding and
Degradation
Medbullets
Team
Folding
Overview
required for a protein to
achieve a proper tertiary
protein structure
involves heat shock proteins
(Hsp)
essential for normal
protein folding
some function as
chaperones and some
function as chaperonins
the more mutated a protein, the more help it needs from
chaperones
if a protein is not folding properly, a chaperone may send it
directly for degradation
clinical relevance
cystic fibrosis
pathogenesis
3 nucleotide deletion on
chromosome 7
ΔF508 mutation in
chloride channel (CFTR)
↓ stability of the protein
and ↑ folding time
instead of insertion into
the plasma membrane
the protein is degraded in
the Golgi apparatus
↓ chloride conductance
results in ↓ Na+
and Cl
reabsorption in sweat
glands
JOIN NOW LOGIN
Topics QBank Evidence Cases Videos Posts Groups Products Help Feedback
1 of 2 images
Overview image of protein folding.
S&TR October/November 2013Seismic Simulations
intensity was highest some distance north of the city because of the
amount of slip and the local geology.
SPECFEM3D_GLOBE, a code that accounts for Earth’s
spherical geometry and three-dimensional variations in
run simulations customized to our needs that show the 1906
San Francisco earthquake at many levels of detail and scales,
with a continuous camera move from a local street view to
a global level.” A single shot early in the film, presenting
(left) A dome show visualization (courtesy of the California Academy of Sciences) of a hypothetical magnitude 7.0 earthquake along the Hayward Fault is
derived from (right) a Livermore simulation of horizontal ground motion generated by Michael Loomis and Rich Cook. The Hayward Fault represents the
highest hazard of all San Francisco Bay Area faults.
Walnut Creek
Livermore
San Jose
Hayward
Oakland
San Francisco
Berkeley
Point Reyes
NATIONAL CENTER FOR
SUPERCOMPUTING APPLICATION
U N I V E R S I T Y O F I L L I N O I S A T U R B A N A - C H A M P A I
News
SCIENTISTS PIO
REAL-TIME GRA
01.25.18 -
Scientists at the National Cen
Illinois at Urbana-Champaign
detection and characterization
study gravitational waves usin
increasing the scientific reach
recently published in Physics
HOME NEWS SCIENTISTS PIONEER USE OF DEEP…
Blue Waters numerical relativity
simulation of two colliding black

12. Thank you!
Alba Cristina Magalhaes Alves de Melo
alves@unb.br