HPC 2012
High Performance Computing, GRIDS and clouds
An International
Advanced Workshop
June 25-29,
2012,
Final Programme
|
|
International Programme Committee |
|||||||||||||||||||||||||||||||||||||||||||||||||||||
Organizing Committee
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||
Sponsors
|
IBM |
|
|
|
|
|
|
|
HEWLETT PACKARD |
|
|
|
|
|
|
|
MICROSOFT |
|
|
|
|
|
|
|
|
|
|
|
JUELICH SUPERCOMPUTING
CENTER, Germany |
|
|
|
|
|
|
|
|
|
|
|
Nvidia Corporation |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Advance project |
|
|
|
|
|
|
|
Amazon Web
Services |
|
|
|
Free Amazon web Service credits for all HPC
2012 delegates Amazon is very pleased to be able to provide
$200 in service credits to all HPC 2012 delegates. Amazon Web Services
provides a collection of scalable high performance and data-intensive
computing services, storage, connectivity and integration tools. From GPUs, to tightly coupled workloads on EC2; from 50k core
scale out systems to map/reduce and Hadoop, utility
computing is a good fit for a variety of HPC workloads. For more information, visit our website: http://aws.amazon.com/hpc |
||
|
|
|
|
|
Bull |
|
|
|
|
|
|
|
The Chain Project |
|
|
|
|
|
|
|
Convey Computer |
|
|
|
|
|
|
|
Cray Inc. |
|
|
|
|
|
|
|
E4 Computer
Engineering |
|
|
|
|
|
|
|
ENEA – Italian
National Agency for New Technologies, Energy and the Environment (t.b.c.) |
|
|
|
|
|
|
|
eXludus |
|
|
|
|
|
|
|
Fujitsu |
|
|
|
|
|
|
|
Loongson |
|
|
|
|
|
|
|
National Research
Council of ICAR - Institute
for High Performance Computing and Networks |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012 Media Sponsors
|
HPCwire is
the #1 resource for news and information from the high performance computing
industry. HPCwire continues to be the portal of
choice for business and technology professionals from the academic,
government, industrial and vendor communities who are interested in high
performance and computationally-intensive computing, including systems,
software, tools and applications, middleware, networking and storage. To receive your complimentary
subscription, visit: http://www.hpcwire.com/xs/register. |
|
HPC in the Cloud is the only
portal dedicated to covering data-intensive cloud computing in science,
industry and the data center. The publication
provides technology decision-makers and stakeholders in the high performance
computing industry (spanning government, industry, and academia) with the
most accurate and current information on developments happening in the point
where high performance and cloud computing intersect. Free subscriptions for the
community! Subscribing is free! Visit:
http://www.hpcinthecloud.com/xs/register. |
|
Datanami is a news portal dedicated to
providing insight, analysis and up-to-the-minute information about emerging
trends and solutions in big data. The portal sheds light on all cutting edge
technologies including networking, storage and applications, and their effect
upon business, industry, government, and research. The publication examines
the avalanche of unprecedented amounts of data and the impact the high-end
data explosion is having across the IT, enterprise, and commercial markets. Subscriptions are complimentary! Visit: www.datanami.com |
|
Speakers Frank Baetke Global
HPC Programs Academia
and Scientific Research Hewlett
Packard Palo Alto, CA USA Natalie Bates Energy
Efficient HPC Working Group Bill Blake Cray Inc. Marian Bubak Department
of Computer Science and ACC Cyfronet, AGH and Informatics
Institute, Asmterdam THE Charlie Catlett Math
& Computer Science Div. and Computation
Institute of The Alok Choudhary Northwestern University Evanston, IL USA Marcello Coppola STMicroelectronics Advanced
System Technology Grenoble Lab Timothy David Centre
for Bioengineering Christchurh NEW
ZEALAND Erik D’Hollander Ghent University BELGIUM Beniamino Di Martino Dipartimento di Ingegneria dell'Informazione Seconda Universita' di Napoli Aversa,
CE Jack Dongarra Innovative
Computing Laboratory and USA Sudip S. Dosanjh SANDIA National Labs Albuquerque,
NM USA Ron Dror D. E. Shaw
Research New York USA Massimiliano Fatica NVIDIA Corporation Santa Clara, CA USA Ian Foster Argonne
National Laboratory and Dept. of Computer
Science The Argonne
& Chicago, IL USA Geoffrey Fox Community
Grid Computing Laboratory Guang Gao Department
of Electrical and Computer Engineering Newark, Delaware USA Carlos Garcia Garino Information
and Communication Technologies
Institute Universidad
Nacional de Cuyo Dale Geldart eXludus
Technologies, Inc. Corporate
Headquarters Montréal,
Québec Wolfgang Gentzsch HPC
Consultant formerly SUN
Microsystems and Vladimir Getov Alfredo Goldman Department
of Computer Science Sergei Gorlatch Universitaet Muenster Institut
fuer Informatik Muenster Weiwu Hu Institute
of Computing Technology Peter Kacsuk MTA
SZTAKI Odej Kao Complex
and Distributed IT Systems Technische Universitat Janusz Kowalik University of
Gdansk POLAND Thomas Lippert Institute
for Advanced Simulation and and John von
Neumann Institute for Computing (NIC) also Europen
PRACE IP Projects and of the DEEP Exascale Project Yutong Lu Bob Lucas Computational
Sciences Division Information
Sciences Institute Patrick Martin School of
Computing Queen’s
University Ken Miura Center
for Grid Research and Development National Jean-Pierre Panziera Extreme Computing Division Bull Valerio Pascucci Center for
Extreme Data Management, Analysis and Visualization, Scientific
Computing and Imaging Institute School of
Computing and Dana Petcu Computer
Science Department Tadeusz Puzniakowski Judy Qiu and Pervasive
Technology Institute Bloomington, IN USA Mark Seager INTEL Corporation Santa Clara, CA USA Alex Shafarenko Department
of Computer Science Hatfield Sunil Sherlekar Parallel
Computing Research INTEL
Labs Thomas and Alex Szalay Department
of Physics and Department
of Computer Science Baltimore,
MD USA Gregory Tallant Lockheed
Martin Aeronautics Company Kenji Takeda Microsoft
Research Cambridge UK Domenico Talia Dept. of
Electronics, Informatics and Systems Yoshio Tanaka AIST –
National Institute of Advanced Industrial Science and Technology Tsukuba William M. Tang Dept. of Astrophysical
Sciences, Plasma Physics Section Fusion
Simulation Program and Princeton
Institute for Computational Science and Engineering Princeton USA Jose Luis Vazquez-Poletti Dpt. de Arquitectura de Computadores y Automática Universidad
Complutense de Madrid Steve Wallach Convey
Computer Corporation Amy Wang Institute
for Interdisciplinary Information Sciences Akinori Yonezawa RIKEN Advanced
Institute of Computational Science and Department
of Computer Science JAPAN |
Workshop Agenda
Monday, June 25th
|
Session |
Time |
Speaker/Activity |
|
|
9:00
– 9:10 |
Welcome Address |
|
|
State of the Art and Future Scenarios |
|
|
|
9:15
– 9:45 |
J. Dongarra On the Future of High
Performance Computing: How to Think for Peta and Exascale Computing |
|
|
9:45
– 10:15 |
I. Foster |
|
|
10:15
– 10:45 |
G. FOX Scientific Computing Supported by
Clouds, Grids and Exascale Systems |
|
|
10:45
– 11:15 |
K. Takeda |
|
|
11:15
– 11:45 |
COFFEE BREAK |
|
|
11:45
– 12:15 |
A. Szalay |
|
|
12:15
– 12:45 |
S. Wallach |
|
|
12:45
– 13:00 |
CONCLUDING REMARKS |
|
|
Emerging Computer Systems
and Solutions |
|
|
|
17:00
– 17:30 |
F. Baetke |
|
|
17:30
– 18:00 |
J.P. Panziera |
|
|
18:00
– 18:30 |
W. Gentzsch |
|
|
18:30
– 19:00 |
COFFEE BREAK |
|
|
19:00
– 19:30 |
B. BLAKE Supercomputing and Big Data: where
are the real boundaries and opportunities for synergy |
|
|
19:30
– 20:00 |
S. Wallach |
|
|
20:00
– 20:10 |
CONCLUDING REMARKS |
Tuesday, June 26th
|
Session |
Time |
Speaker/Activity |
|
|
Advances in HPC
Technology and Systems I |
|
|
|
9:00
– 9:25 |
S. Sherlekar |
|
|
9:25
– 9:50 |
W. Hu |
|
|
9:50
– 10:15 |
D.
GELDART |
|
|
10:15
– 10:40 |
M.
COPPOLA From Multi-Processor
System-on-Chip to High Performance Computing |
|
|
10:40
– 11:05 |
e. d’hollander Programming and Performance of
a combined GPU/FPGA Super Desktop |
|
|
11:05
– 11:35 |
COFFEE BREAK |
|
|
11:35
– 12:00 |
M. Fatica Efficient utilization of
computational resources in hybrid clusters |
|
|
12:00
– 12:25 |
J. Kowalik Is heterogeneous computing a next
mainstream technology in HPC? |
|
|
12:25
– 12:50 |
T. PuzniakoWski |
|
|
12:50
– 13:00 |
CONCLUDING REMARKS |
|
|
Advances in HPC
Technology and Systems II |
|
|
|
17:00 – 17:30 |
S. Gorlatch A Uniform High-Level Approach to
Programming Systems with Many Cores and Multiple GPUs |
|
|
17:30 – 18.00 |
G. GAO A Codelet Based Execution
Model and Its Memory Semantics |
|
|
18:00 – 18:30 |
M.
BUBAK Environments for Collaborative
Applications on e-Infrastructures |
|
|
18:30 - 19:00 |
COFFEE BREAK |
|
|
19:00 -19:30 |
A. Yonezawa Applications on K computer and Advanced
Institute of Computational Science |
|
|
19:30 - 20:00 |
K. Miura |
|
|
20:00
– 20:10 |
CONCLUDING REMARKS |
Wednesday, June 27th
|
Session |
Time |
Speaker/Activity |
|
|
Software and Architecture
for Extreme Scale Computing I |
|
|
|
9:00 – 9:30 |
M. Seager |
|
|
9:30 – 10:00 |
R. NAIR |
|
|
10:00 – 10:30 |
T. |
|
|
10:30 – 11:00 |
B. Lucas |
|
|
11:00 – 11:30 |
COFFEE BREAK |
|
|
11:30 – 12:00 |
S. Dosanjh |
|
|
12:00 – 12:30 |
T. Lippert The EU Exascale Project DEEP -
Towards a Dynamical Exascale Entry Platform |
|
|
12:30 – 13:00 |
Y.
LU |
|
|
13:00 – 13:10 |
CONCLUDING REMARKS |
|
|
Software and
Architecture for Extreme Scale Computing II |
|
|
|
16:30 – 17:00 |
W. TANG Extreme Scale Computational Science Challenges in Fusion Energy
Research |
|
|
17:00 – 17:30 |
N. Bates |
|
|
17:30 – 18:00 |
COFFEE
BREAK |
|
|
18:00 – 20:00 |
PANEL DISCUSSION: Five
years into exascale exploration: what have we
learned? Chairman: P. Messina Participants: F. Baetke, N. Bates, W. Blake, S. Dosanjh, T. Lippert,
Y. Lu, B. Lucas, K. Miura, R. Nair, M. Seager, T. Sterling, W. Tang, S. Wallach |
Thursday, June 28th
|
Session |
Time |
Speaker/Activity |
|
|
Cloud Computing
Technology and Systems I |
|
|
|
9:00 – 9:25 |
V. Getov |
|
|
9:25 – 9:50 |
R. Martin |
|
|
9:50– 10:15 |
J. Vazquez-Poletti |
|
|
10:15 – 10:40 |
O. Kao |
|
|
10:40 – 11:05 |
D. Talia A Cloud Framework for Knowledge
Discovery Workflows on Azure |
|
|
11:05 – 11:35 |
COFFEE BREAK |
|
|
11:35 – 12:00 |
G. Fox FutureGrid exploring Next
Generation Research and Education |
|
|
12:00 – 12:25 |
P. Kacsuk Executing Multi-workflow simulations
on a mixed grid/cloud infrastructure using the SHIWA Technology |
|
|
12:25 – 12:50 |
D. Petcu Open-source platform-as-a-service:
requirements and implementation challenges |
|
|
12:50 – 13:00 |
CONCLUDING REMARKS |
|
|
Cloud Computing
Technology and Systems II |
|
|
|
15:45 – 16:10 |
Y. Tanaka Building Secure and Transparent Inter-Cloud
Infrastructure for Scientific Applications |
|
|
16:10 – 16:35 |
J. Qiu |
|
|
16:35 – 17:00 |
A. Goldman |
|
|
17:00 – 17:30 |
COFFEE BREAK |
|
|
BIG DATA and
Data-Intensive Computing |
|
|
|
17:30 – 17:55 |
V. Pascucci |
|
|
17:55 – 18:20 |
W. Gentzsch EUDAT - European scientists and
data centers turn to big data collaboration |
|
|
18:20 – 18:45 |
C. Catlett Smart Cities and Opportunities
for Convergence of Open Data and Computational Modeling |
|
|
18:45 – 19:10 |
A. Choudhary Discovering Knowledge from
Massive Social Networks and Science Data - Next Frontier for HPC |
|
|
19:15 – 20:15 |
PANEL DISCUSSION: Cloud
Computing and Big Data: Challenges and Opportunities Chairmen: C. Catlett and V. Getov Participants: A. Choudhary,
P. Martin, V. Pascucci, D. Talia |
Friday, June 29th
|
Session |
Time |
Speaker/Activity |
|
|
Challenging
Applications of HPC, Grids and Clouds |
|
|
|
9:00 – 9:25 |
G. Tallant High Performance Computing Challenges from an Aerospace
Perspective |
|
|
9:25 – 9:50 |
T. David Macro-scale phenomena of arterial coupled cells: a Massively
Parallel simulation |
|
|
9:50 – 10:15 |
R. Dror |
|
|
10:15 – 10:40 |
C. Garcia Garino Job scheduling of parametric
computational mechanics studies on cloud computing infrastructure |
|
|
10:40 – 11:05 |
V. PASCUCCI |
|
|
11:05 – 11:35 |
COFFEE BREAK |
|
|
Advanced
Infrastructures and Projects of HPC, Grids and Clouds |
|
|
|
11:35 – 12:00 |
B. Di Martino |
|
|
12:00 – 12:25 |
A. Wang Smart Sensing for Discovering and
Reducing Energy Wastes in Office Buildings |
|
|
12:25 – 12:50 |
A. SHAFARENKO Project ADVANCE: Ant Colony
Optimisation (ACO) using coordination programming based on S-Net |
|
|
12:50 – 13:00 |
CONCLUDING REMARKS |
CHAIRMEN
Paul Messina
Argonne
National Laboratory
Argonne,
IL
Gerhard Joubert
Jack Dongarra
Innovative Computing
Laboratory
and
Ian Foster
and
Department of Computer Science
The
USA
Bill Blake
Cray Inc.
Bill Blake
Cray Inc.
Wolfgang Gentzsch
HPC Consultant
formerly
SUN Microsystems and
Wolfgang Gentzsch
HPC Consultant
formerly
SUN Microsystems and
Bob Lucas
Computational Sciences
Division
Information Sciences Institute
Patrick Martin
School of Computing
Queen’s University
Patrick Martin
School of Computing
Queen’s University
PANELS
|
Five years into exascale
exploration: what have we learned? It
has already been five years since the first three workshops on exascale computing were organized. Literally dozens of
additional workshops on various aspects of exascale
computing have been held, Research&Development
efforts have been launched by various countries, computer manufacturers have
worked on roadmaps that would lead to affordable exascale
systems, and computational scientists have identified myriad exciting
advances that such systems would enable. What lessons have we learned from
these activities that might help guide the considerable additional R&D
that is needed on component technologies, system architecture integration,
programming models, system and application software? The panelists will voice
their opinions about the lessons learned and debate about the most fruitful
future directions. Chairman:
P. Messina Panelists: F. Baetke, N. Bates, W. Blake, S. Dosanjh, T. Lippert, Y. Lu, B. Lucas, K. Miura, R. Nair, M. Seager, T.
Sterling, W. Tang, S. Wallach |
|
Cloud Computing and Big Data: Challenges and
Opportunities Cloud
computing represents a fundamental shift in the delivery of information technology
services and has been changing the computing landscape over the last several
years. Concurrently, an increasing number of application areas are grappling
with challenges related to the scale and/or complexity of data - collectively
called "big data" challenges. In both areas we see commercial
successes as well as continuing research challenges. What
are the overlaps between cloud computing, particularly at global scale, and
big data? Is there room for working towards joint solutions? What classes of
"big data" problems can be addressed via a cloud approach, and are
there classes of data that are less effectively handled in a cloud
environment? In this panel session, each of the panelists will present their
position statements covering certain important aspects of this subject
followed by a discussion of the future directions for research and
development. Chairmen:
C. Catlett and V. Getov Participants:
A. Choudhary, P. Martin, V. Pascucci,
D. Talia |
ABSTRACTS
|
On the Future of High Performance Computing: How
to Think for Peta and Exascale
Computing Jack Dongarra In this
talk we examine how high performance computing has changed over the last 10-year
and look toward the future in terms of trends. These changes have had and
will continue to have a major impact on our software. Some of the
software and algorithm challenges have already been encountered, such as
management of communication and memory hierarchies through a combination of
compile--time and run--time techniques, but the increased scale of
computation, depth of memory hierarchies, range of latencies, and increased
run--time environment variability will make these problems much harder. |
|
Ian
Foster Computation Institute Argonne National Laboratory & We have made much progress over the past decade
toward effectively harnessing the collective power of IT resources
distributed across the globe. In fields such as high-energy physics,
astronomy, and climate, thousands benefit daily from tools that manage and
analyze large quantities of data produced and consumed by large collaborative
teams. But we now face a far greater challenge: Exploding
data volumes and powerful simulation tools mean that far more—ultimately
most?--researchers will soon require capabilities not so different from those
used by these big-science teams. How is the general population of researchers
and institutions to meet these needs? Must every lab be filled with computers
loaded with sophisticated software, and every researcher become an
information technology (IT) specialist? Can we possibly afford to equip our
labs in this way, and where would we find the experts to operate them? Consumers and businesses face similar challenges, and
industry has responded by moving IT out of homes and offices to so-called
cloud providers (e.g., Google, Netflix, Amazon, Salesforce),
slashing costs and complexity. I suggest that by similarly moving research IT
out of the lab, we can realize comparable economies of scale and reductions
in complexity. More importantly, we can free researchers from the
burden of managing IT, giving them back their time to focus on research and
empowering them to go beyond the scope of what was previously possible. I describe work we are doing at the Computation
Institute to realize this approach, focusing initially on research data
lifecycle management. I present promising results obtained to date with the Globus Online system, and suggest a path towards
large-scale delivery of these capabilities. |
|
Scientific Computing Supported by Clouds,
Grids and Exascale Systems Geoffrey
Fox Community Grid Computing Laboratory We analyze scientific computing into classes of
applications and their suitability for different architectures covering both
compute and data analysis cases and both high end and long tail users. We
propose an architecture for next generation Cyberinfrastructure
and outline some of the research challenges. |
|
Cloud computing for research and innovation Microsoft Research Connections EMEA Cloud computing is challenging the way we think
about parallel and distributed computing, particularly in the context of HPC andthe Grid. It opens up many possibilities for how
research, development and businesses can exploit compute, storage and
services on-demand to exploit new opportunities across the whole spectrum of
applications and domains. In this talk we discuss how the community has been
exploring the use of Cloud Computing, including through the European Union
Framework Programme 7 VENUS-C project, and the global Azure Research Engagement
programme. We conclude with thoughts on how cloud computing is potentially
reshaping the landscape of research and innovation. |
|
Extreme Data-Intensive Scientific Computing A. Szalay Department
of Physics and Department of Computer Science Scientific
computing is increasingly revolving around massive amounts of data. From
physical sciences to numerical simulations to high throughput genomics and
homeland security, we are soon dealing with Petabytes
deployed various scientific test cases, mostly drawn from astronomy, over
different architectures and compare performance and scaling laws. We discuss
a hypothetical cheap, yet high performance multi-petabyte
system currently under consideration at JHU. We will
also explore strategies of interacting with very if not Exabytes
of data. This new, data-centric computing requires a new look at computing
architectures and strategies. We will revisit Amdahl's Law establishing the
relation between CPU and I/O in a balanced computer system, and use this to
analyze current computing architectures and workloads. We will
discuss how existing hardware can be used to build systems that are much
closer to an ideal Amdahl machine. We have large amounts of data, and compare
various large scale data analysis platforms. |
|
Steve
Wallach Convey Computer Corporation Big Data has been processed for decades. Classically
the database size was constant or
gradually increasing. With the advent of searching, directed
advertisement, social networking, worldwide electronic messaging and
web-based applications, the database increases in real-time. This coupled
with the availability of petabytes of storage,
naturally leads to the need for new types of power aware computer
architectures and knowledge discovery algorithms. This talk will focus on new types of algorithms, and
architectures that are dynamically chosen based on the data type and data
base size. |
|
Technology Trends in High Performance
Computing Frank Baetke Global HPC Programs Academia and Scientific Research Hewlett Packard Palo
Alto, CA, USA HP’s HPC
product portfolio which has always been based on standards at the processor,
node and interconnect level lead to a successful penetration of the High
Performance Computing market across all application segments. The rich
portfolio of the Proliant BL-series and the
well-established rack-based Proliant DL family of
nodes has been complemented bythe SL-series with
proven Petascale scalability and leading energy efficiency.Very recently this portfolio has been extended
by a new family of severs announced under the name “Moonshot”.
Power and cooling
efficiency is primarily an issue of cost, but also extends for the power and
thermal density of what can be managed in a data center. To leverage the economics of scale established
HPC centers as well as providers of HPC Cloud
services are evaluating new concepts which have the potential to make
classical data center designs obsolete. Those new
concepts provide significant advantages in terms of energy efficiency,
deployment flexibility and manageability. Examples of this new approach,
often dubbed POD for Performance Optimized Datacenter,
including a concept to scale to multiple PFLOPS at highest energy efficiency
will be shown. Finally an outlook
will be given towards systems families due end of the decade that will
provide performance in excess of a 1000 Petaflops
or 1 Exaflop. |
|
Efficient Architecture for Exascale Applications Jean-Pierre
Panziera Extreme Computing Division Bull, France Now that more Petaflop
systems are becoming available, the HPC industry is turning to the next
challenge: Exascale. |
|
Wolfgang
Gentzsch Executive HPC Consultant Fujitsu (external) With the
K Computer installed at the RIKEN Advanced Institute for Computational
Science in While K
and its commercial PRIMEHPC FX-10 joined the top systems, the x86 based
PRIMERGY systems are completing the pyramid’s mid and bottom layers for
mainstream HPC. All Fujitsu
HPC systems are bundled into user-friendly ready-to-go solutions consisting
of HPC hardware, middleware, HPC portal, and services, providing ease-of-use
HPC for the different application segments. In addition, collaborations
enabled by the SynfiniWay integrated software
framework for virtualized distributed HPC. This
presentation aims at providing an overview of Fujitsu’s HPC solution
portfolio, from top-end supercomputing, to mid-market HPC, and technical
cloud computing. We will demonstrate how Fujitsu as the world's third-largest
IT services provider drives innovation in high performance computing for
industry and research. |
|
Supercomputing and Big Data: where are the real boundaries
and opportunities for synergy Bill Blake CTO and SVP, Cray, Inc., Supercomputing provides an increasing high fidelity view of the world
through numerically intensive modeling and
simulation techniques that support complex decision making and discoveries in
the scientific and technical fields. Big Data Analytics, as it is called
today, also provides an accurate view of the world through data intensive
search, aggregation, sorting and grouping techniques that support complex
decision making and knowledge discovery in the web and business transaction
fields. The talk will explore the architectures, data models and programming
models of Supercomputing and Big Data and in particular the implication to
Cray's Adaptive Supercomputing Vision. |
|
Steve Wallach Convey Computer Corporation An overview of the architectural aspects, both
hardware and software, of the convey’s thrust into
data intensive computing. |
|
A confluence of Technology, Architectures &
Algorithms Sunil Sherlekar Parallel Computing Research INTEL Labs For an engineer
engaged in the design of (say) an aircraft, the ideal design tool is a
computational Appliance — one that
would be optimised for his/her computational needs in terms of performance,
cost of capital (hardware), cost of operation (power consumption) and user
interface. For aircraft design, these computational needs would typically
involve computing the lift that the wings would generate and the atmosphericdrag that the aircraft would experience. A
similar scenario can be painted for any designer who uses simulation and
optimisation in his/her design flow. A custom-built appliance — right down to
the compute engines in silicon — is, however, an expensive proposition, both
in terms of design and fabrication. This becomes more so as we progress
further into nanometre semiconductor fabrication technologies. Over the
years, therefore, using general-purpose compute engines or processors has
become a commonplace. For the
last three decades or so, processors have shown a steady improvement in
performance. Most of this is an outcome of A
virtuous cycle has been established between the semiconductor industry and
application developers: while application developers eagerly use up
increasing processor performance, they also set the expectation of higher
performance from future processors. Over the
last few years, however, this fairy-tale-like increase in clock frequency has
hit a wall. This is because increasing clock frequency means increasing power
consumption. Besides the economic downside of higher operating costs for HPC,
this has now created the additional problem of dissipating the resulting
heat. The only
way to tackle the problem of heat dissipation is to produce less heat! The
only way to produce less heat is to operate the processors at a lower clock
frequency and lower operating voltage. If this is done, it also means — unfortunately
— that each processor also has a lower performance! A lower performance at
the system level is, of course, not
acceptable. The
semiconductor industry has tackled this dilemma by providing increasing
performance through a technique that the HPC community has always used:
increasing parallelism! The increasing proliferation of multi-core and
many-core chips is a result of this strategy. The
multi-core chips from Intel’s Xeon family provide for fine-grained
parallelism through vector instructions or SIMD and coarse-gained parallelism
through several cores on the same chip. This idea is taken further in Intel’s
Knights or MIC (Many-Integrated Core) family. MIC provides for even greater
parallelism through a larger SIMD width and a much larger number of cores on
one chip. KNC, the first in this family to be made commercially available,
provides a 1 TF performance on DGEMM as announced during SC’11. In the future, as we go into smaller fabrication process geometries, increasing performance will be provided through increasing parallelism on a chip while attempting to keep the power dissipation per chip constant. This will require addressing several issues: §
Reducing operating voltage while avoiding bit errors or minimising them and handling them at “higher levels”
through error correction. §
Reducing bus power by using techniques such as current-mode signalling. §
Developing circuit design techniques to handle variations in
transistor characteristics with a minimum impact on performance. §
Avoiding clocking and using “transition signalling”
where possible. The other
serious “wall” the semiconductor industry faces today is that of moving data.
This problem has two facets. One, while the speed of moving data is
increasing, it is not keeping pace with the increasing speed of computation.
This is true both for moving data to and from memory into processors and for
moving data between compute nodes in a system. This means the overall speed
of computation is increasingly being limited by the bandwidth of memory and
of interconnect networks. Secondly, the reduction in power consumed per unit
of computation is happening faster than the reduction in power consumed to
move data. This means it is getting increasingly cheaper, in terms of power
consumption, to perform computation on data than to move it around! The
technologies being pursued by the semiconductor industry to tackle the data
movement wall include: §
Bringing the memory closer to the processors and increasing the data
bus width by using the chip area and not just the perimeter (3D chip stacking
with TSV’s or Through-Silicon Vias). §
Increasing the data rate by using optical signals. While Intel’s
silicon photonics technologies help achieve this, electro-optical conversion
at a miniaturised level is still a challenge. §
Better interconnect topologies. §
Obviating the constraints of topologies by using free-space
communication using steered laser beams: still up in the air! Even if
all of the above technologies were to bear fruition, the problem will only be
alleviated; it is quite unlikely that it will actually go away. The key,
therefore, it to develop “communication avoiding” algorithms — those that
reduce data movement even at the cost of increased computation. This can be
done at several levels of abstraction. Going
forward, we at Intel are committed to expand our design strategy to encompass
a top-down approach. This means designing architectures that explicitly take the requirements of
application developers into account. In the near to mid-term, the following
are some of the ideas that may deserve consideration: §
Should the high-speed memory that can be created using 3D chip
stacking be a program-addressable memory or a (last-level) cache? §
Should we have cache memory at all or should all memory be
program-addressable? What are the implications for power consumption? §
For a program-addressable memory hierarchy — when the data traffic is
program generated and not for cache coherence — what on-chip interconnect
architectures would be most suitable? §
If all memory is program-addressable, can compiler technology
alleviate the programmer’s burden to manage date transfer between various
levels of memory? §
If, say for legacy reasons, it is necessary to have a cache hierarchy,
would it help if the cache replacement policy were to take care of the data
access patterns of a given application? Can data access patterns be characterised
for this purpose? Would it still help to allow the programmer to define
his/her own cache replacement policy? §
With a large — and perhaps increasing — SIMD width such as that on
Intel’s MIC processors, would it help if, instead of SIMD, we could carry out
more than one operation on different parts of the SIMD register? In
particular, is VLIW better than SIMD? Should the architecture allow a
programmer-controlled, application-driven trade-off? §
Are hardware blocks specific to application domains a good idea? The point
about all the above ideas — and many others — is not that they are
particularly radical. It is that evaluating their impact in terms of various
applications needs a huge investment in design time and prototyping costs. If
this analysis can be carried out without the need of prototyping, it would be
a great boon. As a first step, it would help create a formal description of
hardware that is more abstract than RTL so as to be tractable but less
abstract than ISA so as to be useful. As a
company, Intel’s commitment to application-driven architecture design is
enabled by the fact that we can optimise all aspects of the design and
fabrication process. In the final analysis, the biggest problems that need to
be solved in the long-term are
those that involve fabrication. We are also committed to ensure backward
compatibility (to support all “legacy applications) and to support a
continuity of programming paradigms (to minimise programming effort). This
brings us back to the issue of providing Design
Appliances which are tailored to specific application domains. Especially
with the increasing cost of foundries that cater to nanometre-scale
geometries, it seems impractical to use hardware that is application
specific. We can arrive at a solution, however, by looking at the exact requirements
of an HPC appliance: §
Efficient computing that can solve HPC problems in a reasonable amount
of time at a reasonable cost. §
A user-interface that is tailored to the application domain and
“talks” the language of the domain (instead of the language of computer
science or electronics). §
A service that is provided on-demand and independent of the location
of the user. A
possible way of providing such appliances would be to use the “Cloud” model.
This would entail: §
Setting up several petascale HPC systems
based on standard, general-purpose processors and a generous repertoire of
application software. §
Connecting these systems to one another and to all the users through a
high-speed network. §
Implementing application-specific, user-interface software on end-user
devices for visualisation and interaction with the
application software on the HPC systems. Besides
the continuing improvements in computing technologies, creating such
appliances will need: a) Developing highly reliable, truly high-bandwidth
wireless communication technologies. This is needed to support the transfer
of huge amounts of data that some application generate to end-user devices on
the go and b) Flexible display panels that can
be rolled up or folded to be easily carried and temporarily pinned or stuck
on walls for use. This is to support high-quality visualisation
of simulation results on the go. If this
is done, we would have created, for each application domain, a Virtual Appliance — something that
combines the customised experience of a real appliance with the economy of a
general-purpose shared system. |
|
The Chinese Godson Microprocessor for HPC Weiwu HU Institute of Computing Technology The presentation will briefly introduce the
Godson CPU roadmap for high performance computers (HPC). Servers and HPCs use the same CPU before the year of 2012. Under the
background of building 100PFLOPS HPC in the year of 2015, the CPU for HPC
should reach TeraFLOPS performance. Different CPUs will be designed for servers
and HPCs. Server CPU will take the traditional
multi-core architecture, while HPC CPU will take many-core or long vector
architecture. Bandwidth limitation and power consumption limitation
will be the big challenge for HPC CPU design. |
|
Dale Geldart eXludus Technologies, Inc. Corporate Headquarters Montréal, As core counts continue to rise, the need to safely and reliably run
more concurrent tasks on each system also increases if we are to maximize
processor and energy efficiency. Concurrently running more tasks, however,
can lead to increased shared resource conflicts that can degrade efficiency,
especially as in many cases memory per core is decreasing, which puts more
pressure on memory resources. New lightweight micro-virtualization strategies
can help users improve system efficiency while avoiding these shared resource
conflicts. |
|
From Multi-Processor System-on-Chip to High
Performance Computing Marcello
Coppola STMicroelectronics, Advanced System TechnologyGrenoble
Lab Current high-end multicore architectures when
designed for maximum speed waste available transistors, computation time,
memory bandwidth , pipeline flow (optimized for sequential operation)
resulting in a power efficiency that is one or two orders of magnitude away
from what HPC demands. Today,
architecture designed for mobile and embedded systems, employing
energy-efficient components, represent a valid alternative is to standard multicore architecture.
.In this presentation, first some example of MPSoC
architectures used in high end consumer markets is presented. Next, we
introduce how technology and innovative heterogeneous architecture could be
used to implement modern HPC. Finally we conclude the presentation showing
the power of MPSoC architectures in delivering
substantial performance improvements in high-performance computing
applications. |
|
Programming
and Performance of a combined GPU/FPGA Super Desktop Erik D’Hollander The high-performance
of GPUs have made personal supercomputing a reality
in many applications exhibiting single program multiple data parallelism.
Programs with less obvious parallelism may be accelerated by field
programmable gate arrays or FPGAs, which complement
the computing power by a very flexible and massively parallel architecture. Field programmable
gate arrays provide a programmable architecture which allows to embed an
algorithm into hardware and drive it with data streams. A multicore
CPU accelerated by GPUs and FPGAs
is a hybrid heterogeneous system with a huge computational power and a large
application area. We present a super
desktop computer consisting of a GPU and two FPGAs
and describe the interconnections, the tool chain and the programming
environment. The performance of GPUs and FPGAs as accelerators
of desktops and supercomputers is restricted by the traffic lanes between the
processor and the accelerator. The roofline model by Williams et al. is able
to represent both the raw computing performance and the input-output
bottleneck in a single graph. Whereas the roofline is completely determined
by the characteristics of processors with a fixed architecture, this is not the
case for reconfigurable processing elements such as FPGAs.
On the contrary, in this case the roofline model may be used to optimize the
resource utilization and the input-output channels as to obtain the maximum
performance for a particular application. The design and quality of different
hardware implementations of the same algorithm is enhanced by the strength of
modern high-level synthesis tools such as AutoESL
and ROCCC, which facilitate the development of powerful reconfigurable
systems. We present the results of a number of image processing algorithms
where the roofline model was used to obtain the maximum performance with a
balanced resource usage and maximum input-output yield. It is shown that the
modern high level tools vary significantly with respect to development time
and performance of the resulting computational architecture. |
|
Efficient utilization of computational resources in
hybrid clusters Massimiliano Fatica NVIDIA Corporation Santa Clara, CA,
USA Efficient
utilization of computational resources in hybrid clusters. Hybrid
clusters composed by node accelerated with Graphics Processor Units (GPUs) are moving quickly from the experimental stage into
production systems. This talk
will present two examples in which the computational workload is split
between CPU cores and GPUs in order to fully
utilize the computational capabilities of hybrid clusters. The first
example will describe a library that accelerates matrix multiplications,
currently used in the CUDA accelerated HPL code and in quantum chemistry
codes. The
second example is from TeraTF, a CFD code part of
the SPEC-MPI suite. In both
cases close to optimal performances could be achieved taking particular care
of the data movement and by using a combination of MPI, OpenMP
and CUDA. |
|
Is heterogeneous computing a next
mainstream technology in HPC? Janusz Kowalik Heterogeneous
computing is regarded as a technology on the path to the exascale
computation. However current architectural and programming trends point to
significant changes that may replace the notion of the heterogeneous
computing. by the
classic idea of SMP with massive parallelism. Hence the answer to the title
question is a good topic for a workshop discussion. |
|
Tadeusz Puźniakowski The OpenCL standard is a
relatively new standard that allows for computation on heterogeneous
architectures. The first part of the presentation summarizes basic rules and
abstractions used in OpenCL. The main part will
contain the experimental results related to a linear algebra algorithm
implemented with different methods of optimization and run on different
hardware as well as the same algorithm run using OpenMP. |
|
A Uniform High-Level Approach to Programming Systems with Many Cores
and Multiple GPUs Universitaet Münster Institut für Informatik Application programming for modern heterogeneous
systems which comprise multiple multi-core CPUs and GPUs
is complex and error-prone. Approaches like OpenCL
and CUDA are low-level and offer neither support for multiple GPUs within a stand-alone computer nor for systems that
integrate several computers. Distributed systems require programmers to use a
mix of different programming models, e.g., MPI together with Pthreads, OpenCL or CUDA. We propose a uniform approach based on the OpenCL standard for programming both stand-alone and
distributed systems with GPUs. The approach is based on two parts: 1) the SkelCL library
for high-level application programming on stand-alone computers with
multi-core CPUs and multiple GPUs, and 2) the dOpenCL
middleware for transparent execution of OpenCL
programs on several stand-alone computers connected over a network. Both parts are built on top of the OpenCL standard which ensures their high portability
across different kinds of processors and GPUs. The SkelCL library
offers a set of pre-implemented patterns (skeletons) of parallel computation
and communication which greatly simplify programming for multi-GPU systems.
The library also provides an abstract vector data type and a high-level data
(re)distribution mechanism to shield the programmer from the low-level data
transfers between a system's main memory and multiple GPUs.The
dOpenCL middleware extends OpenCL,
such that arbitrary computing devices (multi-core CPUs and GPUs) in a distributed system can be used within a single
application, with data and program code moved to these devices transparently. In this talk, we describe SkelCL
and dOpenCL and illustrate how they are used
together to simplify programming of heterogeneous HPC systems with many cores
and multiple GPUs. |
|
Environments
for Collaborative Applications on e-Infrastructures Marian Bubak Department of
Computer Science and ACC Cyfronet, AGH Institute for
Informatics, Development and
execution of e-science applications is a very demanding task. They are
collaborative, used in dynamics scenarios (similar to experiments) and there
is a need to link them with publications [8]. Most of them are used to solve
problems which are multi-physics and multi-scale what results in various
levels of coupling of applications components. Besides of being compute
intensive, more and more often they are data also intensive. This talk presents
and evaluates a few approaches to development and execution of such e-science
applications on currently available e-infrastructures like grids and clouds
[1]. Resources of these infrastructure are shared between different
organisations and may change dynamically, so there is a need for methods and
tools to master them in an efficient way [2]. We present the
WS-VLAM workflow system which aims at covering the entire life cycle of
scientific workflows: end-users are able to share workflows, reuse each other
workflow components, and execute workflow on resources across multiple
organizations [3]. GridSpace [4] is a novel virtual laboratory framework
enabling to conduct virtual experiments on grid-based infrastructures. It
facilitates exploratory development of experiments by means of scripts which
can be expressed in a number of popular languages, including Ruby, Python and
Perl. One of most demanding applications are those from the area of Virtual
Physiological Human. Cloud Data and Compute Platform enables efficient
development and execution of such applications by providing methods and tools
to install services on available resources, execute workflows and standalone
applications, and to manage data in a hybrid cloud-grid infrastructure [5]. Common Information
Space is a service-based framework for processing of sensor data streams and to
run early warning systems applications and manage their results. Although
originally it was elaborated for building and running flood early warning
systems, it may be applicable as an environment for any e-science
applications [6]. On top of the GridSpace we have elaborated an environment for composing
multi-scale applications [7] built from single scale models implemented as
scientific software components, distributed in various e-infrastructures. Applications
structure is described with the Multiscale Modelling
Language (MML). The environment consists of a semantic-aware persistence
store to record metadata about models and scales, a visual composition tool
transforming high level MML description into executable GridSpace
experiment, and finally, the GridSpace supports
execution and result management of generated experiments. The talk will be
concluded with an analysis and evaluation of these different approaches to
construction of environments supporting collaborative e-science applications. References [1] M. Bubak, T. Szepieniec, K. Wiatr (Eds.): Building a National Distributed e-Infrastructure -
Pl-Grid. Scientific and Technical Achievements. Springer, LNCS 7136, 2012. [2] J.T. Moscicki; M. Lamanna; M.T. Bubak and P.M.A. Sloot:
Processing moldable tasks on the grid: Late job
binding with lightweight user-level overlay, Future Generation Computer
Systems, vol. 27, nr 6 pp. 725-736. June 2011. ISSN 0167-739X. (DOI:
10.1016/j.future.2011.02.002) [3] Adam Belloum, Márcia A. Inda, Dmitry Vasunin, Vladimir Korkhov, Zhiming Zhao, Han Rauwerda, Timo M. Breit, Marian Bubak, Louis O. Hertzberger: Collaborative e-Science Experiments and
Scientific Workflows. IEEE Internet Computing (INTERNET) 15(4):39-47 (2011) [4] E. Ciepiela, D. Harezlak, J. Kocot, T. Bartynski, M. Kasztelnik, P. Nowakowski, T. Gubała, M. Malawski, M. Bubak: Exploratory Programming in the Virtual Laboratory.
In: Proceedings of the International Multiconference
on Computer Science and Information Technology, pp. 621-628 (October 2010),
[5] VPH-Share Cloud Platform:
http://dice.cyfronet.pl/projects/details/VPH-Share [6] Bartosz Balis, Marek Kasztelnik, Marian Bubak, Tomasz Bartynski, Tomasz
Gubala, Piotr Nowakowski, Jeroen Broekhuijsen: The UrbanFlood
Common Information Space for Early Warning Systems. Procedia
CS 4: 96-105 (2011) [7] Katarzyna Rycerz and Marian Bubak: Building and Running Collaborative Distributed Multiscale Applications, in: W. Dubitzky,
K. Kurowsky, B. Schott (Eds),
Chapter 6, Large Scale Computing, J. Wiley and Sons, 2012 [8] Marian Bubak, Piotr Nowakowski, Tomasz Gubala, Eryk Ciepiela: QUILT –
Interactive Publications, FET11 – The European Future Technologies Conference
and Exhibition, |
|
Applications
on K computer and Advanced Institute of Computational Science Akinori Yonezawa Advanced
Institute of Computational Science (AICS) Some notable
applications running on the K supercomputer will be presented, which include
Tsunami simulations and mitigation of their damage as well as simulation of a
whole human heart. Also the talk
describes Riken Advanced Institute of Computational Science (AICS) which is the
research organization for K computer and the next generation HPC. |
|
Open Petascale
Libraries (OPL) Project Dr. Kenichi Miura National and Fujitsu Laboratories Limited, With the advent of the petascale
supercomputing systems, we need to rethink the programming model and
numerical libraries. For one thing, we need to make efficient use of
multi-core CPUs. For example, the K Computer at RIKEN contains over 700
thousand cores, and features fast inter-core communication, sharing of the
programmable L2 cache, and so on. The Open Petascale
Libraries Project has been initiated by Fujitsu Laboratories of Europe (FLE)
to address this issue. It is a global collaboration that aims to promote the
development of open-source thread-parallel and hybrid numerical libraries. My
talk introduces the project, provides an update on progress, and seeks to
obtain feedback from the wider community on future directions. At this time,
the project includes: dense linear algebra, sparse solvers and adaptive
meshing, Fast Fourier Transforms, and random number generators. In
particular, I am interested in the development of highly scalable parallel
random number generators, and a wider use of the Further information on the OPL Project is
available at http://www.openpetascale.org/. |
|
Future Exascale
systems, so what’s different? Mark Seager INTEL Corporation The challenges of Exascale have been discussed at length. Addressing the
power and resiliency challenges require an aggressive near threshold voltage
(NTV) circuit designs that actually make the resiliency problem worse. In
this talk, I discuss a hierarchal approach to dealing with these issues and
also the impacts on applications, algorithms, computation &
communications methods and IO. |
|
Software
Implications of New Exascale Technologies Continuing on the high-end
high-performance computing trajectory towards Exascale
requires the overcoming of several obstacles. A lot of attention has been
paid in the community to the hardware challenges arising principally from the
slowing down of Dennard scaling. Several innovative
approaches have been proposed to dealing with these challenges. However, most
of these approaches only add to the software hurdles that already need to be
overcome in order to make Exascale systems
successful. This talk will provide examples of hardware innovations that have
been proposed or would be needed to build an Exascale
system and will describe new software challenges that these innovations would
present. |
|
Achieving Scalability in the Presence
of Asynchrony Thomas Sterling, Ph.D Professor of Informatics and Computing The last 35 years of mainstream parallel computing
have depended upon the assumption of synchronous operation; the expectation
that the time measure of actions was knowable and exploitable in the
management of physical resources and abstract actions. This was true with the
architecture and programming methods for basic vector computing of the
1970’s, the SIMD Array processing systems of the 1980’s, and the
communicating sequential processes based message passing programming of MPPs and commodity clusters of the 1990’s. This
philosophy promoted explicit programmer specification of resource management
and task scheduling with compile time assistance. Now in the Petaflops era with the inflation of number of cores
(either multicore sockets or GPU structures),
widely disparate latencies, and algorithms exhibiting increasingly irregular
structures and time varying response times, asynchronous behavior
is increasingly manifest in terms of degradation of efficiency and
limitations to scalability. Combined with the effects of overhead in
determining effective granularity, and therefore indirectly concurrency,
these factors may demand a revolutionary change to dynamic adaptive
strategies through the implementation and application of runtime system
software as an intermediary to mitigate asynchrony; the independence and
uncertainty of timing of execution events. This presentation will borrow from
the experimental ParalleX execution model to
consider a set of runtime mechanisms(some from prior art in computer science
research) that address these interrelated challenges all contributing to
growing asynchrony within high performance computing systems and their
implications for future architectures and programming methods that will
enable Exascale computing by the end of this
decade. |
|
Bob Lucas Computational Sciences Division Information Sciences Institute With the end of Dennard
scaling, there appear to be three paths forward to greater computing
capability: massive scaling of general purpose processors, purpose built
systems, or pursuit of new physical phenomenon to exploit. Adiabatic quantum
computing is an example of the latter. It is a new modeling
of computing, first proposed in 2000. The |
|
Exascale Design Space Exploration Sudip Dosanjh Extreme-scale Computing Sandia National Laboratories The U.S.
Department of Energy's mission needs in energy, national security and science
require a thousand-fold increase in supercomputing technology during the next
decade. It will not be possible to build a usable exascale
system within an affordable power budget based on computer industry roadmaps.
Both architectures and applications will need to change dramatically.
Although exascale is an important driver, these
changes will impact all scales of computing from single nodes to racks to
supercomputers. The entire computing industry faces the same power, memory,
concurrency and programmability challenges. Exascale
computing has additional challenges, notably scalability and reliability,
that are related to the extreme size of systems of interest. In order
to influence the design of future systems we must partner with computer
companies and application developers to explore the design space. Benefits of
proposed changes must be quantified relative to costs. Costs could be related
to energy and silicon area as well as software development. In order for
computer companies to adopt changes the benefits must be quantified with
trusted and validated models across a broad range of applications. The wider
this range the easier it will be to leverage industry roadmaps. In the past
it has been difficult to perform design tradeoff
studies due to the lack of validated simulation/emulation tools and the
complexity of HPC applications, which can be millions of lines of code. Our
proposed methodology for design space exploration is to use multi-scale
architectural simulation coupled with mini- and skeleton- applications to
analyze a range of abstract machine models. Close collaboration with
application teams will be needed to enable the reformulation of key
algorithms that accommodate machine constraints. |
|
The EU Exascale Project DEEP - Towards a Dynamical Exascale Entry Platform Thomas Lippert Institute for
Advanced Simulation, Jülich Supercomputing Centre and and John von
Neumann Institute for Computing (NIC) also Europen PRACE IP Projects and of the DEEP Exascale Project Since begin of |
|
Hybrid system
architecture and application Yutong Lu With more and more Petaflops systems deployed, many debates come from how we
could use them efficiently. This talk introduces the efforts on hybridarchitecture and software of Tianhe-1A to address
the performance, scalability and reliability issues. In additional, the
update applications running on the Tianhe-1A will be introduced to analyses
the usability and feasibility of the hybrid system.The
brief prospect of the next generation HPC system will also be given. |
|
Extreme Scale
Computational Science Challenges in Fusion Energy Research William M. Tang Advanced computing is
generally recognized to be an increasingly vital tool for accelerating
progress in scientific research in the 21st Century. The imperative is to
translate the combination of the rapid advances in super-computing power
together with the emergence of effective new algorithms and computational
methodologies to help enable corresponding increases in the physics fidelity
and the performance of the scientific codes used to model complex physical
systems. If properly validated against experimental measurements and verified
with mathematical tests and computational benchmarks, these codes can provide
reliable predictive capability for the behavior of
fusion energy relevant high temperature plasmas. The fusion energy research
community has made excellent progress in developing advanced codes for which
computer run-time and problem size scale well with the number of processors
on massively parallel supercomputers. A good example is the effective usage
of the full power of modern leadership class computational platforms from the
terascale to the petascale
and beyond to produce nonlinear particle-in-cell simulations which have
accelerated progress in understanding the nature of plasma turbulence in
magnetically-confined high temperature plasmas. Illustrative results provide
great encouragement for being able to include increasingly realistic dynamics
in extreme-scale computing campaigns to enable predictive simulations with
unprecedented physics fidelity. Some key aspects of application issues for
extreme scale computing will be included within this brief overview of
computational science challenges in the Fusion Energy Sciences area. |
|
Achieving the
20MW Target: Energy Efficiency for Exascale Natalie Bates Energy Efficient
HPC Working Group The growth rate in
energy consumed by data centers in the |
|
Cloud Adoption Issues:
Interoperability and Security Vladimir Getov The concept of a hybrid cloud is an attractive one
for many organisations, allowing an organisation with an existing private
cloud to partner with a public cloud provider. This can be a valuable resource
as it allows end users to keep some of their operation in-house, but benefit
from the scalability and on-demand nature of the public cloud. There are,
however, a number of issues that organisations must consider before opting
for a hybrid cloud set-up. The single most pressing issue that must be
addressed is that, by definition, the hybrid cloud is never ‘yours’ – part of
it is owned or operated by a third party, which can lead to security
concerns. With a true ‘private cloud’ – hosted entirely on your own premises,
then the security concerns are no different to those associated with any
other complex distributed system. Indeed, ‘Cloud computing’ as a term has become
very overloaded – it is doubtful whether this type of internal private cloud system
qualifies as cloud computing at all, as it does not bring the core benefits
associated with cloud computing, including taking the pressure off in-house
IT resources and providing a quickly scalable “elastic” solution using the
new pay-as-you-go business model. However, when this ‘private cloud’ is
hosted by a third party, the security issues facing end users become more
complex. Although this cloud is in theory, still private, the fact that it
relies on external resources means that IT Managers are no longer in sole
control of their data. Security remains a major adoption concern, as many
service providers put the burden of cloud security on the customer, leading
some to explore costly ideas like third party insurance. It is a huge risk,
as well as impractical, to ignore the high potential risk from losing
expensive and/or sensitive data. Another issue that organisations must
consider is interoperability – internal and external systems must work
together before security issues can be considered. It could be said, therefore, that a true hybrid
cloud is actually quite difficult to achieve, when interoperability and
security issues are considered. One solution might be a regulatory framework
that would allow cloud subscribers to undergo a risk assessment prior to data
migration, helping to make service providers accountable and provide
transparency and assurance. Concerns with hybrid cloud are indicative of the
anxiety that many companies feel when considering cloud computing as a viable
business option. We need to see a global consensus on regulation and
standards to increase trust in this technology and lower the risks that many
organisations feel goes hand-in-hand with entrusting key data or processing
capabilities to third parties. Once this hurdle is removed then the true
benefits of cloud computing can finally be realised. |
|
Qos-Aware Management of Cloud Applications Patrick Martin School of
Computing, Queen’s University Many organizations
are considering moving their applications and data to a cloud environment in
order to take advantage of its flexibility and potential cost savings. There
are numerous challenges associated with making this move including selecting
a cloud service provider, deploying and provisioning an application in the
cloud to meet required QoS levels, monitoring
application performance and dynamically re-provisioning as demand fluctuates
in order to maintain QoS commitments and minimize
costs. In the talk I will
propose a framework for QoS-aware management of
cloud applications to address these challenges. I will discuss the structure
of the framework and highlight the key research questions that must be
answered in order to develop the framework. |
|
Automatic IaaS Elasticity
for the PaaS Cloud of the Future Jose Luis Vazquez-Poletti Dpt. de Arquitectura de Computadores y Automática Universidad Complutense de Madrid Cloud computing is essentially
changing the way services are built, provided and consumed. Despite simple
access to Clouds, building elastic services is still an elitist domain and
proprietary technologies are an entry barrier especially to SMEs and consequently, it remains largely within the
domain of established players. The 4CaaSt project (http://4CaaSt.eu/) aims to
create an advanced PaaS Cloud platform which
supports the optimized and elastic hosting of Internet-scale multi-tier
applications. 4CaaSt embeds all the necessary features, easing programming of
rich applications and enabling the creation of a true business ecosystem
where applications coming from different providers can be tailored to
different users, mashed up and traded together. This talk will describe the research
efforts, involving |
|
Stratosphere - data management on the
cloud Odej Kao Complex and Distributed IT Systems Technische Universitat Data Intensive Scalable Computing is a much investigated topic in
current research. Next to parallel databases, new flavors of data processors
have established themselves - most prominently the map/reduce programming and
execution model. The new systems provide key features that current parallel
databases lack, such as flexibility in the data models, the ability to
parallelize custom functions, and fault tolerance that enables them to scale
out to thousands of machines. In this talk, we will present the Nephele
system – an execution engine for massive-parallel virtualized environments
centered around a programming model of so called Parallelization Contracts (PACTs). Nephele is part of the
large system Stratosphere, which is as generic as map/reduce systems, while
overcoming several of their major weaknesses. The focus will be set on the
underlying cloud model, the execution strategies, the detection of
communication bottlenecks and network topology, and on light-weight fault
tolerance methods. Resume: Dr. Odej Kao is a Full
Professor at the Technische Universität
Dr. Kao is a graduate from the Technische
Universität Clausthal,
where he earned a Master’s degree in Computer Science and Electrical
Engineering in 1995. Thereafter, he spent two years working on his PhD thesis
dealing with high performance image processing and defended his dissertation
in December In April 2002 Dr. Kao joined the University of Since 1998, he has published over 220 peer-reviewed
papers at prestigious scientific conferences and journals. Dr. Kao is member
of many international program committees and editorial boards of Journals
such as Parallel Computing. His research interests include Cloud computing, Virtualisation, data and resource management, Quality of
Service and SLAs, identity management, and
peer2peer based resource description and discovery. |
|
A Cloud Framework for Knowledge
Discovery Workflows on Azure Domenico Talia Dept. of Electronics, Informatics and
Systems Cloud platforms provide scalable
processing and data storage and access services that can be effectively exploited
for implementing high-performance knowledge discovery systems and applications. We
designed a Cloud framework that supports the composition and scalable
execution of knowledge discovery applications on the Windows Azure platform.
Here we describe the system architecture, its implementation, and current
work aimed at supporting the design and execution of knowledge discovery
applications modeled as workflows. |
|
Executing Multi-workflow
simulations on a mixed grid/cloud infrastructure using the SHIWA Technology Peter Kacsuk MTA SZTAKI Various scientific communities use different kind
of scientific workflow systems that can run workflows on a specific DCI (Distributed
Computing Infrastructure). The problem with the current workflow usage
scenario is that user communities are locked in their workflow system, i.e.
they cannot share their workflows with scientists using in the same field but
selected a different workflow system. They are also locked into the DCI that
is supported by the selected workflow system, i.e., they cannot run their
workflow application in another DCI that is not supported by the selected
workflow system. The SHIWA technology enables to avoid these pitfalls and
makes it possible to share workflows written in various workflow languages
among different user communities. It also enables the creation of so-called
meta-workflows that combine workflow applications into a higher level
workflow system. The other important feature of the SHIWA technology is the
support of multi-DCI execution of these meta-workflows both on various grids
and cloud systems. The talk wil describe in detail
how such meta-workflows can be created and executed on a mixed grid/cloud
infrastructure. |
|
Open-source
platform-as-a-service: requirements and implementation challenges Dana Petcu While at the infrastructure (‐as‐a‐) service level the adoption of
emerging standards is slowly progressing as solution for interoperability in
agreed or ad‐hoc
federation of Clouds, the market of platforms (‐as‐a‐) services is still struggling
with the variety of proprietary offers and approaches, leading the
application developers to a vendor lock‐in. Open‐source platforms that are
currently emerging as middleware build on top of multiple Clouds have a high
potential to help the development of applications that are vendor agnostic
and a click away from the Clouds, and, by this, to boost the migration
towards the Clouds. Due to the complexity of such platforms the number of
existing solutions is currently low. We will present a short analysis of the
available implementations, including VMware’s Cloud Foundry or Red Hat’s OpenShift, as well as with a special focus on mOSAIC’s platform [1]. While fulfilling the user
requirements, the platform needs also to automate the processes running on
the providers’ sites. In this context, special components to be developed
when implementing an opensource platform are
related to the main characteristics of the Cloud, like elasticity (through
auto‐scaling
mechanisms for example) or high availability (through adaptive scheduling for
example). The requirements of auto‐scaling and adaptive scheduling in the case
of using services from multiple Clouds will be discussed and the recent
approaches exposed in [2,3] will be detailed. [1] mOSAIC Consortium. Project details at http://www.mosaic‐cloud.eu. Platform implementation at https://bitbucket.org/mosaic. Documentation at: http://developers.mosaic‐cloud.eu. [2] N.M. Calcavecchia, B.A.Caprarescu, E. Di Nitto, D. J. Dubois, D. Petcu, DEPAS: A Decentralized Probabilistic Algorithm for Auto‐Scaling, http://arxiv.org/abs/1202.2509, 2012 [3] M. Frincu,
N. Villegas, D. Petcu, H.A. Mueller, R. Rouvoy, Self‐Healing Distributed Scheduling Platform, Procs. 11th IEEE/ACM International Symposium on Cluster,
Cloud and Grid Computing (CCGrid'11), IEEE Computer Press, 225 ‐ 234 |
|
Building Secure and Transparent
Inter-Cloud Infrastructure for Scientific Applications Yoshio Tanaka National Institute of Advanced Industrial Science
and Technology (AIST) On 11 March 2011 In this
presentation, I’ll talk about our experiences on building secure and
transparent Inter-Cloud infrastructure for scientific applications. Current
status and future issues will be presented as well. |
|
Scientific Data Analysis on Cloud
and HPC Platforms Judy Qiu and Pervasive Technology Institute We are in the era
of data deluge and future success in science depends on the ability to leverage
and utilize large-scale data. Systems such as MapReduce
have been applied to a wide range of “big data” applications and the
open-source Hadoop system has increasingly been
adopted by researchers of HPC, Grid and Cloud community. These applications
include pleasingly parallel applications and many loosely coupled data mining
and data analysis problems where we will use genomics, information retrieval
and particle physics as examples. We will introduce the key features of Hadoop and Twister (MapReduce
variant). Then, we will discuss important issues of interoperability between
HPC and commercial clouds and reproducibility using cloud computing
environments. |
|
The suitability of
BSP/CGM model for HPC on Clouds Alfredo Goldman Department of Computer Science Nowadays
the concepts and infrastructures of Cloud Computing are becoming a standard for
several applications. Scalability is not only a buzzword anymore, but is
being used effectively. However, despite the economical advantages of
virtualization and scalability, some factors as latency, bandwidth and
processor sharing can be a problem for doing High Performance Computing on
the Cloud. We will
provide an overview on how to tackle these problems using the BSP (Bulk
Synchronous Parallel). We will also introduce the main advantages of CGM
(Coarse Grained Model), where the main goal is to minimize the number of
communication rounds, which can have an important impact on BSP algorithms
performance. We will also present our experience on using BSP in an
opportunistic grid computing environment.
Then we will show several recent models for distributed computing
initiatives based on BSP. We will also provide some research directions to
improve the performance of BSP applications on Clouds. Finally we
will present some preliminary experiments comparing the performance of BSP
and MapReduce model. |
|
Big Data Analytics for Science
Discovery Valerio Pascucci Director, Center
for Extreme Data Management Analysis and Visualization (CEDMAV) Associate Director, Scientific Computing and
Imaging Institute Professor, School of Computing, University
of Laboratory Fellow, CTO, ViSUS Inc. (visus.us) Advanced techniques for analyzing and
understanding Big Data models are a crucial ingredient for the success of any
supercomputing center and data intensive scientific
investigation. Such techniques involve a number of major challenges such as
developing scalable algorithms that run efficiently on the simulation data
generated on the largest supercomputers in the world or incorporating robust
methods are provably correct and complete in their extraction of features
from the data. In this talk, I will present the application of a
discrete topological framework for the representation and analysis of large
scale scientific data. Due to the combinatorial nature of this framework, we
can implement the core constructs of Morse theory without the approximations
and instabilities of classical numerical techniques. The inherent robustness
of the combinatorial algorithms allows us to address the high complexity of
the feature extraction problem for high resolution scientific data. Our approach has enabled the successful
quantitative analysis for several massively parallel simulations including
the study turbulent hydrodynamic instabilities, porous material under stress
and failure, the energy transport of eddies in ocean data used for climate modeling, and lifted flames that lead to clean energy
production. During the talk, I will provide a live
demonstration of some software tools for topological analysis of large scale
scientific data and discuss the evolution of the organization of the project,
highlighting key aspects that enabled us to successfully deploy this new
family of tools to scientists in several disciplines. BIOGRAPHY Valerio Pascucci is the funding Director, Center for Extreme Data Management Analysis and
Visualization (CEDMAV), recently established as a permanent organization at
the |
|
EUDAT - European scientists and
data centers turn to big data collaboration Wolfgang Gentzsch Advisor, EUDAT EUDAT is a pan-European
big data project, bringing together a unique consortium of research
communities and national data and high performance omputing
centers, aiming to contribute to the production of
a collaborative data infrastructure to support The aim of this talk is to
highlight the main objectives of the EUDAT project and its Collaborative Data
Infrastructure, and to discuss a set of cross-disciplinary data services
designed to service all European research communities, such as safe
replication of data sets among different sites, data staging to compute
facilities, easy data storage, metadata, single sign-on, and persistent
identifiers. |
|
Charlie Catlett Argonne National Laboratory and The The increasing scale of new urban infrastructure projects
and the accelerating rate of demand for such projects bring into focus
several opportunities, indeed mandates, to harness information technologies
that have not been traditionally applied to urban design, development, and
evaluation. Architectural planning tools in use today rely on simplified
models, typically lacking adequate treatment of complexity, underlying
physical processes, or socio-economic factors which are at the heart of
stated city objectives such as "safe," "harmonious," or
"sustainable." To date these objectives have been difficult to
measure due to lack of data, however the trend toward transparency and public
access to "open data" is already enabling interdisciplinary
scientific analysis and performance prediction at unprecedented detail. Our
experience with cities over the past century suggests that the traditional
approach of simplified models, combined with heuristics, often produces
unintended results that are manifest only after they are difficult, or
impractical, to unravel. Embracing open data and computational modeling into urban planning and design has the potential
to radically shorten this experience loop, reducing risk while also allowing
for innovation that would be otherwise impractical. |
|
Discovering Knowledge from Massive
Social Networks and Science Data - ¬Next Frontier for HPC Prof. Alok
N. Choudhary John G. Searle Professor Electrical Engineering and Computer Science Northwestern University Knowledge
discovery in science and engineering has been driven by theory, experiments
and more recently by large-scale simulations suing high-performance
computers. Modern experiments and simulations involving satellites,
telescopes, high-throughput instruments, imaging devices, sensor networks,
accelerators, and supercomputers yield massive amounts of data. At the same
time, the world, including social communities is creating massive amounts of
data at an astonishing pace. Just consider Facebook,
Google, Articles, Papers, Images, Videos and others. But, even more complex
is the network that connects the creators of data. There is knowledge to be
discovered in both. This represents a significant and interesting challenge
for HPC and opens opportunities for accelerating knowledge discovery. In this
talk, followed by an introduction to high-end data mining and the basic
knowledge discovery paradigm, we present the process, challenges and
potential for this approach. We will present many case examples, results and
future directions including (1) mining sentiments from massive datasets on
the web, (2) Real-time stream mining of text from millions of and tweets to
identify influencers and sentiments of people; (3) Discovering knowledge from
massive social networks containing millions of nodes and hundreds of billions
of edges from real world Facebook, twitter and
other social network data (E.g., Can anyone follow Presidential campaigns and
real-time?) and (4) Discovering knowledge from massive datasets from science
applications including climate, medicine, biology and sensors. Biography:
Alok Choudhary is a John G. Searle Professor of Electrical
Engineering and Computer Science at He
received the National Science Foundation's Young Investigator Award in 1993. He
has also received an IEEE Engineering Foundation award, an IBM Faculty
Development award, an Intel Research Council award. He is a
fellow of IEEE, ACM and AAAS. His research interests are in high-performance
computing, data intensive computing, scalable data mining, computer
architecture, high-performance I/O systems and software and their
applications. Alok Choudhary
has published more than 350 papers in various journals and conferences and
has graduated 30 PhD students. Techniques
developed by his group can be found on every modern processor and scalable
software developed by his group can be found on most supercomputers. |
|
High Performance Computing
Challenges from an Aerospace Perspective Greg Tallant Lockheed Martin is a world leader in the
design development, and integration of large complex systems. In this role
Lockheed Martin is involved in many areas of technology that have an
extremely broad range of computational challenges. These challenges range
from engineering problems like computational fluid dynamics (CFD) and
structural analysis to real time signal processing and embedded systems
control. One of the biggest challenges currently being faced by Lockheed
Martin is developing more affordable system solutions that are insensitive to
increases in system complexity. To address this challenge, Lockheed Martin
has assembled a team spanning our business areas and the academic community
to explore quantum computing resources for application to new systems
engineering capabilities and reduced costs for software development and
testing. The objective of our effort is to develop a system-level
verification & validation (V&V) approach and enabling tools that
generate probabilistic measures of correctness for an entire large-scale
cyber-physical system, where V&V costs are insensitive to system
complexity. In this presentation we will provide an overview of our current
research and present some of the initial results obtained to date. |
|
Macro-scale phenomena of arterial
coupled cells: a Massively Parallel simulation Timothy David Centre for Bioengineering Impaired mass transfer characteristics of blood borne
vasoactive species such as ATP in regions such as
an arterial bifurcation have been hypothesized as a prospective mechanism in
the etiology of atherosclerotic lesions. Arterial
endothelial (EC) and smooth muscle cells (SMC) respond differentially to altered
local hemodynamics and produce coordinated
macro-scale responses via intercellular communication. Using a
computationally designed arterial segment comprising large populations of
mathematically modelled coupled ECs \& SMCs, we investigate their response to spatial gradients
of blood borne agonist concentrations and the effect of the micro-scale
driven perturbation on a macro-scale. Altering homocellular
(between same cell type) and heterocellular
(between different cell types) intercellular coupling we simulated four cases
of normal and pathological arterial segments experiencing an identical
gradient in the concentration of the agonist. Results show that the heterocellular calcium coupling between ECs and SMCs is important in
eliciting a rapid response when the vessel segment is stimulated by the
agonist gradient. In the absence of heterocellular
coupling, homocellular calcium coupling between
smooth muscle cells is necessary for propagation of calcium waves from
downstream to upstream cells axially. Desynchronized intracellular calcium
oscillations in coupled smooth muscle cells are mandatory for this
propagation. Upon decoupling the heterocellular
membrane potential, the arterial segment looses the inhibitory effect of
endothelial cells on the calcium dynamics of underlying smooth muscle cells.
The full system comprising hundreds of thousands of coupled nonlinear
ordinary differential equations simulated on the massively parallel Blue Gene
architecture. The use of massively parallel computational architectures shows
the capability of this approach to address macro-scale phenomena driven by
elementary micro-scale components of the system. |
|
Overcoming Communication Latency Barriers
in Massively Parallel Molecular Dynamics Simulation on Anton Ron Dror D. E. Shaw Research Strong scaling of scientific applications on
parallel architectures is increasingly limited by communication latency. This talk will describe the techniques used
to reduce latency and mitigate its effects on performance in Anton, a
massively parallel special-purpose machine that accelerates molecular
dynamics (MD) simulations by orders of magnitude compared with the previous
state of the art. Achieving this
speedup required both specialized hardware mechanisms and a restructuring of
the application software to reduce network latency, sender and receiver
overhead, and synchronization costs.
Key elements of Anton’s approach, in addition to tightly integrated
communication hardware, include formulating data transfer in terms of counted
remote writes and leveraging fine-grained communication. Anton delivers end-to-end inter-node
latency significantly lower than any other large-scale parallel machine, and
the total critical-path communication time for an Anton MD simulation is less
than 3% that of the next-fastest MD platform. |
|
Job
scheduling of parametric computational mechanics studies on cloud computing infrastructure Carlos
García Garino a,b , Cristian Mateos c,d
and Elina Pacini a
a Information & Communication
Technologies Institute (ITIC) and b
cISISTAN Institute, UNICEN
University, Tandil, Buenos Aires, Argentina d Consejo Nacional de
Investigaciones Científicas y Técnicas (CONICET) Parameter Sweep Experiments (PSEs) allow
scientists and engineers to conduct experiments by running the same program
code against different input data. In particular non linear Computational
Mechanical Problems are addressed in this case [1] This usually results in
many jobs with high computational requirements. Thus, distributed
environments, particularly Clouds, can be employed to fulfill these demands.
However, job scheduling is challenging as it is an NP-complete problem.
Recently, Cloud schedulers based on bio-inspired techniques–which work well
in approximating problems– have been reviewed [2]. Sadly, existing proposals
ignore job priority, which is a very important aspect in PSEs since it allows
accelerating PSE results processing and visualization in scientific Clouds. In this work a new Cloud scheduler based on
Ant Colony Optimization, the most popular bio-inspired technique, which also
exploits well known notions from operating systems theory is proposed.
Simulated experiments performed with real PSE job data and other Cloud
scheduling policies indicate that this new proposal allows for a more agile
job handling while reducing PSE completion time. [1] Pacini, E., Ribero, M., Mateos, C.,
Mirasso, A., García Garino, C.: Simulation on cloud computing infrastructures
of parametric studies of nonlinear solids problems. In: F. V. Cipolla-Ficarra
et al. (ed.) Advances in New Technologies, Interactive Interfaces and
Communicability (ADNTIIC 2011). pp. 56–68. Lecture Notes in Computer Science
(2011), to appear. [2] Pacini, E., Mateos, C., García Garino,
C.: Schedulers based on Ant Colony Optimization for Parameter Sweep
Experiments in Distributed Environments. in S. Bhattacharyya and P. Dutta
(Editors), Handbook of Research on Computational Intelligence for
Engineering, Science and Business. IGI Global, |
|
Multi-Resolution Stream Valerio Pascucci Director, Center
for Extreme Data Management Analysis and Visualization (CEDMAV) Associate Director, Scientific Computing and
Imaging Institute Professor, School of Computing, University
of Laboratory Fellow, CTO, ViSUS Inc. (visus.us) Effective use of data management techniques for
massive scientific data is a crucial ingredient for the success of any
supercomputing center and data intensive scientific
investigation. Developing such techniques involves a number of major
challenges such as the real-time management of massive data, or the
quantitative analysis of scientific features of unprecedented complexity.
Addressing these challenges requires interdisciplinary research in diverse
topics including the mathematical foundations of data representations, the
design of robust, efficient algorithms, and the integration with relevant
applications in physics, biology, or medicine. In this
talk, I will present a scalable approach for processing large scale
scientific data with high performance selective queries on multiple terabytes
of raw data. The combination of this data model with progressive streaming techniques
allows achieving interactive processing rates on a variety of computing
devices ranging from handheld devices like an iPhone,
to simple workstations, to the I/O of parallel supercomputers. I will
demonstrate how our system has enabled the real time streaming of massive
combustion simulations from DOE platforms such as Hopper2 at LBNL and
Intrepid at ANL. During the talk, I will provide a live
demonstration of the effectiveness of some software tools developed in this
effort and discuss the deployment strategies in an increasing heterogeneous
computing environment. BIOGRAPHY Valerio Pascucci is the funding Director, Center
for Extreme Data Management Analysis and Visualization (CEDMAV), recently
established as a permanent organization at the |
|
Portability and Interoperability
in Clouds: Agents, Semantic and Volunteer computing can help - the mOSAIC and Cloud@Home projects Beniamino Di Martino Second University of Naples - mOSAIC Project Coordinator Cloud vendor lock-in and interoperability gaps
arise (among many reasons) when semantics of resources and services, and of
Application Programming Interfaces is not shared. Standards and techniques borrowed from SOA and
Semantic Web Services areas might help in gaining shared, machine readable
description of Cloud offerings (resources, Services at Platform and
Application level, and their API groundings), thus allowing automatic
discovery, matchmaking, and thus supporting selection, brokering,
interoperability end even composition of Cloud Services among multiple
Clouds. The EU funded mOSAIC
project (http://www.mosaic-cloud.eu) aims at designing and developing an
innovative open-source API and platform that enables applications to be Cloud
providers' neutral and to negotiate Cloud services as requested by their
users. Using the mOSAIC Cloud ontology and Semantic
Engine, cloud applications' developers will be able to specify their services
and resources requirements and communicate them to the mOSAIC
Platform and Cloud Agency. The mOSAIC Cloud Agency
will implement a multi-agent brokering mechanism that will search for Cloud
services matching the applications’ request, and possibly compose the
requested service. The PRIN (National Relevance Research Project) Prject Cloud@Home
(http://cloudathome.unime.it/) aims at implementing a volunteer Cloud, a
paradigm which mixes aspects of both Cloud and Volunteer computing. The main
enhancement of Cloud@Home is the capability of a
host to be at the same time both contributing and consumer host, establishing
a symbiotic interaction with the Cloud@Home
environment. |
|
Smart Sensing for Discovering and
Reducing Energy Wastes in Office Buildings Amy Wang Institute for Interdisciplinary Information
Sciences Recent survey shows that in our offices up to 70%
of computers and related equipments are left on all the time. Equipment
energy costs can be reduced by 20% just by turning off when not in use.
However, it is very challenging to develop an automatic control system to
discover and reduce the energy wastes. Particularly, to discover the energy
wastes, the running states of the massive appliances need to be tracked in
real-time and checked against the real-time user requirements to judge
whether an electrical appliance is wasting energy or not. Because the
electrical appliances are massive and the user requirements are highly
dynamic, it is generally very difficult and cost inefficient to track the
states of the electrical appliances and the real-time user requirements. In
this talk, we report how the recent advantages of smart metering and
compressive sensing technologies can be exploited to solve above challenging
problems. Although the real-time electrical appliance states and the real-time
user requirements compose very high dimensional dynamic signals, they are
converted to sparse signals by temporal and spatial transformations
respectively. Compressive sensing systems by smart meters and infrared
sensors are designed to track these sparsified
signals using lightweight metering and sequential decoding. Particularly in
this talk, the design methodologies, theoretical bounds and experimental
results will be introduced. |
|
Project ADVANCE: Ant Colony
Optimisation (ACO) using coordination programming based on S-Net Alex Shafarenko Department of Computer Science This talk presents some of the results of the EU
Framework 7 project ADVANCE. We report our experiences of applying an HPC
structuring technique: dataflow coordination programming, and the specific
programming environment: the language S-Net, to restructuring existing
numerical code developed by SAP AG. The code implements an ACO solution to the
Travelling Salesman Problem. We have converted the ACO algorithm to a
stream-processing network and encoded it as a coordination program. We then
implemented this solution by using either explicit thread management in C (a
manually coded version) or by applying our coordination compiler, and
compared the results. We find that the use of S-Net results in a low
code-development cost while achieving the same scaling characteristics and very similar
performance compared to the manually coded solution at large system sizes.
The message-driven (as opposed to message-passing) nature of the coordinating
streaming code creates the prerequisites for a large scale distributed, but
still easily manageable and maintainable implementation. We argue that it is that
maintainability and manageability that makes our approach uniquely suitable
for industrial uptake of HPC. |
