This workshop is a SHARCNET-wide event primarily targetted at academic
researchers in Ontario. Talks will be held on Days 1 and 4 with no
fees or limits on places for academic attendees. Days 2 and 3 will
consist of an intensive course run by Hewlett-Packard on using the
Alpha hardware and standard (parallel) programming tools. There will
be a $200 fee for this portion and a limited number of places.
We will be providing financial assistance with
registration, travel and accomodation costs to students.
We estimate the amount will start at $50-$100 per student.
Exact details and amounts will be determined when
we know the level of demand.
This workshop is being sponsored by SHARCNET and C3.ca. C3.ca is an umbrella organisation
representing the High Performance computing (HPC) interests of
Canadian researchers, universities and vendors. It sponsors workshops
across Canada, funds HPC support activity and is currently in the
process of developing a Long Range Plan for Canadian HPC. For other
c3.ca sponsored workshops, please see
Day 1, Tuesday Oct 22
Welcome, Introduction to SHARCNET James Wadsley, SHARCNET
This talk will give an overview of the SHARCNET facilities and
support, including hardware, personnel and funding. We will describe
how to access the hardware, including the new serial farm at McMaster.
SHARCNET has HPC (High Performance Computing) Consultants to help
researchers develop new HPC applications and parallelize existing
codes. SHARCNET's research support includes faculty teaching
buy-outs, research visitors, post-docs, graduate and summer students.
Some HPC Research Highlights will be presented and future plans
discussed (eg. new CFI applications).
Parallel Computing Overview Ken Tan, SHARCNET
This talk will introduce users to parallel programming. It will cover
reasons to parallelize (eg. speed, memory limitations) and help users
assess which techniques (eg. OpenMP or MPI) and machines
(eg. Shared Memory, High Bandwidth Clusters, Beowulf Clusters) are
best suited to their applications. Measures of efficiency of parallel
codes and the limitations of parallel computing will be discussed.
Basic familiarity with Unix and Fortran and/or C will be assumed.
Programming with OpenMP James Wadsley, SHARCNET
This talk will introduce compiler directive based parallel
programming. On the HP AlphaServer SC system, OpenMP is a relatively
easy way to exploit shared memory architectures (eg. ES40 nodes of
Idra (4 processors) and the 16-way SMP machine,Typhon). With this
approach work is divided among a team of light weight threads that run
simultaneously on the same or nearby data. The types of code that are
suited to OpenMP are discussed, including potential gains and
limitations. Basic OpenMP constructs are described with short code
examples. Basic familiarity with Unix and Fortran and/or C will be assumed.
Programming with MPI Baolai Ge, SHARCNET
The talk with introduce Message Passing as a method for parallelism on
distributed memory architectures (clusters) that also works for shared
memory. Basic components of the MPI (Message Passsing Interface) api
will be covered, including the basic atom send/receive and collective
data movement operations. Explicit data distribution and movement
(ie. Domain Decomposition) and work division (ie. Load Balancing) will
be discussed. Basic familiarity with Unix and Fortran and/or C will
Day 2, Wednesday Oct 23
Programming the HP Alphaserver SC Supercomputer, I HP
Day 3, Thursday Oct 24
Programming the HP AlphaServer SC Supercomputer, II HP
Days 2 and 3 are a hands on introduction to the HP AlphaServer SC
Supercomputer. The objective of this session is to teach users how to
program HP's AlphaServer SC systems using instructor lead lectures and
hands-on workshops. An intensive overview of the HPTC environment for
AlphaServer SC ystems is provided for technical users.
The course will explain and demonstrate how to develop and optimize
applications for performance and exploit the power of AlphaServer SC
systems. Topics and examples will include compiling MPI
message-passing codes, executing those cdes with batch and cluster
management software, debugging those codes with the Etnus TotalView
debugger and tuning those codes the Pallas VAMPIR event-analyser.
SHARCNET is installing these tools on all the SHARCNET clusters, not
just those running Tru64 unix. Most of the information and techniques
covered by this course will be just as useful for people working with
Day 4, Friday Oct 25
Globus Unleashed: Platform's Commercial Support of the Globus Toolkit(TM) Ian Lumb, Platform Computing Inc.
Platform Globus is the industry's first commercially supported distribution
of the Globus Toolkit. By way of analogy with Linux distribution providers
and the Linux kernel, Platform Globus represents the redhatted version of
the `grid kernel' based on the Globus Toolkit. By focusing on pre-built
distributions, documentation, support, training and professional services,
plus enhanced interoperability with Platform LSF, Platform Globus remains
consistent with the vision and mandate of the Globus Project. This
presentation provides an introduction to Platform Globus, and updates on
several projects that are currently underway.
Globus Testbed: Running non-trivial applications over the Grid. , Dr. Gabriel Mateescu, NRC
Gabriel Mateescu has been responsible for the deployment and
customization of middleware for cluster and grid computing. He has
extended the Portable Batch System (PBS), a cluster scheduling
sys-tem, to support a flexible scheduling strategy suitable for
dynamic workloads. He has deployed Globus to several sites across
Canada, from Fredericton NB, to Montreal, to Ottawa, and has built
several extensions to Globus, from adding new information providers to
parallel job submission tools. In addition, Gabriel has contributed to
the construction of several Beowulf clusters and has designed and
implemented new algorithms for load balancing and collective
communication for parallel and distributed applications. Dr. Mateescu
will describe his experiences and outline of the steps necessary for
users to port their applications to the Globus framework.
Dr. James Wadsley is an HPC Consultant and Researcher with
SHARCNET at McMaster University. His interests include massively
parallel computation, computational fluid dynamics and astrophysics.
He holds a Ph.D. in Astronomy from the University of Toronto.
He is a co-author of the parallel comsology gas/N-body code gasoline.
Dr. Kenneth Tan is an HPC Consultant with SHARCNET
at the University of Guelph. His expertise is in the areas of
computational linear algebra, parallel stochastic simulation,
computational number theory, and high performance computing. He is
the inventor of the PLFG parallel random number generator, now part of
the OptimaNumerics Parallel Random Number Generator Library marketed
by Heuchera Technologies, as well as a number of randomized linear
solvers for massively parallel machines.
Baolai Ge is an HPC Consultant and Researcher with SHARCNET at
the University of Western Ontario. He hold an M.Sc. in Mathematics. His research area is in
numerical analysis and scientific computing. He joined SHARCNET
from Liberate Technologies, where he was a system programmer for a few
Hewlett-Packard HPC Workshop Instructors are personnel from HP's
Consulting and Integration Services and High Performance Computing
Ian Lumb , Grid Solutions Manager, Platform Computing has led a career in
scientific research and distributed computing on an international scale. He
joined Platform Computing Inc. in August 1998 to establish the company's
training practice. He subsequently served as the technical liaison for the
company's system vendor partners, before focusing on Grid Computing. Prior
to joining Platform, he served as a researcher and educator in physical
sciences at Toronto's York University, where he also got his introduction
to distributed computing.
Dr. Gabriel Mateescu works as a systems and applications analyst
at the National Research Coun-cil, in the Research Computing Support
Group. Gabriel holds a Ph.D. in Computer Science from Virginia Tech,
Blacksburg, VA, USA. Before joining NRC, Gabriel has worked at
Platform Computing in Toronto, participating in the development of the
Load Sharing Facility (LSF) distributed workload management
system. Previously, he has worked in scientific computing at Los
Alamos National Laboratory and at Virginia Tech, developing parallel
scalable software and algorithms for solving very large systems of
linear equations and for fluid flow problems.
Anticipated Audience and Fee
This is the first annual SHARCNET wide workshop. SHARCNET runs
introductory workshops at the local sites on a regular basis with
hands-on components. This workshop will be more intensive, bringing
in expertise from outside.
Our target audience is Canadian researchers interested in HPC and
parallel applications. Our goal is to attract users from
throughout Ontario. The number of attendees for the talk session will
be limited by the room sizes. The HP Workshop sessions will have a
limited number of places.
We intend to charge $200 per academic user for the HP Workshops
portion. For non-academic users please contact the organisers.
Travel assistance will be available for Graduate students
coming from outside McMaster. You will be able to apply through
the registration form.