The HPNumSOA Project:
High Performance Numerical Computing on Service-Oriented Architectures
This is a project on high performance grid and cluster computing between
Platform (TM) Computing, the Department of Computer Science at the
Australian National University and
the Department of Computer
Science at the University of
Adelaide. It began in July 2007 and is funded by the ARC Linkage Grant LP0669762 and
Platform (TM) Computing.
The project's Chief Investigators are Dr Peter Strazdins
(ANU) and Dr Paul
Coddington (Adelaide). Associated Platform staff are Dr Khalid
Ahmed (Senior Architect), Mr Chris Smith (Principal Product Architect),
and Dr Jingwen Wang (VP Engineering).
The project's APAI PhD scholar is Mr Jaison Mulerikkal.
The development of Performance Modelling, Evaluation and Programmability
Issues of Parallel Scientific Applications using Service Oriented
Grid-Oriented Infrastructures is of chief interest to Platform (TM)
Computing, as is evident from its products such as the Platform LSF
environment for Grids/Clusters, Platform Enterprise Grid Orchestrator
(EGO) grid resource management platform and Platform Symphony
Enterprise Grid package for financial services. The latter is largely
comprised of the Service Oriented Middleware (SOAM), which provides a
high-level infrastructure for enabling grid services.
The SOA approach has worked well for financial applications; this
project will investigate the applicability of this approach to
(scientific and engineering) numerical applications.
The overall theme is to investigate the applicability of Service
Oriented Middleware (SOAM) to high performance numerical computing.
In particular to:
- Develop programming models that extend shared memory paradigms such
as OpenMP to a distributed environment.
- Characterize the scientific applications or libraries that can be ported
to run well on a service-oriented communication infrastructures (SOAM),
running in a dynamic (EGO) and static environment.
- Investigate the programmability and optimization of such applications
including things such as tracing and debugging tools.
- Devise a performance modelling methodology that can predict the
performance of such applications, and also be used to make run-time
decisions in a dynamic environment.
Keywords: Parallel Computing, Grid Computing, Service Oriented Architecture
Computer Performance Modelling and Evaluation, Cluster Computing
Requirements:
Honours I or IIA degree in a computing-related discipline (or
equivalent qualification). Preferable attributes include experience in
any of the following areas: high performance computing, grid computing
MPI programming, complex software systems, scientific computation.
Applicants normally must be Australian citizens or residents or
New Zealand citizens
to be eligible for ARC-funded APAI_IT scholarships.
Benefits: APAI_IT PhD scholarship stipend ($25,627 pa) +
relocation costs, visits to Platform (Toronto) and an internship at
Platform (Beijing). Opportunities for cross-institutional studies at the
University of Adelaide. APAI_IT stipends are over 3 years, extendable up
to 3.5 years.
Links and Key References
- the Distributed and High
Performance Computing group at the University of Adelaide and the South Australia Partnership for Advanced
Computing.
- The HPCCommunity.org Initiative
- White Papers under products Symphony and EGO
from the Platform web site,
including
Developers Guide to Building High
Performance Service-Oriented Applications,
Grid-Enabling and Virtualizing Mission-Critical Financial Services
Applications and ENTERPRISE GRID - The Next Generation
Architecture for Capital Markets.
-
Google Video by Khaild Ahmed,
Building a Scalable Resource Management for Grid Computing, Seattle, 23 Jun 2007
- Open Cluster Stack
- The Open Group SOA Working Group,
Service-Oriented Architecture: white paper, July 2007.
- OMNI Open MP compiler
- some general references on grid programming models (including a little
on service(-oriented) models):
Soh et al,
Grid Programming Models And Environments;
Craig Lee and D. Talia,
Grid Programming Models: Current Tools, Issues and Directions
- D.A. Grove and P.D. Coddington. Modelling Message-Passing Programs
with a Performance Evaluating Virtual Parallel Machine. Performance
Evaluation, 60:165--187, 2005
- the
Accurate Performance Modelling and Prediction of Cluster Computers project,
part of the Jabberwocky Project
last modified: Peter Strazdins 05/06/08