SHARCNET

1
SHARCNET Building a distributed HPC
Infrastructure ORION RE Summit June 2005 Hugh
Couchman, Scientific Director McMaster
University Canadian Institute for Advanced
Research
2
Overview

• What is SHARCNET?
• The provincial national context
• New infrastructure

3
What is SHARCNET?

• Shared
• Hierarchical
• Academic
• Research
• Computing
• Network

To establish a world-leading, multi-university
and college, interdisciplinary institute with an
active academic-industry partnership, enabling
forefront computational research in critical
areas of science, engineering and business. 11
universities and colleges in SW Ontario, Canada
4
Partners

• Academic
• University of Guelph
• McMaster University
• The University of Western Ontario
• University of Windsor
• Wilfrid Laurier University
• Sheridan College
• Fanshawe College

Government CFI OIT ORDCF
Private Sector Hewlett Packard Quadrics
Supercomputing World Platform Computing Nortel
Networks Bell Canada
5
Academic Partners
6
Private Sector and Government Partners
7
Building an HPC Culture

• Competitive research in an increasing number of
  disciplines requires effective use of HPC. An
  important component of SHARCNETs philosophy is
  to actively build expertise in the user community
  to ensure that Canada has discipline innovators
  in HPC. Somewhat orthogonal, or at least
  complementary, to the objective of dumbing-down
  HPC.
• HPC equipment
• System administrators
• HPC consultation/training
• Fellowship programmes

8
SHARCNET 1 Quick Facts

• 42M funding CFI/OIT/Compaq 17M
  ORDCF/Industry/Institutions 25M
• gt 960 user accounts 320 user groups
• 7 technical staff
• 13 SHARCNET Chairs from Canada(7), France(1),
  Germany(2), US(2), Mexico(1)
• 123 Fellowships (undergraduate, graduate,
  postdoctoral, visitors, senior fellowships)
• Contributed equipment gt400 processors
  aggregate capacity exceeds original CFI equipment

9
Clusters across institutions(Nov 2004)
10
Discipline use of SHARCNET
11
The Provincial National Context

• Enthusiastic partner in initiatives to promote
  sharing and effective use, including
• Long-Range Plan for HPC
• Provincial coordination
• Coordinated NSERC MFA
• CFI National Platforms Fund

12
(No Transcript)
13

14
SHARCNET 2

• CFI RFP closed Award March 2004 provincial match
  summer 2004
• 50M (19.3M each CFI, Ontario government 5.8M
  IOF over 5yrs)
• Dec 2004 (completing negotiations with vendors)
• Ongoing machine room renovations
• Equipment focus remains on cluster computing
  (leverage fixed s for maximum computational
  impact), but recognize wide range of discipline
  and user requirements
• Dedicated network (10Gb/1Gb network on ORION)
  exploring connections to other Consortia
• Infrastructure expected August, September 2005

15
SHARCNET2 Infrastructure

• Key components
• 4 primary clusters throughput utility SMP
  capability
• Mid-range SMP
• Labs experimental cluster GridLab
• Point-of-Presence clusters (inc. visualization
  capability)
• Central Archival/long-term storage
• Remote collaboration capability (AccessGrid)
• Consolidate SHARCNET1 Alphas into single cluster
• All resources available via dedicated network on
  ORION

16
(No Transcript)
17
Operational design

• All infrastructure resides on SHARCNET private
  network no obvious institutional ownership
• Infrastructure to facilitate collaboration
• SHARCNET-wide accounts single sign-on
• Uniform policies on storage, use/scheduling
• SHARCNET-wide accounting
• Aggressive scheduling policy dedicated time
  allocation
• Provide development environment cost-share
  commercial software licenses
• Rolling introduction/implementation of user
  abstractions, web services etc.