Development of Virtual Supercomputer Service using Academic Network

1
Development of Virtual Supercomputer Service
using Academic Network
http//www.lpds.sztaki.hu/vss
2
Partners

• Co-ordinator
• MTA SZTAKI LPDS (Laboratory of Parallel and
  Distributed Systems)
• Consortium members
• BUTE DCEIT (Department of Control Engineering and
  Information Technology)
• BUTE CIT (Centre of Information Technology)
• INNOTECH Ltd., Innovation Park of BUTE

3
Background

• Available solutions for high performance
  computing have been changed significantly
• The dominance of traditionalsupercomputers
  decreased
• Computer farms (clusters) consisting of
  general-purpose computers is gaining more and
  more interest
• Growing bandwidth of WAN technologies
• Hungarian academic computer network (operated by
  Office for National Information Infrastructure
  Development )

4
Main goalVirtual Supercomputer Service

• Efficient interconnection of geographically
  distributed computer farms
• increase the available computational power (high
  performance computing)
• more efficient utilisation of the computers(high
  throughput computing)

Academic network
5
Results- analysis, interconnection, development -

• Analysis and comparison of different methods and
  tools used for the development of virtual
  supercomputers
• GLOBUS
• CONDOR
• Successful interconnection of heterogeneous
  computer systems installed at different sites of
  academic institutes
• Development and establishment of a virtual
  supercomputer-service

6
Layers of virtual supercomputer service
Computational intensive tasks
Parallel Development
PVM
MW
MPI
Global job management
Condor-G
Middleware
Globus
Local job management
Resources
Condor
Condor
Condor
Condor
Clusters
7
Results- pilot application, knowledge-centre,
dissemination -

• Public demonstration on the usability
• pilot applications
• Establishment of a national knowledge-centre
• responsible for the distribution sharing of the
  experiences
• Dissemination of the results
• 1st 2nd US-Hungarian Workshop on Cluster and
  Grid Computing (Univ. of Wisconsin Compaq,
  Hungary)
• Seminar activity

8
International relationships

• European Grid Forum participation in Cactus
  testbed
• Collaboration with several supercomputing and
  research centres from the EU
• Simulation collision of black holes
• Successful demonstration at Supercomputing 2000,
  Dallas
• More information http//www.cactuscode.org/Deve
  lopment/Egrid.html
• GridLab (EU FP6) project from 2002
• Monitoring workpackage
• More information
• http//www.gridlab.org