Henri Bal

1

• Henri Bal

Vrije Universiteit Amsterdam
2
Our Grid Adventures onDAS-2, GridLab and
Grid'5000

• DAS-2 homogeneous Computer Science grid
• A grid that works
• Ideal for clean measurements
• Bad for heterogeneous or long-latency
  experiments
• GridLab ad-hoc testbed for one EC project
• Useful for heterogeneous experiments
• Small-scale, unstable
• Grid5000 large-scale French CS grid
• Excellent for large-scale experiments
• Bad connectivity to outside world (DAS-2)

3
Are experimental Grid platforms useful?

• Some DAS-2 (2002-2006) statistics
• 1 M investment (5 Myrinet clusters, 200 nodes)
• gt 200 users from many universities, 25 Ph.D.
  theses
• Major incentive for VL-e (Virtual Laboratory
  fore-Science) initiative ? 20 M government
  funding
• VL-e was major incentive for Big-Grid ( 30 M)
• Funding for DAS-3 (Aug. 2006)
• Collaboration DAS-3 /SURFnet 80 Gbs optical
  network
• Collaboration with Grid5000 towards a European
  scale

gt DAS has a major impact on Dutch computer
science
4
What is the experimental scope?

• Essential to distinguish between
• Experimental grids for Computer Scientists
• Proof of concept environments for application
  scientists
• Production grids (with massive data storage)
• In plain Dutch DAS / VL-e PoC / Big-Grid
• Computer Science grids are needed to
• Do clean experiments with reproducible results
• Optimize utilization degree
• Be able to change everything (OS, network
  protocols)

lt 20
5
Is there room for an international testbed?

• Need large-scale systems
• E.g., for research on P2P computing, algorithms,
  programming systems, communication protocols,
  etc.
• There are many collaborations that may boost an
  international testbed
• NL ASCI school ? DAS-1, DAS-2, DAS-3
• EU CoreGRID NoE ? ..