Distributed test-lab: Our take

1
Distributed test-lab Our take

• Peter Druschel
• Rice University

2
Minimal requirements

• Ability to deploy, run approved
  services/experiments from any site
• Un-bureaucratic approval process
• Trace collection facilitites

3
Capability wish-list

• Service deployment, bootstrap, sw update
• Data collection
• Distributed workload generation

4
Existing building blocks to be incorporated

• Per-service VMs, class loading (Java?)
• PKI to authenticate nodes, code, services
• Universal DHT for deployment, admin, bootstrap of
  service overlays
• Simple network storage (CFS, PAST?)
• Event notification (Bayeux, CAN-mc, Scribe?)

5
Experiments

• Study
• network proximity in DHTs
• Algorithms, proximity metrics
• node availability/failure modes
• Impact on DHTs and their apps
• node heterogeneity
• DHT based multicast
• in a real overlay network

6
Stumbling blocks

• Gaining broad participation
• Arbitrating use
• Effectively controlling resource consumption
• Security needs