The Design and Operation of CloudLab
Talkat USENIX ATC '19 · on Wednesday, 2019-07-10· at 11:20· USENIX ATC Track I: Grand Ballroom I–VI·
summary
CloudLab is a distributed hardware testbed spanning Utah Wisconsin and Clemson that serves 4,000 users who run over 79,000 experiments on 2,250 servers and switches. The paper measures usage patterns, shows reservations lead to longer and larger runs, and attributes 86% of mapped failures to a few error classes fixed by custom heuristics. The implemented reservation system, constraint system, and error reporting improve feasibility and predictability for large-scale experiments.
AI-written summary · about
chapters (8)
- 0:00Introduction to CloudLab
- 1:46Research mission and design
- 4:22Reservation system analysis
- 7:15Measuring reservation impact
- 10:37Error reporting mechanisms
- 12:42Constraint system benefits
- 14:33Conclusion and resources
- 16:05Q&A session
related talks
25mLessons Learned from the Chameleon Testbed
20mThe New Linaro Enterprise Lab
39mThe campus as a living lab: Collaboration with data lakes
57mWhat Is the Cloud?
16mA novel Cloud High Performance Computing infrastructure for user
16mPreview: Cloud Computing
16mServiceLab: Preventing Tiny Performance Regressions at Hyperscale through...
12mExponential Load Testing...