Burst Computing: Quick, Sudden, Massively Parallel Processing on Serverless...
summary
Claims a serverless system built atop OpenWhisk that launches large jobs with a group invocation primitive (flare) to guarantee simultaneity, packs many workers into fewer containers to speed startup and enable locality, and provides a BCM middleware with zero-copy messaging for in-memory collectives. Evaluation shows 960 workers start in 1.7 s (≈40× faster than baseline), 2× speed-up on TeraSort, and a 98.5% drop in remote communication with a 13× PageRank speed-up.
AI-written summary · about
chapters (7)
- 0:00Burst Computing paradigm
- 1:46Serverless ideal for bursts
- 3:31Limitations of FaaS for stateful jobs
- 5:43Three friction points in FaaS
- 8:30Burst Computing design
- 11:21Evaluation results and gains
- 14:37Conclusion and future
related talks
36mServerless and Event-driven Applications with Cloud Foundry and OpenWhisk
23mCF Server-Less: CloudFoundry as a Platform for Server-less Computing
27mCloud Native, Event Drive, Serverless, Microservices Framework - OpenWhisk
17mCloud Computing Made Better: An AWS View on Serverless Computing
25mCF Serverless: CloudFoundry as a Platform for Serverless Computing
17mFollowing the Data, Not the Function: Rethinking Function Orchestration in Serverless
05mServerless architectures built on an open source platform
31mFullstack Serverless at Scale