Ownership: A Distributed Futures System for Fine-Grained Tasks
https://www.usenix.org/conference/nsdi21/presentation/cheng







System Requirements and Challenges





Ownership: Achieving fault tolerance without giving up performance



Runtime overhead is low: keep metadata local
Scale: nested function calls






What's the smallest timescale the system can currently support, and what keeps it from being even shorter?
1ms overhead
Data serialization and deserialization
Continuations? not support. Get the future
Submit another task that depends on the previous one
Dynamic get
PreviousFault-Tolerant Replication with Pull-Based Consensus in MongoDBNextCaerus: NIMBLE Task Scheduling for Serverless Analytics
Last updated
Was this helpful?