Elastic Scaling of Stateful Network Functions
summary
S6 implements a distributed shared object (DSO) space that keeps state objects local for common accesses but makes them visible and migratable across instances, compiles object methods into microthreads in C++ and runs on EC2 containers, and avoids system-wide pauses during rehashing by gradual per-object migration and packet-context checkpointing. Evaluation shows 100x improvement during scaling events and 2–5x higher throughput in steady state, with median latency staying sub-millisecond during large scale-outs.
AI-written summary · about
chapters (10)
- 0:00Introduction to scaling
- 1:03Challenges in state management
- 1:59State distribution problems
- 4:08Limitations of prior approaches
- 7:07The Six framework proposal
- 8:48Performance of Six framework
- 10:46Design and abstraction
- 13:55Architectural implementation
- 17:31Implementation and evaluation
- 20:24Conclusion and Q&A
related talks
25mStretch: A Fault-Driven DSM Runtime for Distributed Multithreaded Applications
38mState is Hard: An SDK for Building Stateful Applications
46mWriting Stateful Frameworks
19mSponge: Fast Reactive Scaling for Stream Processing with Serverless Frameworks
37mScaling a Distributed Stateful System: A LinkedIn Case Study
46mPerfect Scalability
27mEmpowering Resilient Scalability: The Power of Choreography in Distributed Systems
43mSystems @Scale Summer Q&A Week 2