A Reliability Benchmark for Actor-Based Server Languages
summary
Designed and released Supervised Communicating Processes (SCP), implemented in Erlang/OTP and Scala/Akka with a deterministic fault injector that triggers progressive permanent failures (example: 5% of pairs killed every 5s) and recovery modes (uniform, random, burst) across supervisor:worker ratios from 1:1 to 1:128 and failure rates up to 20%. Results show both systems provide strong process isolation; Erlang supervisors add measurable overhead so supervisor count critically affects throughput (more supervisors help handle bursts), while Scala/Akka shows higher latency variance but outperforms Erlang on uniform recoveries when tuned.
AI-written summary · about
chapters (7)
- 0:00Start
- 8:11SCP Benchmarking
- 9:29Erlang Implementation
- 13:07Scala/Akka Implementation
- 18:58Failures with Recovery
- 23:02Different SCP Failure Patterns: Erlang
- 29:26Summary
related talks
37mBeyond Fault Tolerance
18mReliable Actors with Retry Orchestration
27mFour Principles of Reactive Architectures
01h 01mThe Erlang Ecosystem
22mPARTISAN: Scaling the Distributed Actor Runtime
40mSo I used Erlang...is my system as scalable as they say it'd be?
52mDistributed Computing Dealing with Time and Failure in the Wild
48mReliable messaging in the world of actors