Beehive: A Scalable Disaggregated Memory Runtime Exploiting Asynchrony of Multithreaded..
summary
Beehive builds a Rust compile-time transform plus a runtime that turns synchronous multithreaded code into asynchronous pararoutine units and a lightweight cooperative scheduler, using ownership rules to avoid pointer-alias false dependencies. The runtime launches and suspends pararoutines with ~10 ns launch/switch overhead to overlap microsecond remote-memory I/O, evaluated on eight workloads and open-sourced. It outperforms Fastswap, Hermit, and AIFM by 4.26×, 3.05×, and 1.58× on average and improves throughput and tail latency by reducing cache misses and scheduling cost.
AI-written summary · about
chapters (8)
- 0:00Beehive runtime intro
- 1:21Disaggregated memory challenges
- 3:13Thread switching limits
- 5:25Async execution insights
- 6:55Sync code challenges
- 8:46Beehive design
- 10:34Performance results
- 12:23Takeaways
related talks
14mMultitenancy for Fast and Programmable Networks in the Cloud
31mOptimizing Cost and Performance with arm64
17mFollowing the Data, Not the Function: Rethinking Function Orchestration in Serverless
16mResource Efficient Stream Processing Platform with Latency-Aware Scheduling
19mEnabling Big Memory workloads to run
14mCanvas: Isolated and Adaptive Swapping for Multi-Applications on Remote Memory
22mRight Abstraction
47mNHD - A Topology-Aware Scheduler for K8s for Low-Latency & HPC Applications