Fast and Scalable In-network Lock Management Using Lock Fission
Talkat 18th USENIX Symposium on Operating Systems Design and Implementation (OSDI '24) · 2024-07-10..07-12·
summary
FISSLOCK implements lock fission by running the grant decision in a programmable switch pipeline and keeping holders/waiters in server-side agent pools, with on‑demand agent migration for load balance. The prototype supports over one million locks on a single rack switch, cuts median lock‑grant time by up to 79.1% versus prior switch designs, and improves end‑to‑end transaction throughput (TATP 1.76×, TPC‑C 2.28×); evaluation shows low, stable tail latency and resilience to dynamic hotspot shifts.
AI-written summary · about
chapters (8)
- 0:00Distributed lock management intro
- 2:30Limitations of lock managers
- 4:58Lock fission principle
- 6:21FLock operation overview
- 8:30Distributed lock fission challenges
- 10:33Switch memory optimization
- 10:53Performance evaluation
- 13:01Conclusion and future
related talks
18mFARLock: Asymmetric RDMA Locking Made Fair
23mBalancing on the Edge: Transport Affinity without Network State
17mIn Link We Trust: BFT at the Speed of CFT using Switches
15mShiftLock: Mitigate One-sided RDMA Lock Contention via Handover
19mDeveloping, Deploying, Operating In-House Network Switches at a Massive Scale
25mShoal: A Network Architecture for Disaggregated Racks
16mUnlocking the Power of Inline Floating-Point Operations on Programmable Switches
23mProtecting your resources with sanlock