Reader/writer Range Locking

*
Refereed Presentation
Scheduled: Thursday, September 14, 2017 from 4:00 – 4:45pm in Platinum D

One Line Summary

Discuss features and design of a fair, sleepable range reader/writer lock for the linux kernel.

Abstract

Range locking improves parallelism by fine graining locks which can cause bottlenecks and loss of performance, such as the infamous mmap_sem. As such there as been active efforts to upstream a scalable reader/writer range lock (which can also aid DAX and lustre, to mention some users). The session will update and discuss key implementation details such as fairness, performance and comparisons with a more traditional rw-semaphore.

Audio

Tags

Synchronization, performance, locking, address space

Presentation Materials

slides

Speaker