Scalable reader-writer locks

Abstract

We present three new reader-writer lock algorithms that scale under high read-only contention. Many previous reader-writer locks suffer significant degradation when many readers attempt to acquire the lock concurrently, even though they are all allowed to hold the lock at the same time. In contrast, our locks scale almost perfectly when there is only read contention on a 4-chip system with a total of 256 hardware threads.Two of the algorithms extend the MCS queue mutex to provide reader-writer synchronization with low overhead, and can be used when busy-waiting synchronization is appropriate. The third algorithm is an improvement on a production-quality reader-writer lock used in the Solaris TM kernel, which provides robust priority and flexible fairness guarantees.A key tool we developed to implement our reader-writer locks is the closable scalable nonzero indicator (C-SNZI), a variation on the SNZI object. C-SNZI objects allow us to significantly reduce the contention among reads when many readers try to acquire the lock concurrently, but keeps the acquisition overhead small in the absence of read contention. We present an algorithm for C-SNZI that achieves this goal, and show how it can be used by each of our lock algorithms to provide scalable reader-writer locks with different fairness guarantees.

NUMA-aware reader-writer locks

“…This, however, requires the ability to allow newly arriving readers to bypass writers that arrived earlier but that are still waiting to acquire the lock. This observation was made in the earliest work on RW locks by Courtois et al [2], and then followed by other works that make the same trade off between fairness and scalability [11,15,16]. Our C-RW-RP algorithm also entails the same trade off.…”

“…In our microbenchmark experiments, our locks outperform the prior best RW lock (the SNZI-based ROLL lock [15]) by up to a factor of 10.…”

“…The Reader-Preference Lock (C-RW-RP). lock, an approach that has been used in the past [15,16]. Figure 3 depicts the high-level pseudo-code of C-RW-NP.…”

“…RW locks have been studied extensively for several decades [1,2,11,[13][14][15][16], with proposals ranging from simple counter-or semaphore-based solutions [2], to solutions leveraging centralized wait-queues [14,16], to solutions that use more sophisticated data structures such as Scalable Non-Zero Indicators (SNZI) [15]. Of these, all but the SNZI-based solutions rely on centralized structures to coordinate threads, and thus encounter scalability impediments [15].…”

“…Of these, all but the SNZI-based solutions rely on centralized structures to coordinate threads, and thus encounter scalability impediments [15]. The SNZI-based algorithms keep track of readers -threads acquiring the RW lock in read mode -with each reader arriving at a leaf in the "SNZI tree".…”