Transactional Mutex Locks

Mutual exclusion locks limit concurrency but offer low latency. Software transactional memory (STM) typically has higher latency, but scales well. In this paper we propose transactional mutex locks (TML), which attempt to achieve the best of both worlds for read-dominated workloads. TML has much lower latency than STM, enabling it to perform competitively with mutexes. It also scales as well as STM when critical sections rarely perform writes. In this paper, we describe the design of TML and evaluate its performance using microbenchmarks on the x86, SPARC, and POWER architectures. Our experiments show that while TML is not general enough to completely subsume both STM and locks, it offers compelling performance for the targeted workloads, without performing substantially worse than locks when writes are frequent.

• Type: Conference
• Year: 2010
• Where: EUROPAR
• Authors: Luke Dalessandro, David Dice, Michael L. Scott, Nir Shavit, Michael F. Spear
• Keywords: Distributed And Parallel Computing, EUROPAR 2010, Locks Limit Concurrency, Offer Low Latency, Transactional Mutex Locks