SymS: a symmetrical scheduler to improve multi-threaded program performance on NUMA systems

摘要

The nonuniform memory access (NUMA) architecture has been used extensively in data centers. Most ofrnthe previous works used single-threaded multiprogrammed workloads to study the performance of NUMArnsystems, which mainly focus on two classes of problems: resource contention and data locality. However,rnwhen running multi-threaded programs on NUMA systems, the critical thread of these programs significantlyrninfluences the system performance and brings new challenges that are different from those in arnsingle-threaded situation. In particular, an additional scheduling scheme is desired to avoid the performancerndegradation caused by the critical thread of multi-threaded programs running on NUMA systems. This workrnpresents a scheduler, Symmetrical Scheduler, which successfully solves the lagging problem by balancingrnthe number of the costly remote shared data accesses for threads on NUMA systems. To the best of ourrnknowledge, little work has been conducted to examine the performance impacted by the critical thread ofrnmulti-threaded programs on NUMA systems. By running the PARSEC benchmark on such systems, ourrnmethodology can improve the program performance by a factor of 6% on average and achieve maximallyrn25.3% improvement compared with Linux kernel scheduling mechanism.