High-performance execution of service compositions: a multicore-aware engine design

摘要

Although modern computer hardware offers an increasing number of processing elements organized inrnnonuniform memory access (NUMA) architectures, prevailing middleware engines for executing businessrnprocesses, workflows, and Web service compositions have not been optimized for properly exploiting thernabundant processing resources of such machines. Amongst others, factors limiting performance are inefficientrnthread scheduling by the operating system, which can result in suboptimal use of system memory andrnCPU caches, and sequential code sections that cannot take advantage of multiple available cores.rnIn this article, we study the performance of the JOpera process execution engine on recent multicorernmachines. We first evaluate its performance without any dedicated optimization for multicore hardware,rnshowing that additional cores do not significantly improve performance, although the engine has a multithreadedrndesign. Therefore, we apply optimizations on the basis of replication together with an improved,rnhardware-aware usage of the underlying resources such as NUMA nodes and CPU caches. Thanks to ourrnoptimizations, we achieve speedups from a factor of 2 up to a factor of 20 (depending on the target machine)rnwhen compared with a baseline execution ‘as is’.