Improving thread scheduling by thread grouping in heavily loaded many-core processor systems

Basic scheduling techniques currently employed in operating systems usually don't account for hierarchy of processors cores and caches resulting in suboptimal system efficiency, especially on heavy loaded systems. In this paper we explore possible improvements in scheduling for such heavy loaded many-core systems. Basic idea is to group threads of the same process as close as possible, preferably even on the same processor. We expect that doing so may improve efficiency of processor's cache usage, resulting in better overall performance. The idea is tested by adapting the Linux's simplest scheduler (in code) - a round-robin scheduler. Achieved results are presented in this paper.