Abstract

As multi-core processors with tens or hundreds of cores begin to proliferate, system optimization issues once faced only by the high-performance computing (HPC) community will become important to all programmers. However, unlike with HPC, the focus of the multi-core programmer will be on programming productivity and portability as much as performance. We introduce in this paper a novel scheduling framework for multi-core processors that strikes a balance between control over the system and the level of abstraction. Our framework uses high-level information supplied by the user to guide thread scheduling and also, where necessary, gives the programmer fine control over thread placement.

Citations