Load distribution on microkernels

Abstract

The authors investigate the appropriateness of a /spl mu/kernel for load distribution. They demonstrate /spl mu/kernel benefits by presenting a load distribution implementation on top of the Mach /spl mu/kernel. The load distribution scheme is based on three main parts: task migration, load information management, and distributed scheduling. It is shown that it is relatively easy to implement transparent task migration on top of a message passing /spl mu/kernel, such as Mach. Changing the distributed scheduler so that it considers IPC and VM load in addition to processor load is straightforward, and it yields up to a 30% improvement in average execution time. The authors investigate distributed scheduling strategies in the light of the extended load information management. Finally, they compare load distribution on various OS architectures.<>