Performance of adaptive space-sharing policies in dedicated heterogeneous cluster systems

Abstract

In this paper, we propose a new two-level space-sharing policy for dedicated heterogeneous systems. This policy is based on the concept of basic processor unit (BPU). At a higher level, the policy computes the partition size in terms of BPUs. At the second-level, the BPUs of the first level are mapped to physical processors. The proposed policy is an adaptive space-sharing policy as this type of policy has been shown, in homogeneous systems, to provide better performance than the fixed and static policies by taking system load and user requirements into account. We compare the performance of our new policy with a previous policy. The results presented here show that the proposed policy provides substantial performance improvements at system loads of interest (i.e., medium to high system loads).