Modelling and Developing Co-scheduling Strategies on Multicore Processors

• Published in: 2015 44th International Conference on Parallel Processing pp. 220 - 229
• Main Authors: Zhu, Huanzhou, He, Ligang, Gao, Bo, Li, Kenli, Sun, Jianhua, Chen, Hao, Li, Keqin
• Format: Conference Proceeding Journal Article
• Language: English
• Published: IEEE 01-09-2015
• Subjects: Computation; Degradation; Graphs; Hydroxyapatite; IP networks; Linear programming; Mathematical models; Multicore processing; Optimization; Performance degradation; Processor scheduling; Processors; Program processors; Serials; Silicon
• ISSN: 0190-3918; 2332-5690
• DOI: 10.1109/ICPP.2015.31

On-chip cache is often shared between processes that run concurrently on different cores of the same processor. Resource contention of this type causes performance degradation to the co-running processes. Contention-aware co-scheduling refers to the class of scheduling techniques to reduce the performance degradation. Most existing contention-aware co-schedulers only consider serial jobs. However, there often exist both parallel and serial jobs in computing systems. In this paper, the problem of co-scheduling a mix of serial and parallel jobs is modelled as an Integer Programming (IP) problem. Then the existing IP solver can be used to find the optimal co-scheduling solution that minimizes the performance degradation. However, we find that the IP-based method incurs high time overhead and can only be used to solve small-scale problems. Therefore, a graph-based method is also proposed in this paper to tackle this problem. We construct a co-scheduling graph to represent the co-scheduling problem and model the problem of finding the optimal co-scheduling solution as the problem of finding the shortest valid path in the co-scheduling graph. A heuristic A*-search algorithm (HA*) is then developed to find the near-optimal solutions efficiently. The extensive experiments have been conducted to verify the effectiveness and efficiency of the proposed methods. The experimental results show that compared with the IP-based method, HA* is able to find the near-optimal solutions with much less time.