Optimizing Scheduling Stability for Runtime Data Alignment

Runtime data alignment has been paid attention recently since it can allocate data segment to processors dynamically according to applications’ requirement. One of the key optimizations of this problem is to schedule simultaneous communications to avoid contention and to minimize the overall communication costs. The NP-completeness of the problem has instigated researchers to propose different heuristic algorithms. In this paper, we present an algorithm independent technique for optimizing scheduling stability of different scheduling heuristics. The proposed technique introduces a new scheduling policy, Local Message Reduction (LMR), to obtain better communication schedule adaptive to different environments. To evaluate the performance of the proposed technique, we have implemented LMR along with two existing algorithms, the two-phase degree reduction and the list scheduling algorithms. The experimental results show that the proposed technique is effective in terms of scheduling stability, communication efficiency and easy to implement.