Edge Scheduling Algorithms in Parallel and Distributed Systems

Many research efforts have been done in the domain of static scheduling algorithms based on DAG. However, most of these literatures assume that all processors are fully connected and receive communication data concurrently, while ignoring the contentions and delays on network links in real applications, which leads to low efficiency. This paper focuses on the issue of edge scheduling for dependent task set in parallel and distributed environment. Combined with conventionally efficient heuristics, two contention-aware scheduling algorithms are proposed in the paper: OIHSA (Optimal Insertion Hybrid Scheduling Algorithm) and BBSA (Bandwidth Based Scheduling Algorithm). Both the proposed algorithms start from the inherent characteristic of the edge scheduling problem, and select route paths with relatively low network workload to transfer communication data by modified routing algorithm. OISHA optimizes the start time of communication data transferred on links in form of theorems. BBSA exploits bandwidth of network links fully to transfer communication data as soon as possible. Therefore, the makespan yielded by our algorithms can be reduced efficiently. Moreover, the proposed algorithms adapt to not only homogeneous systems but also heterogeneous systems. The experiment results indicate that the proposed algorithms obviously outperform other algorithms so far in terms of makespan.