The scheduling of Out-Tree task graphs is one of the critical factors in implementing the compilers of parallel languages and improving the performance of parallel computing. When applied to Out-Tree task graphs, many previous classical heterogeneity based algorithms always ignored the economization on processors and the minimization of the schedule length, which led to low efficiency in real applications. This paper proposes a heterogeneity based greedy algorithm for scheduling Out-Tree task graphs, which is based on list and task duplication, tries to find the best point between balancing loads and shortening the schedule length and improves the schedule performance without increasing the time complexity of the algorithm. The comparative experimental results demonstrate that the proposed algorithm could achieve shorter schedule length while using less number of processors.
On cluster resource allocation for multiple parallel task graphs
