Advanced reservation-based scheduling of task graphs on clusters
HiPC'06 Proceedings of the 13th international conference on High Performance Computing
| Hi-index | 0.00 |
By means of a comprehensive test bench of 36000 test cases we evaluated the efficiency of functional parallel programs. For all the test cases schedules have been computed by various well known heuristics. We assumed a homogeneous target system (e.g. a compute cluster of equally powerful interconnected nodes) that can be part of a grid computing environment which supports the execution of parallel programs. Unfortunately, the efficiencies of the investigated schedules were pretty low. For this reason, we propose a new Hyper-scheduling approach that reduces the amount of idle times by interweaving subsequent schedules from the parallel job queue. First results confirm that Hyperscheduling significantly improves efficiency.