Study of the behaviour of heuristics relying on the Historical Trace Manager in a (multi)client-agent-server system
Résumé
We compare some dynamic scheduling heuristics that have shown good performances on simulation study against MCT on experiments on real solving platforms. The heuristics rely on a prediction module, the Historical Trace Manager. They have been implemented in NetSolve, a Problem Solver Environment built on the client-agent-server model. Numerous different scenarios have been examined and many metrics have been considered. We show that the predicting module allows a better precision in task duration estimation and that our heuristics optimize several metrics at the same time while outperforming MCT.
Loading...