SANTHOSH RAMAIAH

Our main aim of this project is to propose and compare algorithms for obtaining low energy mapping onto many-cores. To cope with complex dynamic behavior of applications, it is relevant to perform task mapping during runtime, hence, heuristic approaches are taken instead of exhaustive search of the solution space. Some heuristic-based algorithms proposed previously consider placing tasks in order of largest communication volume. But the Placed Communication First (PCF) mapping algorithm introduced by Trahan and Elbidweihy and refined here performs task mapping based on largest volume of communication to already placed tasks rather than overall communication volume. Experimental results show that PCF provides average reduction of 41% communication cost compared to other algorithms on real application task graphs. To explore the problem space, we also consider randomly generated task graphs and make use of a scheduler with region selection in addition to mapping, showing similar results...