| In this thesis, a class of scheduling problem is examined in which independent tasks with compatibility constraints are assigned to a bounded number of processors. A unique feature of the scheduling problem is that the scheduling algorithm itself contributes to the overall execution time of the application. In this situation, it is the combination of the scheduling algorithm and the resulting schedule that must be minimized.;Experimental results show that the performance of many scheduling heuristics is dependent upon the combination of input parameters to the application, which can be difficult to predict. A new adaptable approach to scheduling tasks in an unpredictable environment is proposed. The proposed approach provides a mechanism for multiple heuristics to compete with each other at runtime. The advantages of this approach include the ability to allow an application to fine-tune its runtime performance dynamically, as well as the ability to protect an application from the unpredictability or unreliability of any particular scheduling heuristic when faced with undesirable input parameters. |
