In this paper a method for estimating task execution times ispresented, in order to facilitate dynamic scheduling in a heterogeneousmetacomputing environment. Execution time is treated as a randomvariable and is statistically estimated from past observations. Thismethod predicts the execution time as a function of several parametersof the input data, and does not require any direct information about thealgorithms used by the tasks or the architecture of the machines.Techniques based upon the concept of analytic benchmarking/codeprofiling are used to accurately determine the performance differencesbetween machines, allowing observations to be shared between machines.Experimental results using real data are presented
展开▼