摘要：Current performance prediction analyticalmodels try to characterize the performance behaviorof actual machines through a small set of parameters.Due to different factors, the predicted times suffer sub-stantial deviations. A natural approach is to associate adifferent proportionality constant with each basic blockof computation. In particular, the paper deals with askeleton designed for parallel divide and conquer algo-rithms that provide hypercubical communications amongprocesses. Our proposal is to introduce different kindsof components to the analytical model by associatinga performance constant for each conceptual block of askeleton. The trace files obtained from the execution ofthe resulting code using the programming skeleton areused by lineal regression techniques giving us, amongother information, the values of the parameters of thoseblocks. The accuracy of the proposed model is analyzedby means of two instances of skeleton
关键词：Paralellism; Timing Model; Skeleton;Multivariate Analysis; Divide and Conquer
