Predictive mapping framework in heterogeneous computing environment
Goran Martinović, Leo Budin, Željko Hocenski · IEEE Transactions on Image Processing · 2004
Heterogeneity of computer environment allows execution of large diverse applications, but it also simultaneously aggravates mapping. We present a combined multicriterial mapping based on the MEVSP model. Model parameters originate from the level of application, platform, mapping and user. The most suitable prediction methods are long-term performance prediction, short-term prediction of threshold violations and short-term predictions of categorical data. Heterogeneity level can be expressed by single or stochastic values and by consistency level. The application of the MEVSP model and prediction procedures can improve the application execution, as shown in a simplified environment.