Using Machine Learning Ensemble Methods to Predict Execution Time of e-Science Workflows in Heterogeneous Distributed Systems

摘要: Effective planning and optimized execution of the e-Science workflows in distributed systems, such as Grid, need predictions times workflows. However, predicting heterogeneous systems is a challenging job due to complex structure workflows, variations input problem-sizes, dynamic nature shared resources. To this end, we propose two novel workflow time-prediction methods based on machine learning ensemble models. In paper, showcase our approach for different real Grid environments. Our can effectively predict time scientific applications various problem sizes, sites, runtime We characterized performance using attributes that define well environment. Contrary common ensembles, employed three strong learners, which balance weaknesses each other by their strengths model times. The proposed have been thoroughly evaluated real-world e-science applications. experimental results demonstrated multi-model models significantly decrease prediction error (by 50%, average) compared with radial basis function neural network, local learning, templates. also be applied similar effectiveness without any major modification environments, Cloud.