A survey on methodologies for runtime prediction on grid environments

The accurate prediction of runtimes of future job tasks on the nodes of a grid supplies vital information for the users to make CPU resource usage decisions. There are number of different approaches to predict runtimes of the future job tasks. These approaches range from the statistical to non statistical and some of them require expensive search algorithms or availability of the source code of job tasks. In this paper we discuss about the existing such methods and categorise them according to a certain taxonomy. Then we compare the advantages and disadvantages of them with that of the Task Profiling Model for Host Load Profile Prediction which is developed by the authors.