Task profiling model for load profile prediction

The accurate prediction of load profiles of future job tasks on the nodes of a cluster or grid supplies vital information for the users to make CPU/Disk resource usage decisions. At present, the Unix five-second host load is collected and used to predict the host loads, but forecasting can be improved if CPU and Disk load data are collected separately for each user on each host. The Free Load Profile or footprint of a job task on a load free node is a necessary input to the proposed Performance Prediction Model. To this end, the Task Profiling Model for Load Profile Prediction is proposed, which forecasts the load profiles of job tasks of individual machines based on current and historical data. The data is collected by agents running on the nodes of the cluster/grid. The data so obtained aids in choosing the most suitable set of computers for the deployment of the tasks in time optimal manner. Also, accurately predicted load profiles are useful inputs to the cost prediction models. The Task Profiling Model has been implemented in a software framework and evaluated for its prediction accuracy.

Research highlights
► The Task Profiling Model predicts the computer load profiles of future job tasks.
► The foundation of the Model is the DoL that is applied at the Os kernel level.
► Its runtime predictions are found to be better than the available methodologies.