Improving a Run Time Job Prediction Model for Distributed Computing Based on Two Level Predictions

Nowadays, distributed computing environment faces many difficulties because the number of submitted jobs is increasing dramatically. One of the most used method to serve the jobs is to find the accurate run time of the submitted jobs. This paper proposes a new job prediction method, to predict on jobs’ run time using two level prediction namely linear regression model and fitting model. The proposed model uses six variables including user ID, group ID, executable ID, number of CPUs, memory size and average CPU time, furthermore to solve the problem of the categorical variables (i.e. user ID, group ID and executable ID) a dummy code is used. To adjust and to find the best combination between linear regression model and fitting models, different fitting models are used by combining linear and nonlinear fitting models. By simulation the results show that the proposed model is better than previous models when smoothing spline fitting is used, also the results indicate that proposed model is efficient with low error and high prediction rate compared with previous models.