Bandwidth variability prediction with rolling interval least squares (RILS)
Proceedings of the 50th Annual Southeast Regional Conference
Hi-index | 0.00 |
Growing complexity in Grid environment makes bandwidth monitoring and prediction both increasingly difficult and increasingly important. The challenge is to design a low-cost and online prediction system to provide Grid users and developers with detailed bandwidth information. Bandwidth prediction is designed based on Grid Service Infrastructure, it is composed of a series of Grid services including monitoring service, prediction service, optimizing service and publication service. Support Vector Regression is used as prediction technology and optimizing strategy is proposed to enhance prediction accuracy. Bandwidth prediction is tested using benchmark data set and experimental results indicate that the prediction service is efficient in optimizing model hyperparameters and prediction error is low on both familiar and unfamiliar bandwidth samples.