Host load prediction for grid computing using free load profiles

Authors:
Sena Seneviratne;David Levy
Affiliations:
School of Electrical and Information Engineering, The University of Sydney, Sydney, NSW, Australia;School of Electrical and Information Engineering, The University of Sydney, Sydney, NSW, Australia
Venue:
ICA3PP'05 Proceedings of the 6th international conference on Algorithms and Architectures for Parallel Processing
Year:
2005

Citing 6
Cited 1

The network weather service: a distributed resource performance forecasting service for metacomputing

Future Generation Computer Systems - Special issue on metacomputing
Dynamically forecasting network performance using the Network Weather Service

Cluster Computing
Host load prediction using linear models

Cluster Computing
Online Prediction of the Running Time of Tasks

Cluster Computing
Realistic CPU Workloads through Host Load Trace Playback

LCR '00 Selected Papers from the 5th International Workshop on Languages, Compilers, and Run-Time Systems for Scalable Computers
Design and Evaluation of a Resource Selection Framework for Grid Applications

HPDC '02 Proceedings of the 11th IEEE International Symposium on High Performance Distributed Computing

CPU load prediction using neuro-fuzzy and Bayesian inferences

Neurocomputing

Quantified Score

Hi-index	0.00

Visualization

Abstract

In Order to increase the overall performance, we have studied methods for improving load prediction, which would help improve load balancing in the Grid. Current software designed to handle distributed applications does focus on the problem of forecasting the computer’s future load. The UNIX five-second-host load has been collected and used to predict the host load, but the solution of forecasting can be further improved if CPU historical load data had been collected separately for each login user. Another important aspect of historical data collection is that before submission to the grid, the user separates his HPC program into sizable parallel programs and test runs them supposedly on load free computers. This means the user can obtain the load profile of the parallel program on a load free computer together with other important information. Once the free load profile is known, load behaviour of a job under certain variable background load conditions can be predicted. Thus the forecast can be performed for each user before adding the weighted values towards the final solution of prediction. In this paper we have proved that load prediction using free load profiles provided better results. In fact once the user based load data are collected, the forecasting is somewhat like that of the Stock market.