Abstract
Most attention in the paper is paid to GRID quality parameter, which specifies averaged time of execution of one job waiting in a cluster queue. Detailed analysis of time series of this parameter, which is received from monitoring systems of academic BalticGrid network, is presented. Seasonalities of variation of parameter value are shown graphically. Problematic forecasting cases of this parameter are reviewed. Forecasting methodology, which involves a model of data imputation, is proposed. The experiment is carried out, using forecasting methods that are prevailing in practice (AR, nAR, ARMA, SARIMA, CF technique). Ill. 10, bibl. 10, tabl. 1 (in English; abstracts in English and Lithuanian). http://dx.doi.org/10.5755/j01.eee.114.8.706
Highlights
It is increasingly confronted with the problem of choosing suitable clusters for execution of jobs in GRID networks for commercial and scientific purposes
WMS (Workload Management System) services, which operate according to particular algorithms of a cluster search, are responsible for choice of suitable clusters for jobs sent to GRID network
Since QoGS method is intended for GRID networks with an unlimited amount of clusters, when improving a chosen method, it is necessary to take into consideration that the necessary data, according to the described methods [6] and [8], is not accessible in many cases or its samples are too small
Summary
It is increasingly confronted with the problem of choosing suitable clusters for execution of jobs in GRID networks for commercial and scientific purposes. In continuation of research in this area, it has been chosen to improve the concept of QoGS algorithm [1], which operation is based on the search of the best cluster, using qualitative parameters While improving this algorithm, it is necessary to evaluate the fact that a part of values of parameters, describing operation of GRID network, varies high in time. GRID cluster for a job execution will be chosen more precisely (more suitably), using forecasted values of parameters that describe clusters in QoGS algorithm, since the average of time series of parameters has been used so far. Having done self-training according to this data, it is possible to forecast that execution time of the job, which arrived, will be the same as that of the most similar executed job This method serves perfectly for computing, having GRID network with a very limited amount of clusters. Since QoGS method is intended for GRID networks with an unlimited amount of clusters, when improving a chosen method, it is necessary to take into consideration that the necessary data, according to the described methods [6] and [8], is not accessible in many cases or its samples are too small
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
