• PDF / 226,092 Bytes
• 5 Pages / 595.276 x 793.701 pts Page_size
• 84 Downloads / 218 Views

DOWNLOAD

REPORT

---

SOFTWARE–HARDWARE SYSTEMS IDENTIFYING THE OPTIMAL METHOD OF PREDICTING THE UTILIZATION OF CLUSTER RESOURCES AND GRID NODES S. I. Lavreniuka† and O. L. Perevozchikovaa‡

UDC 004

Abstract. A method of determining the optimal model of predicting the queue of tasks of computing clusters and grid nodes is discussed. It is shown that it is possible to use the PREDICTOR statistical prediction software. Keywords: cluster, grid, prediction, number series, queue. INTRODUCTION When a computing cluster operates at full capacity and all the nodes are constantly in use (for example, as in the SCIT-3 cluster at the Institute of Cybernetics of the NAS of Ukraine), it is important to predict the availability of the necessary resources or their amount during certain time periods. Users from many institutes of the NAS of Ukraine employ the SCIT-3 cluster [1, 2]. This cluster is also a powerful node of the Ukrainian Academic Grid (UAG) [3]. In this connection, it serves several task queues: local (for cluster users) and grid (for UAG users) queue. Let the time it takes a cluster to perform a task be specified by the formula Tall = Tq + Tr ,

(1)

where Tq is the time the task is in a local queue and Tr is the task time. To specify the time it takes a grid queue to perform a task, we will use the formula ¢ = Ts + Tq + Tr + Tu , Tall

(2)

where Ts is the time the task and data arrive at the grid node, Tq is the time the task is in the local queue of the grid node, Tr is the task time, Tu is the time the specified resource obtains the result. Generally, Tq is the time of waiting for the resources necessary to perform the task. To estimate and minimize Tall and ¢ Tall , it is first of all necessary to estimate Tq . Using the estimation results, it is possible to choose the grid nodes with optimal ¢ . Then the problem reduces to finding and choosing optimal methods to Tq and Tr to start the tasks and minimize Tall and Tall

predict the resources of the local and grid queues and to determine the optimal number of periods of acceptable prediction. PROBLEM SOLUTION To solve the problem posed, the PREDICTOR software package for automatic numerical predictions is used. This interactive prediction package is an add-on for MS Excel and has four processing complexity modes with different intellectual means for users [4]. a

V. M. Glushkov Institute of Cybernetics, National Academy of Sciences of Ukraine, Kyiv, Ukraine, [email protected]; ‡[email protected]. Translated from Kibernetika i Sistemnyi Analiz, No. 2, pp. 160–164, March–April 2011. Original article submitted May 17, 2010. †

1060-0396/11/4702-0311

©

2011 Springer Science+Business Media, Inc.

311

1. For beginners, the Master mode allows a prediction to be quickly obtained by step-by-step operations of establishing seasonality, choosing a method (model), visual quality assurance of the model, and recording the result as a number series. The optimal values of all the necessary parameters are chosen automatically by the PREDICTOR. 2. For qualified users, a dialogue environment of iterativ