Grid broker selection strategies using aggregated resource information

摘要

The increasing demand for high performance computing resources has led to new forms of collaboration of distributed systems, such as grid computing systems. Moreover, the need for interoperability among different grid systems through the use of common protocols and standards has become increased in the last few years. In this paper we describe and evaluate scheduling techniques for multiple grid scenarios. In particular, they consist of the proposed "bestBrokerRank" broker selection policy and two different variants. The first one uses the resource information in aggregated forms as input, and the second one also uses the broker average bounded slowdown as a dynamic performance metric. From our evaluations performed with simulation tools, we state that, although the aggregation algorithms lose resource information accuracy, the broker selection policies using aggregated resource data do not penalize their performance significantly. Moreover, we show that the best performance results are obtained with the coordinated policy using dynamic performance information, in addition to aggregated resource information. Therefore, we conclude that delegating part of the scheduling responsibilities to the underlying scheduling layers promotes separation of concerns and is a good way to balance the performance among the different grid systems.