An Adaptive and Historical Approach to Optimize Data Access in Grid Computing Environments

Abstract

The data Grid, a class of Grid Computing, aims at providing services and infrastructure to data-intensive distributed applications which need to access, transfer and modify large data storages. A common issue on Data Grids is the data access optimization, which has been addressed through different approaches such as bio-inspired and replication strategies. However, few of those approaches consider application features to optimize data access operations (read-and-write). Those features define the application behavior, which supports the optimization of operations, consequently, improving the overall system performance. Motivated by the need of efficient data access in large scale distributed environments and by the affordable improvements of application characteristics, this paper proposes a new heuristic to optimize data access operations based on historical behavior of applications. Throughout experiments we concluded that applications are better optimized by anticipating different numbers of future events, which vary over the execution. Then, in order to address such issue, we proposed an adaptive sliding window which automatically and dynamically defines how many future operations must be considered to improve the overall application performance. Simulations were conducted using the OptorSim simulator, which is commonly considered in this research field. Our experimental evaluation confirms that the proposed heuristic reduces application execution times up to 50% when compared to other approaches.