Replication in Data Grids: Metrics and Strategies

TL;DR

This paper analyzes how replica distribution affects data grid performance, introduces new evaluation metrics, and proposes a data mining-based replication strategy with an algorithm for mining correlated patterns.

Contribution

It provides a comprehensive overview of replication strategies, introduces new metrics for distribution quality, and develops a novel data mining-based replication algorithm.

Findings

Distribution quality significantly impacts performance.

New metrics effectively evaluate distribution quality.

Data mining techniques can enhance replication strategies.

Abstract

We focus in this report on two main axes. The first is dedicated to the study of the effect of replicas distribution on data grid performances. In this respect, our main contributions are as follows: 1) An overview of replication strategies mainly from the viewpoints of the considered parameters in their associated steps as well as the used metrics in the literature for their evaluation. 2) A study of the impact of placement strategies on data grid performance which motivated the analysis of the effect of the replicas distribution quality on the performance results of replication strategies. 3) The proposal of new evaluation metrics dedicated to the evaluation of the distribution quality. 4) The setting of an objective evaluation of replication strategies which is based on a beforehand assessment of the distribution quality. The second axis is mainly dedicated to exploiting results of data mining techniques to enhance performances of replication strategies. With respect to this axis, we mainly concentrate on the following contributions listed below: 1) The study of the strengths and the drawbacks of the main replication strategies based on data mining techniques and how these latter are applied in this context. 2) The proposal of a new guideline to data mining application in the context of data grid replication strategies. 3) The proposal of a new algorithm for mining maximal frequent correlated patterns. The input of this algorithm is obtained through a preliminary step focusing on how to adapt the required grid concepts to the data mining algorithm. 4) The design and the implementation of a new replication strategy based on a data mining technique, and more precisely correlated patterns.