Summary
Job Scheduling in Computational Grids is gaining importance due to the need for efficient large-scale Grid-enabled applications. Among different optimization techniques designed for the problem, Genetic Algorithm (GA) is a popular class of solution methods. As GAs are high level algorithms, specific algorithms can be designed by choosing the genetic operators as well as the evolutionary strategies such as Steady State GAs and Struggle GAs. In this paper we focus on Struggle GAs and their tuning for scheduling of independent jobs in computational grids. Our results showed that a careful hash implementation for computing the similarity of solutions was able to alleviate the computational burden of Struggle GA and perform better than standard similarity measures. This is particularly interesting for the scheduling problem in Grid systems, which due to changeability over time, has demanding time restrictions on the computation of planning the jobs to resources.
See the full content of this document
Extract
Tuning Struggle Strategy in Genetic Algorithms for Scheduling in Computational Grids
1. Introduction
With the emerging paradigm of Grid Computing and the development of Grid infrastructures, Grid-based applications are becoming a common approach for solving many complex problems. A key issue in this kind of applications is scheduling jobs into Grid resources efficiently, which is known to be computationally hard and much more difficult than its standard version for sequential or LAN computation environments.Job Scheduling in Computational Grids is gaining importance due to the need for efficient large-scale Grid-enabled applications, e.g. in Optimization (Casanova et al. [8], Goux et al. [13] and Wright [30]), Linderoth et al. [19]), Collaborative/eScience Computing (e.g. Newman et al. [22], Paniagua et al. [24]), DataIntensive Computing (e.g. Beynon al. [3]) and many applications arising from concrete types of Grids such as Science Grids, Access Grids, Knowledge Grids, etc. Scheduling is a challenging problem in a Grid environment because of its dynamic nature and the large number of resources to be managed and jobs to be scheduled. Furthermore, resources can have their own local policies (regarding access, cost, etc.) to be taken into account. The problem is multi-objective in its general definition, as there are several optimization criteria to be matched, such as makespan, flowtime, and resource utilization.Several approaches are being addressed in the literature for the problem of obtaining schedulers capable of delivering fast planning ...See the full content of this document
Sponsored links
