A Multi-Objective Optimization Model for Data-Intensive Workflow Scheduling in Data Grids

摘要

The concept of workflow is used for modeling many of the data-intensive scientific applications executed on data grids. A Workflow is a series of interdependent tasks during which data is processed by different tasks. Scheduling the workflows in the grids is the process of assigning tasks to appropriate resources with the aim of achieving goals such as reducing workflow completion time while considering the data dependencies between the tasks. Data access time, processing time, and waiting time together constitute task completion time in the grids. Workflow scheduling aims to optimize these parameters in such a way that the workflow completion time decreases, and the system efficiency improves. In this paper, a scheduling model based on multi-objective optimization is proposed for scheduling data-intensive workflows in data grids. The scheduling model aims to optimize data communication cost, waiting time, and tasks processing time while considering data dependencies between the tasks. The model defines the data communication cost in terms of data transfer time in various communications between nodes (intra-and inter-cluster communications). This study uses four different Multi-Objective Evolutionary Algorithms (MOEAs) as well as Random Search (RS) algorithm to implement the proposed scheduling model. Convenient coding mechanisms for representing chromosomes, compatible crossover and mutation operators were also designed. Simulation results of the proposed scheduling model using different optimization algorithms are presented. The results are then assessed and compared based on different quality indicators.