Effective Metaheuristic Algorithms for Bag-of-Tasks Scheduling Problems Under Budget Constraints on Hybrid Clouds

Abstract

Cloud computing is an ideal platform for executing bag-of-task (BoT) applications due to its capability of delivering high-quality and pay-per-use computing services. This paper presents a family of genetic algorithm (GA)-based metaheuristics for scheduling the tasks of data-intensive BoT applications on hybrid clouds. The scheduling objective is to minimize the flowtime of BoT applications under a specified budget constraint. We take into account the impact of communication time and communication cost to formulate the optimization model for the data-intensive BoT scheduling problem. By using a task sequence to represent the scheduling solution, the proposed algorithms start with using a low-complexity strategy to generate an initial solution. The generated initial solution is identified as the best chromosome in the initial population of GA framework. We improve the standard crossover operator in GA’s evolutionary procedure by incorporating a probabilistic model. In addition, we design an efficient task dispatching method to evaluate the scheduling quality of each chromosome. Built upon the improved crossover scheme and task dispatching method, the proposed metaheuristic algorithms employ three crossover operators to solve the BoT scheduling problem considered in this work. Extensive experiments are performed to verify the performance of the proposed algorithms in scheduling data-intensive BoT applications.