Genetic-based algorithms applied to a workflow scheduling algorithm with security and deadline constraints in clouds

Abstract

There have been a number of metaheuristic scheduling techniques for cloud described in the literature, as well as their applications. The efficiency of metaheuristic techniques has been established in a wide range of workflow scheduling algorithms for cloud environments. However, it is still unknown whether the metaheuristic that is chosen, is suitable for solving the problem of optimization. This paper examines the effect of both Particle Swarm Optimization (PSO) and Genetic-based algorithms (GA) on attempts to optimize workflow scheduling. A security and cost-aware workflow scheduling algorithm was selected to evaluate the performance of the metaheuristics. Three algorithms were evaluated in three real-world workflows with a risk rate constraint that ranged between 0 and 1 with a 0.1 step. The findings indicate that GA-based algorithms significantly outperformed the PSO both in term of cost-effectiveness and response time.

Introduction

scientific workflow is a recent paradigm for distributed programming that is deployed in computational experiments in scientific A areas such as physics, astronomy, and biology. A workflow can be defined as a procedure involving a series of steps designed to simplify the complexity of executing and managing applications [1]. It is commonly represented as a Directed Acyclic Graph (DAG), where each node carries out a task, and each edge denotes a precedence or flow constraints between the tasks. Previously, workflows were deployed in computational grids. However, owing to the increase in the complexity of the scientific applications for handling big data, more powerful and scalable infrastructures are needed to run complex workflows within a reasonable amount of time [2].

Cloud computing is an infrastructure which can be rapidly and elastically provisioned on demand [3]. It can offer different virtual machine configurations that are capable of executing workflows. Clouds can be classified as private, community-based, public or hybrid clouds. Private cloud is owned and used by a single organization, and it does not charge the services it offers. Community-based clouds are designed for a specific community that has shared concerns. Public cloud shares services with multiple-tenants and these are charged on the basis of their use. Finally, hybrid cloud consists of a mix of public, communal and private clouds, where public/communal cloud resources are integrated with private cloud to create a single environment. There is a wide range of cloud services (e.g., virtual machine (VM) types, storage policy, and cloud data center locations) [1], [4]. For this reason, scheduling is a key factor in a workflow executed in a cloud environment and has been the subject of investigation in recent years [5]. Scheduling algorithms are designed to map out the tasks of the workflow for VMs based on scheduling criteria and subject to the constraints of the users. However, the heterogeneity of these tasks and wide range of cloud services, lead to an NP-hard optimization problem. Carrying out optimal scheduling within a reasonable time is a challenge because some variables have to be taken into account such as the fact that there are (a) several types of VMs with different capacities and prices, (b) tasks with heterogeneous loads, and (c) other attributes that impede the process of optimization.

In view of this, metaheuristic techniques have been employed in workflow scheduling in clouds to find a near-optimal scheduling scheme. However, research on the use of metaheuristics for workflow scheduling in cloud has been largely restricted to a single technique. In a recent study, it was found that Particle Swarm Optimization (PSO) is the most widely employed technique, and more than 50% of the scheduling algorithms were based on metaheuristics [6]. These findings suggest that there is a need to understand whether scheduling algorithms that are optimized by metaheuristics can achieve a better performance than any alternative and whether PSO is, in fact, a good choice.

This paper examines the effects of different metaheuristics on a workflow scheduling algorithm for cloud. Our major objective was to investigate the differences between the optimization obtained by Particle Swarm Optimization, the Genetic Algorithm (GA), and the Multi-Population Genetic Algorithm (MPGA). The approach adopted in this study involves a mixed methodology based on a security-aware and cost-aware workflow scheduling algorithm [7] that applied PSO to optimize the combinatorial scheduling scheme, and Genetic-based algorithms. Real-world workflows were deployed to evaluate the suitability of each metaheuristic. In addition, an analysis was conducted of the time spent until the stagnation of each optimization technique. Statistical Tests were conducted that assumed p ≤ 0.05 to check if there is a significant difference between PSO and GA-based algorithms.

The findings should be a useful addition to solutions provided by workflow scheduling algorithms based on multidimensional optimization, and show an efficient method for achieving an optimization beyond that of PSO. However, the reader should bear in mind that this study does not claim that workflow scheduling problems optimized by GA-based algorithms always outperform PSO. As mentioned earlier, the optimization efficiency of GA-based and PSO algorithms has only been examined in a single workflow scheduling algorithm.

The remainder of the paper is structured as follows: Section 2 summarizes the related work. Section 3 formulates the problem and system models in detail. The methodology employed for the experiments is outlined in Section 4. Section 5 explains the experimental planning and examines the results of the evaluation. Finally, in Section 6 there are some concluding remarks and suggestions for future research in the area.