Optimized task scheduling and resource allocation on cloud computing environment using improved differential evolution algorithm

https://doi.org/10.1016/j.cor.2013.06.012Get rights and content

Abstract

Purpose

The objective of this study is to optimize task scheduling and resource allocation using an improved differential evolution algorithm (IDEA) based on the proposed cost and time models on cloud computing environment.

Methods

The proposed IDEA combines the Taguchi method and a differential evolution algorithm (DEA). The DEA has a powerful global exploration capability on macro-space and uses fewer control parameters. The systematic reasoning ability of the Taguchi method is used to exploit the better individuals on micro-space to be potential offspring. Therefore, the proposed IDEA is well enhanced and balanced on exploration and exploitation. The proposed cost model includes the processing and receiving cost. In addition, the time model incorporates receiving, processing, and waiting time. The multi-objective optimization approach, which is the non-dominated sorting technique, not with normalized single-objective method, is applied to find the Pareto front of total cost and makespan.

Results

In the five-task five-resource problem, the mean coverage ratios C(IDEA, DEA) of 0.368 and C(IDEA, NSGA-II) of 0.3 are superior to the ratios C(DEA, IDEA) of 0.249 and C(NSGA-II, IDEA) of 0.288, respectively. In the ten-task ten-resource problem, the mean coverage ratios C(IDEA, DEA) of 0.506 and C(IDEA, NSGA-II) of 0.701 are superior to the ratios C(DEA, IDEA) of 0.286 and C(NSGA-II, IDEA) of 0.052, respectively. Wilcoxon matched-pairs signed-rank test confirms there is a significant difference between IDEA and the other methods. In summary, the above experimental results confirm that the IDEA outperforms both the DEA and NSGA-II in finding the better Pareto-optimal solutions.

Conclusions

In the study, the IDEA shows its effectiveness to optimize task scheduling and resource allocation compared with both the DEA and the NSGA-II. Moreover, for decision makers, the Gantt charts of task scheduling in terms of having smaller makespan, cost, and both can be selected to make their decision when conflicting objectives are present.

Introduction

Cloud applications are very popular in recent years. Especially, cloud computing has emerged as a promising approach to rent IT infrastructures on a short-term pay-per-usage basis. With cloud computing, companies can scale up to massive capacities in an instant without having to invest in new infrastructure, train new personnel, or license new software. Cloud computing is of particular benefit to small and medium-sized businesses who wish to completely outsource their data-center infrastructure, or large companies who wish to get peak load capacity without incurring the higher cost of building larger data centers internally. In both instances, service consumers use what they need on the Internet and pay only for what they use. Thus, operators of so-called Infrastructure-as-a-Service (IaaS) clouds, like Amazon EC2 [3], let their customers allocate, access, and control a set of virtual machines which run inside their data centers and only charge them for the period of time the machines are allocated. Therefore, workflow management on cloud computing becomes more important, when many tasks are sent to cloud environment at the same time.

Researches on specification and scheduling of workflows have concentrated on temporal and causality constraints, which specify existence and order dependencies among tasks. However, another set of constraints that specify resource allocation is also equally important. Execution of a task has a cost and this may vary depending on the resources allocated. Resource allocation constraints define restrictions on how to allocate resources, and scheduling under resource allocation constraints provide proper resource allocation to tasks [27], [5], [12], [21], [10], [19], [38], [14], [18]. In cloud computing environment, machines are located in different regions and have disparate processing abilities, characteristics (number of CPU cores, amount of main memory, etc.), and cost. In these situations, the cost and makespan associated with the task schedule and the resources allocated should be taken into account. Therefore, task scheduling and resource allocation should be carefully coordinated and optimized jointly in order to achieve an overall cost and time-effective schedule. That is, to find optimal task schedules by minimizing cost and makespan.

This study developed cost and time models for computing the cost and time in a task schedule. The cost model includes the processing and receiving cost and the time model incorporates receiving, processing, and waiting time. A parallel-machine scheduling involving both task processing and resource allocation was studied by using an improved differential evolution algorithm (IDEA). The proposed IDEA combines Taguchi method [23], [22], [32] and a differential evolution algorithm (DEA) [30], [31], [17]. The DEA had a powerful global exploration capability on macro-space and uses fewer control parameters. The systematic reasoning ability of the Taguchi method was used to exploit the better individuals on micro-space to be potential offspring. The objective is to find optimal task schedules by minimizing total cost and makespan. Therefore, the multi-objective optimization approach, which was the non-dominated sorting technique [28], [8], [9], not with normalized single-objective method, was applied to find the Pareto front of total cost and makespan. Finally, based on suitable schedules from the Pareto-optimal solution sets, efficiency of the task distribution and the resource utilization were discussed and analyzed.

The paper is organized as follows. Section 2 defines the research issue. The related works are briefly described in Section 3. The proposed approach is illustrated in Section 4. Section 5 shows case study, results, and discussions. Finally, Section 6 concludes the study.

Section snippets

Problem description

To generalize the discussion, the assumption is that there is a set of cloud customer tasks and each task has many subtasks with precedence constraints. Each subtask is allowed to be processed on any given available resource. A cloud resource has a given level of capacity (e.g., CPU, memory, network, storage). A subtask is processed on one resource at a time and the given resources are available continuously. The cloud computing discussion in this study highlighted the fact that either cloud

Related works

This section gives a brief review about differential evolution algorithms, Taguchi method, and a multi-objective approach used in this study.

The proposed approach for task scheduling and resource allocation

The proposed IDEA includes two calculated models for cloud computing and five new ideas for scheduling. They are shown in the following. The cost and time models are proposed for cloud scheduling. The cost model includes rent cost for processing and receiving subtask. The time model incorporates receiving, processing, and waiting time. (1) A subtask-resource-based technique is proposed to encode a schedule. The technique is useful to avoid generating infeasible individual on initialization,

Design examples and comparisons

The performance index for measuring the quality of Pareto-optimal sets and two examples were used for a performance comparison of the IDEA, DEA (i.e., IDEA without the Taguchi method), and NSGA-II.

Conclusions

By introducing the Taguchi method within the DEA framework, the proposed IDEA approach preserves the global solution finding merit of the DEA as a design tool to optimize task scheduling and resource allocation on cloud computing environment. The major contribution of the IDEA is the use of an OA of the Taguchi method as mask mutation operator to account for task-subtask encoding in order to generate improved offspring. Doing so obtains Pareto-optimal solutions within two objectives based on

Acknowledgment

This work was supported in part by the National Science Council, Taiwan, ROC, under Grant Numbers NSC99-2221-E-151-071-MY3, NSC100-2221-E153-001 and NSC101-2221-E153-003.

References (44)

  • S. Bandyopadhyay et al.

    Multiobjective GAs, quantitative indices, and pattern classification

    IEEE Transactions on Systems, Man, and Cybernetics—Part B: Cybernetics

    (2004)
  • C.A. Coello et al.

    Evolutionary algorithms for solving multi-objective problems

    (2007)
  • K. Deb

    Multi-objective optimization using evolutionary algorithms

    (2004)
  • Deb, K., Agrawal, S., Pratab, A., and Meyarivan, T., 2000. A fast elitist non-dominated sorting genetic algorithm for...
  • K. Deb et al.

    A fast and elitist multiobjective genetic algorithm: NSGA-II

    IEEE Transactions on Evolutionary Computation

    (2002)
  • A. Field

    Discovering statistics using SPSS

    (2006)
  • S. García et al.

    A study on the use of non-parametric tests for analyzing the evolutionary algorithms’ behaviour: a case study on the CEC’2005 special session on real parameter optimization

    Journal of Heuristics

    (2009)
  • D.E. Goldberg

    Genetic algorithms in search, optimization and machine learning

    (1989)
  • C.H. Hsu et al.

    Multicriteria tradeoffs in inventory control using memetic particle swarm optimization

    International Journal of Innovative Computing, Information and Control

    (2009)
  • W.H. Ho et al.

    Parameter identification of chaotic systems using improved differential evolution algorithm

    Nonlinear Dynamics

    (2010)
  • Okabe T, Jin Y, Sendhoff B. A critical survey of performance indices for multi-objective optimization. In: Proceedings...
  • Pandey, S., Wu, L., Guru SM, Buyya R. A particle swarm optimization- based heuristic for scheduling workflow...
  • Cited by (240)

    View all citing articles on Scopus
    View full text