Energy-aware Cloud Workflow Scheduling With Geo-distributed Data

In recent years,the power consumption of data centers has been increasing in the proportion of global power consumption.On the one hand,the electrical consumption brings about a large amount of environmental pollution;on the other hand,the electricity cost of data centers,generated by electricity consumption,accounts for more than 30% of the total operating costs of cloud service providers.How to reduce the energy consumption of data centers becomes an urgent problem.An ECWSD(Energy-aware Cloud Workflow Scheduling with geo-distributed Data)algorithm is proposed to minimize the electricity cost as well as take the optimization of energy consumption for workflow applications into account,which exerts important practical significance and has promising prospects.This paper considers energy-aware workflow scheduling in the geo-distributed data centers.Both two types of data transmission time(generated data transmission time and original data transmission time)and the diversity of electricity prices are concerned to minimize the electricity cost of service providers.Based on involved characteristics and properties of the considered problem,this paper builds the mathematical model for the optimization goal and constraints.Then a heuristic ECWSD algorithm is proposed,the ECWSD algorithm consists of five component: workflow application sequencing,the deadline partitioning,initial task sequence constructing,DVFS-based resource allocating and the VND-based task sequence updating.After finishing the sorting of submitted workflow applications,the workflow application is scheduled alternatively according to the sorted workflow sequence.The subdeadline of each task in the workflow application is partitioned,and an initial task scheduling sequence is subsequently created.The resource allocating method is then proposed to assign each task to the appropriate VM with the minimum electricity cost.Since task sequences have great influence on the scheduling performance.In order to search better solutions for the considered problem,new task sequences are generated by employing the VND with dynamically adjusted neighborhood structures.In order to verify the performance of the proposed algorithm,this paper designs two experimental modules,i.e.parameter calibration and algorithm comparison.Firstly,algorithm parameters and sequence method combinations are analyzed using ANOVA(Analysis of Variance),and the optimal parameter combination and sorting rules are selected.Then considerable scientific workflow cases are introduced to compare the proposed algorithm with two related algorithms.Experimental results indicates that the proposed algorithm has better performance than two benchmark algorithms under different deadline candidates.