The Research Of Difference Evolution Algorithm And Its Applications In Task Scheduling Of Cloud Computing

Cloud computing is a business model, it distributes computing tasks among the resourcepools, which are constituted by a large number of computers, and enables users to get ITservice on demand. The performances of these computers in computing, storage, broadbandare limited. It often requires complex task scheduling to meet users’ diverse and elasticdemand of cloud computing. But it is still too hard to meet the needs of users’ QoS (Quality ofService). Cloud tasks scheduling is proposed to solve this problem.Cloud task scheduling is a NP-hard problem of a kind of combinatorial optimization.There are inadequacies in existing cloud task scheduling algorithms, such as FIFO, fairscheduling, capacity scheduling, and so on. It is difficult to meet the demand for practicalapplication. Therefore, in various studies some new algorithms have been proposed gradually.In addition to improving the classic scheduling algorithm, the intelligent optimizationmethods are gradually introduced, such as genetic algorithms, particle swarm optimization,ant colony algorithm, the simulated annealing algorithm. But they are in the exploration andresearch phase, also still have their advantages and disadvantages.Differential evolution algorithm as a new bionic intelligent optimization method basedon groups of evolution, only in recent years caused many researchers attention. The algorithmis simple to use and has the ability to achieve global optimization, etc., but it also hasinadequate point. It cannot be directly used for discrete problems. It is easy to fall into localoptimum sometimes. It has slow convergence speed sometimes and so on. It has not beenapplied to the area of cloud computing yet. This paper intends to improve the differentialevolution algorithm and applies it to the task scheduling problem of cloud computing.Improvement divides into two steps: Firstly, improve the deficiency of basic differentialevolution algorithm; then, according to the characteristics and needs of the task scheduling ofcloud computing, make further improvements.To overcome the lack of differential evolution algorithm itself, this paper proposesMDDE (Modified Discrete Differential Evolution) algorithm. A discretization principle isproposed to enable the differential evolution algorithm to apply to combinatorial optimizationproblems. Then, to make up the deficiency of optimization performance of differentialevolution algorithm, we propose the concept of anterior and posterior evolutionary process.The parallel MULTIGROUP is setting in the anterior evolutionary process. And configureintegrate differential strategy DE1and DE2in anterior and posterior evolutionary process tobalance the global exploration and local mining capacity, ensuring that the MDDE algorithmhas good global exploration capacity in the anterior evolutionary process while in theposterior evolutionary process it has fast convergence performance with good local mining capacity. The greedy crossover principles and a new mutation operation selection mechanismare proposed to accelerate the speed of convergence of the MDDE algorithm. Then we applythe MDDE algorithm to the TSP, and verify its optimized performance.According to the characteristics of the cloud task scheduling, and based on the MDDEalgorithm, this paper proposes a new task scheduling algorithm of cloud computing:TC-MDDE. According to coding method we configure and discretization principle, we define"absolute value rounded remainder mapping method" to solve the illegal coding problem inthe differential operator. The legalization method can ensure the smooth progress of themutation operation. To meet the diversified demand for cloud computing, flexible fitnessfunction of the QoS parameter is defined. According to the different users’ time and costrequirements, adjust the time and cost weight coefficient to adjust scheduling result.In order to verify the function and optimize performance of the TC-MDDE algorithm,this paper deploys the TC-MDDE algorithm to cloud simulation platform: CloudSim. Werandomly generate the data matrixes, which simulations require, of tasks requirements andresource performance. Do the functional verification experiments by running the TC-MDDEalgorithm in different weight coefficient. Do the verification experiments of optimizeperformance by comparison with other scheduling algorithms. The results of functionalverification experiment show that the TC-MDDE algorithm is able to meet the schedulingresult of the different time and cost requirements, and achieve the diversification of satisfyingthe QoS demand by proper adjustment of weight coefficient. The results of verificationexperiments of optimize performance show that the TC-MDDE algorithm is better than otherscheduling algorithms, the result of the convergence speed and optimization are morecompetitive.