Ant Colony Optimization Of Virtual Machine Placement For Data Latency Minimization In Cloud Systems

Since the rapid development of the Internet hardware and software, people’s lifestyle has undergone tremendous change. Cloud computing as a new model came into service usage and interaction, this pattern occurs in the network services more and more reasonable and efficient. With more and more data-intensive applications displaced to the cloud, big data processing research got people’s attention. In order to effectively and quickly process the application in cloud, people often use the Map Reduce / Hadoop which is distributed processing framework. Big data in this kind of computing framework entered will be divided into several sub-blocks of data, and these data blocks are independent with each other. In this case, if the maximum access latency of data between the data storage nodes and its corresponding computing nodes has not been limited, then the total completion time of the task will be delayed considerably. Moreovercomputing nodes also need to communicate with each other in cloud to collect operating results when processing big data. Therefore the maximum access latency between computing nodes which allocated for processing data node also require restricted to control the total completion time.In this paper, the basic concept and the background of cloud computing, features and architecture will be introduced first. Then network topologies in data center and the similarities and differences between distributed computing cloud computing architecture and traditional cloud will be shown to carry out the main work of this paper. Mainly do the following work: First, according to the characteristics of distributed cloud computing environments virtual machine scheduling problem, the problem is studied in formal mathematical model, and we will analyze the current status of the problem we researched in. Secondly after a brief introduction about the ant colony algorithm and inherent shortcomings of basic ant colony algorithm we adopt the Max-Min ant colony algorithm to be our basic algorithm in this paper, and based on detailed analysis of the specific characteristics of our problem, we will raise some improvement idea in basic Max-Min ant colony algorithm to optimize the performance of algorithm. For example, to solve the random selection of the initial path in ACO we proposed initial position selection based on the optimal iteration and made some improvements in pheromone update strategy in maximum and minimum ant colony algorithm. At the same time we propose two local search strategy: double mutation search technology as well as K-NN search technology to improve the quality of the convergence. Given the ant colony algorithm is a kind of parallel algorithms, so we introduce parallel ant colony algorithm to speed up the convergence rate in engineering way. Third, based on Matlab simulation software, we first evaluate the performance of the algorithm which we adopted by adjusting the initial parameters, and at same time the total access latency, maximum access latency, and the optimal solution get from each iteration are obtained as the basis for judging the advantages and disadvantages between algorithm we proposed and the current of the optimal algorithm. Furthermore, the experimental analysis will be given in four new data center network topologies, and in these framework the maximum access latency is measurable criteria to evaluate the performance of between two algorithms. Through the above two experiments in this paper prove algorithm we mentioned is effective and quality.