Research On Cloud Platform Task Scheduling Algorithm

With the rapidly development of network technology, large data era has arrived, a lot of data has been generated every minute and every hour. As a new method of calculation, cloud computing has inherited and developed the grid computing, parallel computing, distributed computing and clustering technology, etc. It integrates computing resources, services information, storage resources and data services so that it can provide more efficient and reliable personalized services for us. However, given the heterogeneous of cloud computing environment, it is crucial for cloud computing system to efficiently utilize the system resources to response users’ needs, at the same time, ensure the system’s load balancing.PSO and ABC algorithm can achieve load balancing, which have been widely applied as task scheduling algorithm under the cloud computing environment. However, PSO and ABC as one of the swarm intelligent algorithm all have a common defect: they can easily fall into local optima and premature convergence. This thesis mainly studies the principles of PSO and ABC algorithm, puts forward a new fusion algorithm-FPA algorithm. This algorithm uses PSO instead the employed stage of ABC algorithm to accelerate the convergence, meanwhile, it takes the advantage of the excellent lateral search capabilities of ABC’s onlookers stage and scouts stage to increase its ability for searching optimal solutions. Through the MATLAB simulation, FPA algorithm improves the ability to step out from the local optimal solution and avoid premature convergence.Hadoop as an open source cloud computing platform, which has the ability to commendably achieve the Map-Reduce calculation model and HDFS proposed by Google. However, it still needs improvements at the task scheduling and load balancing aspects. For the lack of pre-processing functions of the input data, this thesis proposes a concept of data selection layer which divides the data types into text data, radio data, video data and image data. With the properly preprocessing classification of the data, then the data can be distributed to the corresponding fast nodes to process. In this thesis, we come up with a new task scheduling strategy based on the data selection layer compared with the poor performance about load balancing of Hadoop task scheduling algorithm. Based on the data selection layer and combined with the modified BITS algorithm and FPA algorithm, we add a data selection layer to the original framework, we improve the Hadoop framework, and meanwhile, we present a new task scheduling strategy based on data selection layer. With the simulation by CloudSim, we can see that the strategy can improves the performance of Hadoop and the ability of load balancing.Finally, conclude the research of this thesis, raise the deficiency of FPA algorithm and task scheduling strategy based on data selection layer, discuss follow-up studies.