This article put forward a NCM_Apriori algorithm, which through compressing matrix and reducing the scan times to reduce the database I/O overhead, effectively improve the efficiency of association rule mining. At the same time in the process of generating association rules, computation is greatly reduced by using the nature of probability. And applies the algorithm to the mining of students' course selection system, which can provide decision support for colleges and universities.

As a typical problem of graph theory, Single-Source Shortest Path(SSSP) has a wide range of applications and research. The traditional MapReduce framework such as Hadoop has been applied to SSSP problem. However, in this way, it may need writing and reading the disk frequently, transmitting data largely, beacuse of SSSP’s iterative. Haloop is a parallel programming framework which makes an improvement based on Hadoop framework, and adapts itself to iterative programming. Hence, this article represents the SSSP problem with an iterative way firstly, and then we put forward the implementation mechanism of SSSP algorithm based on Haloop. Though testing and analyzing, the implementation based on Haloop obviously improves the efficiency of program execution, compared with the traditional realization mechanism.

A metadata management mechanism on the basis of HDFS was proposed in this paper, which is applied for cloud storage. The basic unit of consistent hashing is DHU in the scheme. Metadata is distributed in multiple servers in this solution, and the position of metadata is recorded in a mapping table. In addition, prefetching strategy and local cache technology are used to improve the performance of the system. Operation logs and Multi-Paxos algorithm are being used to ensure system’s reliability and consistency. Improved system with a lower latency in reading and writing than the original system was seen in the experiments. Good performance for cloud storage system can be shown by this system.

For the issues of quality of service (QoS) in the cloud computing raised by users, this paper proposes a strategy for QoS classification and builds tasks' priority function modeling and through priority scheduling tasks, assigning tasks to the reasonable resources, and finally to complete the task efficiently, improve the utilization of resources.

In order to make up for the shortage of Min-Min in load balancing, a new task scheduling algorithm T-Max-Int Under the grid computing has been proposed in this paper. In T-Max-Int, the Loss Degree of Max-Int has been brought into Min-Min. T was in the form of percentage, which represents the proportion of selected tasks that have loss degree in the total tasks. Then, experiments of T have been taken to make Makespan the minimum. Finally, T-Max-Int, Max-Min, Min-Min were compared, which proved that T-Max-Min is better than the other two algorithms in aspects of Makespan and load balancing.