Research And Implement On A JXTA-Based Distributed Parallel Computing Middleware System

The rapid rise of the Internet and Intranet, especially the fast development of the P2P overlay network, has been due to the seamless access the information that is distributed in remote locations across the network. The sharing of information is currently the major factor in the growth of networks. With further development of the network, distributed computing becomes the main impetus of the network. By building computing resource network of high usability and fault tolerate capability, the P2P network makes possible for the reality of distributed computing.This paper describe the design and implement of a middleware named DPCF on distributed parallel computing, and it is based on the P2P system JXTA of Sun. It organizes the host that participated in the computing on the network into a distributed virtual machine suitable for parrllel computing. Firstly, based on the research of the remote method invocation system (RMI) of Java, a new remote method invocation system named DPCF Runtime is designed for distributed parrllel computing. It provides the function of the processor for the distributed virtual machine. Secondly, by leveraging the Dirtributed Hash Table, a distributed sharing object model and its related naming service are implemented. It provides a globle sharing object space for the distributed virtual machine. Lastly, two type of parallel computing model are researched for the principle of the peer group organization, and a set of peer group building protocol are implemented.DPCF is written entirely in pure Java and that run on any standard compliant Java virtual machine. Many new features are implemented in DPCF Runtime, such as: No any preprocessor and precompiler; RMI-liked programming model; objects created on remote hosts; asynchronous remote method invocation; migrating objects to heterogeneous hosts; scalable support for multi-protocol of network. The naming service based on JXTA makes good support for sharing information and synchronization between remote hosts. Experimental results show that DPCF middleware system performs good speedup for typical parallel computing. By using...