Distributed Parallel To The Design And Realization Of The Function Library

Along with the development of the computer industry, software applications are complex day by day, and the reliability and the performance are requested more and more. The high speed development of computer hardware can partially solve these problems, but the high performance hardware equipment is too expensive to popularize. The development of the distributed parallel technology and the popularization of computer network enable us to use the inexpensive ordinary computer to develop software with high performance and high reliability. However, the development of distributed parallel system also has problems of high cost and long development cycle and difficult to adapt the fast changing market demand. In view of these problems, we proposed the distributed parallel function library(Dplib). It provides many components which the distributed parallel system needs and solves the problems such as node management, communication, and data consistency.In order to solve problems of the distributed parallel system such as the node management and communication, this thesis designs the"commtor"module and proposes the concept of"communication field"to limit the communication range of distributed parallel systems. The"commtor"module automatically completes the node management work, and designs the communication APIs which have the same semantic of the MPI standard [literature 3-5]. The"commtor"module is the foundation of the entire Dplib. All the other modules rely on this module.The data consistency question is one of the most important issues in the distributed parallel technology. Provides a general consistency solution is the key point of this thesis. The Lamport consistency model [literature 6] points out, we can guarantee the data consistency if the related operations (operating on the same identical source) in each data repetitions have same sequence view. According to this thought, this thesis decomposes the consistency question into two aspects: the transaction result consistency and the transaction sequence consistency. We designed two modules to solve these problems: the"synctor"module and the"pipeline lock"module. The"synctor"module is able to call the same function at the remote nodes. It also wraps...