Design And Implementation Of A Parallel Program Debugger In Cluster System

As the increasing applications of the parallel computation in military affairs, science and technology, and business, more users use the parallel technology as a method to solve technical problems. However the weakness of current developing environment of parallel program limits the widely application of parallel computation. Under such background, an integrated environment for developing parallel programs needs to be developed. It reduces the complexity of using parallel computing system. The parallel program debugger is the main part of the developing environment, the design and implementation of it is still a challenging.By studying existing parallel program debuggers, this paper designs and implements a parallel program debugger in cluster system. This debugger allow users to debug Unix/Linux platform's PVM or MPI parallel programs on Windows platform long-distance, and more Windows programmers to develop the parallel programs efficiently. It also solves the problems raised by the complexity and difficultly of mutual operation using parallel computing system, so that the users can put more energy on application question. It promotes the widely application of parallel computation. Compared with the related work, it reduces the complication of developing parallel programs, and improves the efficiency of using cluster system. It has highly practical value.The major work of this thesis is as below:By analyzing the control block problem of existing parallel program debuggers, the author put forward an improved control model. The debugger's system structure is fit long-distance debugging.By analyzing the inconsequence of control model in existing long-distance parallel program debugger, the author put forward an improved long-distance control mode. It is more fit to debug parallel programs on cluster long-distance, and effectively avoids the system expenses during processes exchange.The author studies the related communication technology about long-distance debugging, and designs a method of controlling command range and a method of managing breakpoint.The author presents an improved blue print which using the new machine interface GDB/MI by studying the portable realization of parallel program debugger and the problems existed in scalability, reliability, and maintainability, so the debugger can run more stably and high effectively.Through study on the global predicate detection technology which handles uncertainty problem, the author brings forward an additional predicate conception, and applies the technology of global predicate detection into sharing memory system.Based on the research above, the author designs a system structure of debugger which possesses the function of long-distance debugging, and implements a parallel programs debugger in cluster systems according to the system structure.