The Analysis Of MPI-2 And Improving I/O Performance On Clusters Using MPI-2

As parallel computers get faster, scientists are increasingly using parallel computers to solve problems where a large amount of computing power is required. In recent years, great advances have been made in the CPU and the communication performance in parallel computers. However, similar advances have not been made, on the input/output(I/O) performance in parallel machines. I/O performance has not been put more intention on.At present, MPI is most widely used in parallel computing environment of high-performance computer systems. MPI has come to be the standard of message passing mode in parallel programming. MPI has been extended to MPI-2 for it's rising impact among parallel computing environment by MPI Forum, which is responsible to define and maintain the MPI standard. Three main topics are extended in MPI-2 including process creation and management ("PCM" for short), remote memory access ("RMA" for short) and parallel I/O.In this thesis, the implementation of MPI-2 standard is studied and analysed. We first give an overview of MPI-2 and high-performance computer systems with MPI. Then we study the details of the design and implementation of RMA and PCM which are the main parts of MPI-2 standard. Application evaluation is also provided.