| Massively parallel computers are increasingly being used to solve large, I/O intensive applications in many different fields. For such applications, the I/O requirements quite often present a significant obstacle in the way of achieving good performance, and an important area of current research is the development of techniques by which these costs can be reduced.Firstly, the sequent I/O and the parallel I/O methods in parallel environment are discussed inthis paper. These methods are implemented by using MPI-1 and MPI-2 respectively. And their I/O performance is also analyzed. Theoretical analysis and experimental results show that the parallel I/O by using MPI-1 can obtain higher I/O bankwidths than other I/O implementation. Therefore, the parallel I/O can provide an efficient way for achieving higher I/O performance.Furthermore, how to achive high I/O performance in the presence of noncontiguous requests is discussed in this paper. The I/O access patterns in parallel programs often consist of accesses to a large number of small, noncontiguous pieces of data. If many small, distinct I/O requests are made in an parallel application, the I/O performance often degrades drastically. To avoid this problem, two approaches are analyzed in this paper. One approach is two-phase I/O, where an I/O strategy is developed to reduce the number of I/O requests and increase the size of I/O requests. A much better use of the I/O subsystem can be achieved by using this I/O strategy. We discribe two-phase I/O algorithms and implement them in some parallel programming examples. Moreover, the I/O performance is also analyzed. Experimental results show two-phase I/O can significantly reduce the cost of performing the non-contiguous I/O. Another approach is view I/O. A key feature of MPI-IO is that it allows users to access several noncontiguous pieces of data from a file with a single I/O function call by defining file views with derived datatypes. We... |
PDF Full Text Request