• Fast I/O (Thursday 1:30-3:00PM)
    Room A110/112
    Chair: Richard F. Barrett, Los Alamos National Laboratory

    • Title: MPI-IO/GPFS, an Optimized Implementation of MPI-IO on top of GPFS
    • Authors:
      Jean-Pierre Prost (IBM T.J. Watson Research Center)
      Richard Treumann (IBM Enterprise Systems Group)
      Richard Hedges (Lawrence Livermore National Laboratory)
      Bin Jia (IBM Enterprise Systems Group)
      Alice Koniges (Lawrence Livermore National Laboratory)
    • Abstract:
      MPI-IO/GPFS is an optimized prototype implementation of the I/O chapter of the Message Passing Interface (MPI) 2 standard. It uses the IBM General Parallel File System (GPFS) Release 3 as the underlying file system. This paper describes optimization features of the prototype that take advantage of new GPFS programming interfaces. It also details how collective data access operations have been optimized by minimizing the number of messages exchanged in sparse accesses and by increasing the overlap of communication with file access. Experimental results show a performance gain. A study of the impact of varying the number of tasks running on the same node is also presented.

    • Title: A Case Study in Application I/O on Linux Clusters
    • Authors:
      Robert Ross (Mathematics and Computer Science Division, Argonne National Laboratory)
      Daniel Nurmi (Mathematics and Computer Science Division, Argonne National Laboratory)
      Albert Cheng (Scientific Data Technologies Group, NCSA)
      Michael Zingale (Center on Astrophysical Thermonuclear Flashes, University of Chicago)
    • Abstract:
      A critical but often ignored component of system performance is the I/O system. Today's applications demand a great deal from underlying storage systems and software, and both high-performance distributed storage and high level interfaces have been developed to fill these needs.

      In this paper we discuss the I/O performance of a parallel scientific application on a Linux cluster, the FLASH astrophysics code. This application relies on three I/O software components to provide high-performance parallel I/O on Linux clusters: the Parallel Virtual File System, the ROMIO MPI-IO implementation, and the Hierarchical Data Format library. Through instrumentation of both the application and underlying system software code we discover the location of major software bottlenecks. We work around the most inhibiting of these bottlenecks, showing substantial performance improvement. We point out similarities between the inefficiencies found here and those found in message passing systems, indicating that research in the message passing field could be leveraged to solve similar problems in high-level I/O interfaces.

    • Title: The Design of I/O-efficient Sparse Direct Solvers
    • Authors:
      Florin Dobrian (Old Dominion University )
      Alex Pothen (Old Dominion Univ. & ICASE )
    • Abstract:
      We consider two problems related to I/O: First, find the minimum primary memory size required to factor a sparse, symmetric matrix when permitted to read and write the data exactly once. Second, find the minimum data traffic between core and external memory when permitted to read and write the data many times. These problems are likely to be intractable in general, but we prove upper and lower bounds on these quantities for several model problems with useful sparsity (i.e., whose computational graphs have small separators). We provide fast algorithms for computing these quantities through simulation for irregular problems. The choice of factorization algorithms (left-looking, right-looking, multifrontal), orderings (nested dissection or minimum degree), and blocking techniques (1- or 2- dimensional blocks) can change the memory size and traffic by orders of magnitude. Explicitly moving the data (files managed by the program) improves performance significantly over implicit data movement (pages managed by the operating system). Thus this work guides us in designing a software library that implements an external memory sparse solver.