Cadre provides insight in petascale I/O patterns of parallel systems

 Urbana 17 May 00 The University of Illionois inititated an US National Science Foundation funded project called Cadre. Cadre is a facility to build a base of empirical data on I/O access patterns of high-performance computing systems. It will enable researchers and educators to overcome limitations imposed by I/O on the overall performance of petascale computing systems. Tools used through the CADRE facility reflect insight gained through development of the Pablo I/O characterization toolkit, software tools for the performance analysis and optimization of parallel and distributed systems, the umbrella project of which Csdere is a part.

The Pablo Research Group, part of the university's department of computer science, investigates the interaction of architecture, system software, and applications on large-scale parallel and distributed computer systems. It is led by Dan Reed, head of the computer science department and director of the National Computational Science Alliance (Alliance).

CADRE is a Web-based facility, www-pablo.cs.uiuc.edu/Project/CADRE/ " target="_new www-pablo.cs.uiuc.edu/Project/CADRE/ ), that is extending, documenting, archiving, and disseminating software tools, sample applications, and experimental data to advance research on I/O system design, analysis, and optimization for high-performance computing environments. CADRE is using:

  • The Pablo portable I/O performance analysis tools that capture and reveal the I/O patterns of data-intensive applications executing on high-performance systems.
  • Tutorial material covering I/O characterization and analysis, as well as material specific to use of Pablo tools for I/O optimization.
  • A repository of I/O traces, obtained from execution of instrumented I/O-intensive applications. This repository contains both the raw traces as well as a database of I/O data that can be searched and queried for specific metrics and data. These traces provide file system and storage hierarchy designers and other I/O researchers with ready access to empirical data exposing the interdependencies among access patterns, I/O APIs, library implementations, file system features and policies, and storage hardware configurations. Trace files contained in the database can be downloaded, ordered for delivery on CD-ROM or DVD, or used with analysis tools online. See bugle.cs.uiuc.edu :1303/cgi-bin/cadre.pl bugle.cs.uiuc.edu :1303/cgi-bin/cadre.pl).

The high-performance community is already tapping the resources made available by the CADRE facility. A team from Northwestern University in the US recently used the Pablo tools to analyze the I/O activity of a parallel 3D cosmology hydrodynamics code called Enzo. Running the code successively on 32, 64, and 128 processors at the National Center Supercomputing Applications (NCSA), trace files were generated and analyzed using the Pablo performance toolkit.

 

Ad Emmen

[News on Advanced IT]   [Calendar]   [Analysis]   [IT in Medicine]