NCSA NEWS

CADRE Expands Resources for Design of Efficient I/O Systems

released March 20, 2001

 

Contacts
Karen Green
Public Information Officer
kareng@ncsa.uiuc.edu
217.265.0748

Deb Israel
Pablo Research Group, UIUC
disrael@cs.uiuc.edu
217.333.8426

IBM, Intel donate hardware to NSF-fund effort

URBANA, IL — CADRE, a National Science Foundation-funded facility for high-performance I/O characterization and optimization, announces a newly expanded repository of resources that aid in the design of efficient I/O systems. The upgrade to the Web-based CADRE repository (http://www-pablo.cs.uiuc.edu/Project/CADRE/) was made possible by hardware contributions from IBM and Intel and by the addition of new software tools.

An SPII Nighthawk, donated by IBM, and a 28-node Linux cluster, donated by Intel, are enabling CADRE to collect and distribute performance data from a wider range of parallel systems. The Intel cluster consists of 28 933MHz Pentium III Xeon systems, each with two 18-gigabyte disks for I/O experiments and measurement. The Nighthawk, a two-node POWER3 SMP with eight 375MHz RISC processors per node and a maximum memory of 64 GB, was donated to a group of computer scientists at the University of Illinois at Urbana-Champaign by IBM as part of its Shared University Research Program.

"As a community of researchers, we've uncovered a piece of the answer," said Luiz DeRose, research member and tools group leader at IBM's Advanced Computing Technology Center. "We know that access pattern variations have deep performance implications for I/O libraries and file systems. Now, tools and data contained in the CADRE repository can help expose another piece: the application usage patterns and economic tradeoffs that must be considered in optimizing the I/O of diversely configured, next-generation environments. IBM's SP donation underscores our support of this objective."

In addition to the new computer systems, CADRE researchers recently released new tools that will help expand the base of empirical data in the CADRE repository. The Pablo Performance Capture Facility (PCF) is a cross-platform tool designed to analyze the I/O activity of application codes running under Linux, Solaris, AIX, and IRIX.

Another new characterization tool, the Pablo Physical I/O Tracing Facility, reveals the correlation between application I/O requests and physical I/O operations. This is significant because physical I/O patterns are strongly affected by data striping mechanisms, file system policies, and disk hardware attributes. Understanding how the operating system translates application I/O requests into physical disk operations can aid in optimizing file policies and data distributions for higher performance. The initial release of the Physical I/O Facility supports Linux systems and Linux clusters.

CADRE was launched last year as a Web-based facility that extends, documents, archives, and disseminates software tools, sample applications, and experimental data to stimulate research on I/O system design, analysis, and optimization for high-performance computing environments.

"Without a substantial base of empirical data on the I/O access patterns from high-performance systems, file and operating system developers haven't been able to overcome the limitations that I/O imposes on performance," said Dan Wells, principal investigator for the CADRE project.

Dan Reed, director of the University of Illinois' National Center for Supercomputing Applications and the National Computational Science Alliance, also noted, "CADRE's focus is to collect and distribute a comprehensive body of data characterizing the I/O behavior and stipulating the requirements of large, next-generation simulations and other data-intensive applications that run on diverse parallel computers."

CADRE is operated by the Pablo™ Research Group. Founded in 1984, the Pablo Research Group (http://www-pablo.cs.uiuc.edu/) is an academic research group within the department of computer science at the University of Illinois at Urbana-Champaign. Under the direction of Professor Dan Reed, members of the group investigate the interaction of architecture, system software, and applications on large-scale parallel and distributed computer systems. Key research foci are exploration of performance analysis techniques, compiler-aided scalability analysis, scalable parallel file systems, and real-time adaptive systems for resource policy control.

Pablo is a registered trademark of the University of Illinois Board of Trustees in the United States and/or other countries. Other product and company names herein may be trademarks of their respective owners.

 

Headlines Archive