-CoPs Project


The COPS project

Memory System
Performance Characterization


Memory System Performance Characterization with 
ECT memperf - Extended Copy Transfer Characterization

ECT memperf is a method to characterize the performance of memory systems (ISCA95 and HPCA97). It captures two aspects of the memory hierarchy, its behavior with temporal locality by varying the working set size and the spatial locality by varying the access pattern (in our case we work with strides). The calculated value is the transfer bandwidth (for a large amount of data). Furthermore we can use the same chart for characterization of local and remote transfers in other words accesses from computation and from communication. And we can do this regardless of the underlying architecture.

More details, documentations and results of the memperf micro benchmark are given in a talk held at an ISCA98 workshop in Barcelona.

New Version:

The ECT memperf now evaluates the measurement values out of multiple run. For Pentium or Athlon systems an optimized copy loop using SSE and MMX commands was implemented to allow a comparison of the best possible result versus the compiler generated result.


Our measured results for the following machines are available online: 

  • Pentium Pro 200 FX
  • Pentium II 233 LX
  • Pentium II 400 BX
  • Pentium III 600 BX
  • Pentium III 600 i820
  • Pentium III Xeon 700 Profusion (Multiprocessor)
  • Pentium III 800 440GX+
  • Pentium III 866 VIA694XDP
  • Pentium III Xeon 800 i840
  • Pentium III 933 ServerSet LE
  • Pentium III 1000 ServerSet LE
  • Pentium III Xeon 1.7 GHz i850
  • Pentium 4 1.8 GHz i850
  • Pentium 4 1.9 GHz i850

  • DEC 8400 (Alpha 300MHz)
  • AlphaStation ES40 (Alpha 21264A 667 MHz)

  • SUN Enterprise SMP (USPARC II 250MHz)
  • SUN Ultra 5-10 (USPARC IIi 440MHz)

  • SGI Origin 2000 (R10000 195 MHz)
  • SGI O2 (R5000 300 MHz)
  • SGI O2 (R12000 270 MHz)

  • CRAY T3E (Alpha 300MHz)

memperf - Download

Download You can download our memory micro benchmark memperf_v0.9h.tar.gz. It runs on standard UNIX systems and Windows. For Pentium-Systems the Pentium Performance Counters are used for time measurement, for Alpha systems the Alpha Clock Counter can be activated.
We provide binaries for x86-Linux (gnu 2.95.2), Windows (Visual C++ 6.0 SP4) as well as Alpha (Compaq C V6.1-015) and Sparc (Sun WorkShop 6 update 2).
Read the README and INSTALL file for further information. You will find the last changes in the Changelog.

Relevant Materials:

  • Ch. Kurmann, T. Stricker Characterizing memory system performance for local and remote accesses in high end SMPs, low end SMPs and clusters of SMPs. 7th Workshop on Scalable Memory Multiprocessors held in conjunction with ISCA98, June 27-28 ,1998, Barcelona, Spain.
    Available formats: abstract, acrobat, postscript, compressed.
    Talk: acrobat, postscript, compressed.
  • Ch. Kurmann, T. Stricker A Comparison of two Gigabit SAN/LAN technologies: Scalable Coherent Interface versus Myrinet. Appears in Proceedings of the SCI Europe'98 Conference, EMMSEC'98, 28-30 Sept 1998, Bordeaux, France.
    Available formats: abstract. acrobat, postscript, compressed.
  • T. Stricker, T.Gross Global Address Space, Non-Uniform Bandwidth: A Memory System Performance Characterization of Parallel Systems Reprint from proceedings of HPCA'97, Feb 1-5,1997, San Antonio, TX.
    Available formats: abstract. acrobat, postscript, compressed.
  • T. Stricker and T. Gross. Optimizing Memory System Performance for Communication in Parallel Computers . Reprint from proceedings of ISCA'95, June 1995.
    Available formats: abstract. acrobat, postscript compressed

ICS - Lab for Computersystems
DINFK - Departement of Computer Science
ETHZ - Swiss Institute of Technology

Departement Informatik der ETH Zürich
Mar 2002
Christian Kurmann, < kurmann@inf.ethz.ch >