WWC-5: IEEE 5th Annual Workshop on Workload Characterization

November 25, 2002

Austin Marriott at the Capitol

Austin, Texas, USA

WWC-5 Advance Program

Sponsored by IEEE and the Technical Committee on Computer Architecture

7:15 AM

Breakfast and Registration

8:00 AM

Session 1: Web Site Benchmarks and Synthetic Trace Models

Chair: Alex Mericas (IBM)

  • Specification and Implementation of Dynamic Web Site Benchmarks   slide

    Cristiana Amza, Anupam Chanda, Alan L. Cox, Sameh Elnikety, Romer Gil (Rice University), Karthick Rajamani (IBM), Willy Zwaenepoel (Rice University), Emmanuel Cecchet and Julie Marguerite (Rice University / INRIA Rhone-Alpes)

  • Benchmarking a Site with Realistic Workload   slide

    G. Ballocca, R. Politi, V. Russo (CSP, Italy) and G. Ruffo (Universita di Torino, Italy)

  • Evaluating Synthetic Trace Models Using Locality Surfaces   slide

    Elizabeth S. Sorenson and J. Kelly Flanagan (Brigham Young University)

9:15 AM

Coffee Break

9:45 AM

Session 2: Commercial Workloads and Platforms

Chair: Mauricio Breternitz (Intel)

  • A Detailed Comparison of Two Transaction Processing Workloads   slide

    Robert Stets, Kourosh Gharachorloo (HP WRL) and Luiz André Barroso (Google)

  • Comparing and Contrasting a Commercial OLTP Workload with CPU2000 on IPF

    Jeff Rupley II, Murali Annavaram, John DeVale, Trung Diep and Bryan Black (Intel)

  • Optimizing Applications for Performance on the Pentium 4 Architecture   slide

    Arrian Mehis and Ramesh Radhakrishnan (Dell)

11:00 AM


12:00 PM


1:00 PM

Session 3: Java and Multimedia

Chair: John McCalpin (IBM)

  • PennBench: A Benchmark Suite for Embedded Java   slide

    G. Chen, M. Kandemir, N. Vijaykrishnan and M. J. Irwin (Pennsylvania State University)

  • Cache Performance in Java Virtual Machines: A Study of Constituent Phases   slide

    Anand S. Rajan (ARM), Shiwen Hu and Juan Rubio (The University of Texas at Austin)

  • MPEG4: Fallacies and Paradoxes   slide

    Zhen Fang (University of Utah) and Sally A. McKee (Cornell University)

  • Estimating Multimedia Instruction Performance Based on Workload Characterization and Measurement

    Adil Gheewala, Jih-Kwon Peir (University of Florida, Gainsville), Yen-Kuang Chen and Konrad Lai (Intel)

2:40 AM

Coffee Break

3:20 PM

Session 4: Network and Handheld Environments

Chair: Ramesh Radhakrishnan (Dell)

  • Energy Impact of Secure Computation on a Handheld Device   slide

    Zhiyuan Li and Rong Xu (Purdue University)

  • A Methodology for Workload Characterization of File-Sharing Peer-to-Peer Networks   slide

    D. Nogueira, L. Rocha, J. Santos, P. Araujo, V. Almeida and W. Meira Jr. (Federal University of Minas Gerais, Brazil)

  • Workload Characterization of Elliptic Curve Cryptography and Other Network Security Algorithms for Constrained Environments   slide

    A. Murat Fiskiran and Ruby B. Lee (Princeton University)

John L. Gustafson

John L. Gustafson leads the HPC workload characterization effort at Sun Labs in Mountain View, California. He joined Sun in 1999 after many years developing novel computer measurement approaches at Sandia National Laboratories and Ames Laboratory. Dr. Gustafson has won three R&D 100 Awards and has several patents, all in the area of computer performance analysis. His 1987 work with colleagues Gary Montry and Robert Benner at Sandia showed that the assumptions of Amdahl's law might not be appropriate for many HPC workloads, and introduced the paradigm referred to as "Gustafson's law" that removes unnecessarily pessimistic predictions about the power of massively parallel computing.

Keynote Abstract

While business benchmarks have incorporated scaling ideas for years (TPC, ECPerf), only a handful of HPC benchmarks have had true scalability to a range of workloads. LINPEAK, SLALOM, and HINT were early efforts to scale the problem size as opposed to measuring execution time on a fixed problem size. We present benchmark design efforts being undertaken at Sun Labs to characterize HPC workloads in a new way that is totally architecture- independent and language-independent. The benchmarks can be scaled in many dimensions, not just size, to capture aspects such as precision requirements, load imbalance, I/O demands, and the variability of the workload as a function of time. Initial work shows that this approach has high predictive value and can yield better than 0.995 correlation with particular HPC workloads. Sun Labs plans to share this scalable workload suite with the HPC community when it is completed.


The workshop will be held at the Austin Marriott at the Capitol.

WWC-5 | Registration | Call for Papers

For assistance on this page, please send mail to blee @ ece.utexas.edu.