Workshop on Workload Characterization
Austin Marriott at the Capitol
Austin, Texas

September 16, 2000

8:00am - 6:00pm

Advance Program

Registration: 7:15am

Opening: 8:10 am

8:15am - 9:30am
Session 1: Locality and Predictability Characteristics
Chair: Lizy Kurian John, The University of Texas

Source Code Level Classification of Programs for Dynamic Prediction
Sangwook P. Kim, Francis Tseng, and Yale N. Patt,
The University of Texas at Austin

Characterization of Value Locality in Java Programs,
Bohuslav Rychlik and John Paul Shen,
Carnegie Mellon University

Using Locality Surfaces to Characterize the SPECInt 2000 Benchmark Suite,
Elizabeth S. Sorenson, J. Kelly Flanagan,
Brigham Young University

9:30am - 10:00am

10:00am - 11:15am
Session 2: I/O and Memory
Chair: Ann Marie G. Maynard, IBM

Towards an Optimal File Allocation Strategy for Specweb99,
Tom. W. Keller, Karthikeyan Sankaralingam, and H. Peter Hofstee,
Austin Research Lab and The University of Texas at Austin

Characterization of Memory Energy Behavior,
H. S. Kim, M. Kandemir, N. Vijayakrishnan, and M. J. Irwin,
The Pennsylvania State University

On the Impact of Workload Burstiness on Disk Performance,
Maria E. Gomez and Vicente Santonja,
Universidad Politecnica de Valencia, Spain

11:15am - 12:15pm
Keynote Address
Speaker: John D. McCalpin, Ph.D., Senior Scientist, IBM
Title:   An Industry Perspective on Performance Characterization:
         Applications vs Benchmarks

12:15pm - 1:45pm

1:45pm - 3:15pm 
Session 3:Tools and Methodology
Chair: Tom Keller, IBM

Adapting the SPEC2000 Benchmark Suite for Simulation Based Computer
Architecture Research,
AJ KleinOsowski, John Flynn, Nancy Meares, David J. Lilja,
University of Minnesota

L-RSIM: A Simulation Environment for I/O Intensive Workloads,
Lambert Schaelicke,
University of Utah

Using the BACH Trace Collection Mechanism to Characterize the SPEC2000
Benchmark Suite,
Niki C. Thornock and J. Kelly Flanagan,
Brigham Young University

Choosing Representative Slices of Program Execution for Microarchitecture
Simulations: A Preliminary Application to the Data Stream,
Thierry Lafage and Andre Seznec,
IRISA, France

3:15pm - 3:45pm

3:45pm - 5:30
Panel Discussion
"Are longer benchmarks more representative?"

How long should benchmarks be? Seconds? Minutes? Hours? Days? Many
SPEC benchmarks run for half an hour or so. All events of interest
in a CPU benchmark including cache misses and page faults take much
shorter than that.  How long do you need to characterize a system to
perform an effective characterization?


Pradip Bose	IBM Yorktown Heights
Jim Browne 	The University of Texas 
Al Davis        University of Utah and Intel
John McCalpin   IBM Austin
John Shen       Intel Corp.

Lizy John	The University of Texas (Moderator)