CRAY-1 instruction analysis: a comparison of two methods
In an effort to obtain data for workload characterization and performance evaluation studies, statistics were gathered on the frequencies of individual instructions in codes executing on the Cray-1. Two methods were used for the collection of data: (1) direct dynamic counting of executing instructions and (2) sampling of instructions at regular intervals during execution of the code. This paper describes the implementation of both methods and compares the results obtained. We conclude that because the Cray-1 is a vector processor, the sampling technique produces skewed results that misrepresent the workload being examined. Therefore, although it is easier to implement, sampling produces inaccurate results and is an invalid approach to performance evaluation studies on a vector computer. The counting of instructions provides a valuable profile of instruction frequencies and is a stable basis on which to build performance estimations.
- Research Organization:
- Los Alamos National Laboratory (LANL), Los Alamos, NM (United States)
- DOE Contract Number:
- W-7405-ENG-36
- OSTI ID:
- 5633829
- Report Number(s):
- LA-UR-83-1500; CONF-831208-1; ON: DE83012778
- Resource Relation:
- Conference: 14. computer measurement group conference, Washington, DC, US, 6 Dec 1983
- Country of Publication:
- United States
- Language:
- English
Similar Records
Evaluating computer program performance on the CRAY-1
An instruction issuing approach to enhancing performance in multiple functional unit processors