Cache characterization surfaces and predicting workload miss rates

  • Authors:
  • E. S. Sorenson;J. K. Flanagan

  • Affiliations:
  • Dept. of Comput. Sci., Brigham Young Univ., Provo, UT, USA;Dept. of Comput. Sci., Brigham Young Univ., Provo, UT, USA

  • Venue:
  • WWC '01 Proceedings of the Workload Characterization, 2001. WWC-4. 2001 IEEE International Workshop
  • Year:
  • 2001

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper, we use locality surfaces to predict cache miss rates. To do this, we introduce two new surfaces. The miss surface characterizes how a trace is filtered by a particular cache in terms of locality. A cache characterization surface helps us examine caches in terms of what stride/delay relationships are likely to cause misses in the cache. The cache characterization surface is independent of any workload. We use these surfaces to quantitatively predict cache miss rates with some degree of accuracy.