Paper
Elizabeth S. Sorenson. Using Locality To Predict Cache Performance. Master's thesis, Brigham
Young University, 2001.
Abstract
Researchers have developed many different methods for evaluating locality. However there
exists no standard method of evaluation that incorporates all aspects of temporal and
spatial locality. In this thesis, we introduce an advanced form of locality surface
that incorporates all aspects of temporal and spatial locality into one
three-dimensional surface. Examining a locality surface for a particular workload
gives significant information about sequential, temporal, and loop constructs. This
gives researchers a new avenue for characterizing workloads and examining locality
quantitatively. We also show how these surfaces are useful for qualitatively
predicting cache performance.
In addition, we present a method for creating cache characterization surfaces. These
surfaces help designers to visualize what types of locality events perform well in a
given cache. Locality surfaces and cache characterization surfaces may be multiplied
together to quantitatively predict cache performance for a given workload and a given
cache.