Software techniques for negating skid and approximating cache miss measurements

Data centric analysis using direct measurements has been established as a successful performance analysis technique. Information gathered with this technique can map cache misses to program variables. These mappings can then be used to address data locality problems and other issues. Existing approaches rely on special hardware support which is needed to negate a ‘skid’ factor. Our approach is viable when the special hardware support is not present, but where skid is still an issue. Prior methods also rely on maintaining runtime information about memory allocation addresses for variables, which may lead to program perturbation. Our approach uses software analysis to eliminate the need for maintaining allocation and free records. We show that by using heuristics our technique can attribute cache misses to program variables while maintaining the approximate rank-order found by using traditional techniques. We also show that there exists a high correlation between the misses attributed by our approximation and the misses assigned by examining direct measurements.

► Software heuristics can be used to approximate cache miss measurement.
► Attributing cache misses to variables takes locality into account.
► Skid is negated through software by using adjustments based on probability.
► Approximations maintain roughly same rank order as direct measurements.
► Approximation techniques run on architectures where skid negation not present.