Modeling RTOS Components for instruction cache hit rate estimation

Abstract

RTOS components have a direct impact on the instruction cache performance - an aspect that has been reported as a bottleneck in several studies. Therefore, there exists a need for insight into the instruction cache hit rates for various RTOS components at design time for efficient design space exploration. In this paper, we propose a technique to model RTOS components for instruction cache hit rate estimation. Our methods rely on rapid generation of hit rate values for different cache sizes for the RTOS components. We then fit the generated instruction cache hit rates using multivariate regression schemes to account for all parameters influencing the hit rate. The estimation techniques were tested for a large range of cache sizes and numerous parametric as well as non-parametric RTOS components. Comparison with hit rates generated by a cache simulator shows that these models can accurately estimate the hit rates with the mean difference in hit rates ranging from 0.00 to 0.05. The proposed technique also offers significant speed-up as compared to other modeling approaches because it does not require exhaustive cache hierarchy simulation for building the models.