This is from a group project I worked on quite a while ago. I predominantly worked on writing the analysis. Here's the abstract from our paper for a little background:
Using random sampling to filter a trace cache should provide similar results to filtering based on the "hot/cold" trace principle due to the fact that hot traces occur more often. Based primarily on the work of Behar, Mendelson, and Kolodny, we have implemented such a trace cache using the SimpleScalar tool set and analyzed the results with regard to performance. Despite some unexpected results, our data was still in line with that of the prior work and showed a slight improvement over not using a trace cache. We suggest further improvements that could be made upon the trace cache structure in light of the performance benefits and low cost of implementing a random sampling filter.