Hit ratio has been extensively used in the study of single-level caches, where higher hit ratios generally imply better performance. In multi-level scenarios, however, the benefit of a hit varies significantly depending on the cache level at which it occurs, making hit ratio a misleading performance metric.
In this paper we use the average response time as the key performance metric. In practice, different algorithms result in different amounts of inter-cache traffic, and in limited bandwidth scenarios, the observed average response time depends more on the inter-cache bandwidth usage than on the number or location of hits. Hence, we measure both the inter-cache bandwidth usage and the average response time (assuming unlimited bandwidth). The actual bandwidth limit depends on the hardware and the number and intensity of other applications using the same network fabric. We provide measurements for some limited bandwidth cases as well.
In our experimental results, error bars are not shown since the average response times over separate runs was within 
%, even for the adaptive algorithms.