The purpose of this visualization was to emphasize that there are two factors at play with cache line size:
The first is the principle of locality. Larger cache line sizes means that memory accesses in a program are more likely to be hits (e.g. imagine iterating over an array; larger cache lines mean that fewer cache lines can "cover" the entire array, so we get fewer cache misses). This factor can be seen in the fact that for Barnes-Hut/Radiosity/Ocean Sim, as cache line size gets larger, miss rate decreases
The second is false sharing. As seen on the prev. slides, larger cache lines cause more false sharing and thus more cache misses. This factor can be seen in Radix Sort.
The purpose of this visualization was to emphasize that there are two factors at play with cache line size:
The first is the principle of locality. Larger cache line sizes means that memory accesses in a program are more likely to be hits (e.g. imagine iterating over an array; larger cache lines mean that fewer cache lines can "cover" the entire array, so we get fewer cache misses). This factor can be seen in the fact that for Barnes-Hut/Radiosity/Ocean Sim, as cache line size gets larger, miss rate decreases
The second is false sharing. As seen on the prev. slides, larger cache lines cause more false sharing and thus more cache misses. This factor can be seen in Radix Sort.