Difference between capacity miss and conflict miss?

Question

For example, Cache contains only 2 sets, SET 1 and SET 2. Problem: If data A maps to SET 1 and it doesn't exist in SET 1 while SET 1 is fully occupied and SET 2 is *not* fully occupied, I know this case is called conflict miss. But what should it be called if the same case but entire cache is fully occupied? Is it capacity miss or conflict miss? Once the cache is fully occupied, will a conflict miss ever happen?

Answer 1

Of the traditional 3 C's of cache misses, capacity and conflict are the hardest to understand. My favorite definition for a conflict miss comes from some early literature on caches.

"Conflict misses are misses that would not occur if the cache were fully associative with LRU replacement" - Norman Jouppi

The big idea behind that is if you increase the associativity to the maximum, and that resolves the miss, then it must be because you did not have enough associativity. Capacity misses on the other hand occur because the cache is not large enough.

So in your particular case, you need a bit more information about the cache. You specified two sets, but you didn't specify whether it's a directly mapped or 2-way set associative cache.

As to whether a conflict miss can ever occur in a cache that is fully occupied, sure, it absolutely can. All you need is to read a few elements that map onto the same set, but are less than the size of your cache.