Limited Directory Scheme (cont)
Assume we are keeping K pointers, and that K processors currently have copies of the data
Another processor requests the data - now have K+1 processors requiring the data
- No more free pointers, will have to tell one of the processors to invalidate their cache entry before the K+1 processor can read the data (data eviction).
- Processor to be selected for cache invalidation can be chosen on a psuedo random basis, or using a more complex scheme (least recently used)