The key idea here is to illustrate the disadvantage of sparse directories when compared to bit vector directories. Although sparse directories do save space, a write miss in a bit vector directory allows for invalidation messages to be sent in parallel, resulting in a faster acknowledgment and overall write speed than the sparse directory format where traversing the linked list of sharers to invalidate each one is necessary. I would imagine this is more of a problem in computers with an extremely large number of processors, where the linked list of sharers for a single cache line could be very long. If, however, we are considering a quad-core machine, then the write latency probably would not be as significant.
The key idea here is to illustrate the disadvantage of sparse directories when compared to bit vector directories. Although sparse directories do save space, a write miss in a bit vector directory allows for invalidation messages to be sent in parallel, resulting in a faster acknowledgment and overall write speed than the sparse directory format where traversing the linked list of sharers to invalidate each one is necessary. I would imagine this is more of a problem in computers with an extremely large number of processors, where the linked list of sharers for a single cache line could be very long. If, however, we are considering a quad-core machine, then the write latency probably would not be as significant.
This comment was marked helpful 0 times.