
576 CHAPTER 8 Directory-Based Cache Coherence
Applying the Framework to the Application Case Studies
Let us now look briefly at each of the applications in Figure 8.9 to interpret the
results in light of these four sharing patterns and to understand how the size distri-
butions might scale.
In the LU factorization program, when a block is written it has previously been
read only by the same processor that is doing the writing (the process to which it is
assigned). This means that no other processor should have a cached copy, and zero
invalidations should be sent. Once it is written, it is read by several other processes
and no longer writte