Directory based cache coherency protocols can be used to build large scale, weakly ordered, shared memory multiprocessors. The salient feature of these protocols is that they are interconnection network independent, making them more scaleable than snoopy bus protocols. The major criticisms of previously defined directory protocols point to the size of memory heeded to store the directory and the amount of communication across the interconnection network required to maintain coherence. This thesis tries solving these problems by changing the entry format of the global table, altering the architecture of the global table, and developing new protocols. Some alternative directory entry formats are described, including a special entry format for implementing queueing semaphores. Evaluation of the various entry formats is done with probabilistic models of shared cache blocks and software simulation. A variable length global table organization is presented which can be used to reduce the size of the global table, regardless of the entry format. Its performance is analyzed using software simulation. A protocol which maintains a linked list of processors which have a particular block cached is presented. Several variations of this protocol induce less interconnection network traffic than traditional protocols.
Date of this Version