A cost model for distributed shared memory using competitive update

This paper presents a new "cost" analysis model for distributed shared memory (DSM) using a competitive update protocol. The cost metric of interest is the overhead of message passing necessary to implement DSM. This approach is based on segment model proposed previously (Kim et al., 1996, 1997). The input parameter for the cost analysis model is the probability density function of the number of remote updates in a segment. This distribution can quite accurately characterize many applications. The proposed model is validated by comparing analytical results obtained using the model to experimental results. The competitive update protocol for shared memory is defined using a parameter called the "update limit" (or threshold). Using the proposed model, we compute the optimal update limit for the competitive update protocol.