In this paper, we introduce new analytical models for predicting the performance of parallel applications under various cache coherence protocol assumptions. The purpose of these models is to determine which protocols are to be used for which data blocks, and, in the case of dynamic protocols, also to determine when to change protocols. Although we focus on tightly-coupled multiprocessor systems, similar models can be derived for loosely-coupled distributed systems, such as networks of workstations. Our models are unique in that they lie between a large body of theoretical models that assume independence and a uniform distribution of memory accesses across processors, and a large body of address-trace oriented models that assume the availability of a precise characterization of interleaving behavior of memory accesses. The former are not very realistic, and the latter are not suitable for compile-time and run-time usage. In contrast, our models enable us to choose different input parameters depending on how the models will be used and depending on the needed accuracy in performance prediction. We present the models and show how the required parameters can be obtained. We assess the accuracy of our models on 15 parallel applications. For these applications, our most complete model predicts performance within a 10 percent margin when compared to a simulation of a sequentially consistent multiprocessor system. As part of this study, we also show the potential advantage of using dynamic hybrid protocols.

A model for predicting relative performance in distributed shared memory systems is presented. The model is intended for use with systems that feature adaptive consistency schemes. Based on estimated read/write access patterns, the model determines the average steady-state communication costs in the system, which serve as a performance indicator. The model is applied to data items which may be either compact data objects or objects consisting of smaller data elements. It is shown that the best results are attained when protocol selection is done at the level of data elements.>

The proposed distributed shared memory model is based on a data replication scheme that provides an environment for a collection of processes that interact to solve a parallel programming problem. In the implementation of the scheme the authors suppose that the replicas of the shared data are present at each node and that an appropriate coherence protocol for maintaining the consistency among the replicas is applied. The performance of the distributed computation is very sensitive to the data-access behavior of the application and to the applied coherence protocol. Communication cost is regarded as an appropriate performance measure. Therefore, the authors first introduce a model characterizing the computation behavior with five workload parameters. Second, they formally describe the coherence protocols as cooperating state machines in order to evaluate their communication costs as functions of workload parameters.>