An analytical model of high performance superscalar-based multiprocessors

Several shared memory multiprocessor models using approximate Mean Value Analysis (MVA) have been developed and used to evaluate a number of system architectures. Since this time, the complexity of multiprocessor systems has increased as superscalar processors and latency reduction techniques are employed in these systems. We present an MVA multiprocessor performance model which incorporates these new features and in addition, increases the level of modeling detail to improve exibility and accuracy. We describe in detail extensions present in our model that allow us to analyze the impact of these new features. We then use the model to demonstrate some of the tradeoos involved in designing modern multiprocessors, including the impact of highly su-perscalar architectures on the scalability of multiprocessor systems. 1 Introduction An analytical modeling technique that has been frequently used to evaluate shared memory multiprocessors is approximate Mean Value Analysis (MVA))12]. In MVA, a set of equations that represent the mean response times and mean waiting times of various performance elements are derived using the mean values of various system parameters as model inputs. For example, a simple multiprocessor MVA model may include the mean response time of a cache miss as a response time equation, the mean bus waiting time as a waiting time equation, and the mean time between cache misses as a model input. Once constructed, these equations have circular dependencies and must be solved iteratively. The power of MVA modeling lies in its computation ee-ciency. Convergence is usually achieved within a second, independent of the number of processors and memories in the model. Many models of multiprocessor systems have been created using this technique and have been used to evaluate the performance of both research prototypee9, 13] and commerciall10] multiprocessors. The results obtained from approximate MVA models have been shownn5] to correlate well with those obtained from trace-driven simulation models. Since these models have been constructed, superscalar