Performance Analysis Of Network On Chip Based On Stochastic Network Calculus

Performance analysis is one of the focuses of NoC. Network calculus, as today’s most promising analysis tools of network performance, also can be applied to NoC. As deterministic network calculus based on worst case analysis, the performance analysis results are usually obtained overly pessimistic, which leads to waste of the resources of NoC. Network performance analysis based on stochastic network calculus, can predict the actual performance situation more accurately. We focus on the NoC performance analysis problem to study the modeling method based on stochastic network calculus. The main work is as follows:1. We have proposed the conflict analysis model based on service expectation.Traditional conflict analysis model is based on flow paths to establish the conflict situation, ignoring the influence of packets in its flow, and causes analysis results and the actual performance has a certain gap. According to the flow of the Poisson arrival process, the possibility of conflict was calculated and stochastic network calculus knowledge was used to create end to end performance analysis model of each flow. Simulation results show that, compared with previous conflict model, the analysis accuracy has been increased at least 18%.2. We have put forward a global redundant TSV fault-tolerant model.The existing partial redundancy TSV fault-tolerant model in promote chip yield ability is limited, we need a more efficient alternative. Global model redundancy fault tolerant TSV proposed in this paper compared with partial redundancy fault tolerant TSV model, in a same redundant number, can promote a higher yield. Experiments show that, in a specific redundant number, the proposed model is compared to the partial fault-tolerant redundancy in terms of yield, the highest improve is 29 percent. Due to TSV failure rate of the probabilistic result in uncertainty of node services, in line with the stochastic network calculus modeling ideas. Therefore, using stochastic network calculus, global fault-tolerant node service model was established in this paper.