Design Of NoC Performance Evaluation Platform

Network on Chip (NoC) is a new proposed solution to deal with global communication problem of complex System on Chip (SoC). Due to its wide research area and issues, NoC is facing with the huge design challenge. The research issues mainly include the problems about System Modeling, Communication Model Abstraction, Network Topology, Protocol, Arbitration, Performance analysis, layout and mapping of IP module in network, On-chip Network design automation, and the test and verification of the NoC system, etc. In an application specific NoC design, from the choosing the topology forms, confirming communication protocol, to designing of the basic component of NoC, the designers have to do a lot of choice and tradeoff for every step in the whole course of design. Due to the huge design space of the NoC, different design considerations will cause the distinct results. A sound evaluation method is needed to evaluate that whether the NoC design could fill the need of the specific system application. Therefore how to evaluate the performance of the NoC design become very significant.Based on the analysis of related research work in and out of country and through the analysis of the characteristic of NoC and borrow the idea from the traditional network, interconnection network, and bus based transaction processing, this paper, at first, describes the definition of the NoC performance metrics, and proposed both the definition and the theoretical formulas for network based and transaction based metrics respectively. These metrics mainly included the average latency, throughput, and link utilization and so on.Then based on the analysis of the traffic attribute of the basic traditional network and the application specific NoC, through the abstracting of the temporal and the spatial traffic attributes, we build a two dimensional temporal-spatial traffic model, construct a NoC traffic generator model which satisfies the Poisson and Self-similar distributions in temporal traffic distribution and satisfies Evenly distribution and locality degrade probability distribution in spatial distribution. Accordingly, the traffic generator model which supports four different traffic distributions was got.Using the hierarchical progressive abstraction modeling method, and combined with traffic attribute of NoC, this paper proposed a scalable performance evaluation platform frame for NoC which supports multi traffic distributions and the random network scale. Based on this frame, we design two configurable performance evaluation platforms which are used for network based performance evaluation and transaction based performance evaluation in NoC respectively.In the end, these performance evaluation platforms test the performance of a specific NoC instance which is also designed by our research group from the aspects of network based and transaction based. On the one hand, we use the performance curves to analyze the performance of this NoC design, on the other hand, we verify the effectiveness and correctness of this platform.