An Accurate Analytical Design Model for Multithreaded Network Processors

Abstract

Network processors are system-on-a-chip programmable devices with special architectural features that are optimized to perform packet-processing functions. With higher line speeds and more demanding functional requirements, design of network processors has turned into a complex task. Therefore, having an accurate design model is necessary and helpful in accelerating and simplifying the design process. In this paper, extending a prior analytical model, we propose a number of analytical expressions to consider multithreading and branch predictor parameters in the model, resulting in having a more accurate representation with more effective design parameters. The multithreading is a selectable option in design of network processors which depending on the application can be enabled or disabled. Using the proposed model we not only can find out how to optimally configure a network processor but also can understand if the multithreading helps to improve the performance. Through simulation we have also observed a significant fraction of dynamic power (3% to 5% of total power consumption) is consumed by the branch predictor access, therefore we have added the branch predictor parameters to the model to consider its impact on the performance. Validity of the proposed model has been verified as performance trends for both of the simulation and proposed model were similar and the average error was equal to only 0.04.