Analytical Performance Analysis of Network-Processor-Based Application Designs

摘要

Network processors (NP) are designed to provide both performance and flexibility through parallel and programmable architecture, making them superior to general-purpose processors on performance and to hardware-based solutions on flexibility. But NPs also introduce new challenges. It is important to study the limitations of NP architectures so that one can take full advantage of NP resources to achieve the required performance for a given application. It is therefore desirable to develop a general framework for analyzing performance of NP-based applications. This paper presents an analytical method for solving this problem. In particular, we devise a queuing network to model NP resources and application work flows. We then use queuing theory and operational analysis to obtain performance metrics on throughput and response time, among other things, at the component level as well as at the system level. We apply our performance model to SpliceNP, a TCP Splicing implementation of content-aware switches on network processors presented in [10], and show that the analytical results using our models match the experimental results from actual implementation.