An effective network processor design framework
Liam Noonan, Colin Flanagan · 2006
In this paper we present a framework for design space exploration of a network processor, that incorporates parameterisation, power and cost analysis. This method utilises multi-objective evolutionary algorithms and object oriented analysis and design. Using this approach an engineer specifies certain hard and soft performance requirements for a multi-processor system, and allows it to be generated automatically by competitive evolution/optimisation, thus obviating the need for detailed design. To make the proposal concrete, we use the Intel IXP1200 network processor as a baseline complex system design and show how various improvements can be make to this architecture by evolutionary/competitive design. Various approaches to multi-objective optimisation (Darwin, Lamarck Baldwin, etc.) are compared and contrasted in their ability to generate architectures meeting various constraints. We also present an assessment of a proposed architecture with reference to four different packet processing roles. The merits of an "island clocking" scheme versus a "common clocking" scheme are also discussed. Our paper highlights the flexibility that this framework bestows on the designer, along with the potential to achieve cost savings and performance improvement.