Predictive processing architecture extension for network processors

Abstract

(~-(~-r~·\ Objective of the proposed packet processing concept is to reduce the latency which result from network layer hierarchies. It is based on two components: Protocol Stack Prediction and Speculative Packet Processing. The decision how to handle and forward a packet is made depending on the information which is contained in the packet header. The header of the data-link layer, the network layer and the transport layer is hierarchically arranged. I.e. the type of a higher layer of the ISO/OSI reference model of communication is derived from the header of the lower layer. To circumvent a serial extraction, protocol stackprediction predicts the protocol stack of the next packet. The scheme is depicted in Fig. 1. met due to architectuallacks [6]. This results from inherent control data dependencies in the multi-layer protocol stack. In [8], the capability of a new NP processing architecture based on data prediction and speculative protocol processing has been demonstrated. While there a fundamentally new architecture was designed, this paper decribes an efficient mapping of the approach to an already existing, commercially available NP. Thus, the benefit of speculative protocol processing can be obtained by an adaptation of software mapping and modified processing control. The remainder of this paper is organized as follows. Section II introduces the speculative protocol-processing model. Its applicability to Intel IXP2400 is shown in section III followed by a quantitative comparison. Finally, the conclusion in section VI highlights the benefit of the proposed concept.

DOI: 10.1109/ICECS.2005.4633587

Cite this paper

@inproceedings{Foag2005PredictivePA, title={Predictive processing architecture extension for network processors}, author={J{\"{u}rgen Foag and Thomas Wild}, booktitle={ICECS}, year={2005} }