Efficient multi-field packet classification for QoS purposes

Abstract

Mechanisms for service differentiation in datagram networks, such as the Internet, rely on packet classification in routers to provide appropriate service. Classification involves matching multiple packet header fields against a possibly large set of filters identifying the different service classes. In this paper, we describe a packet classifier based on tries and binomial trees and we investigate its scaling properties in three QoS scenarios that are likely to occur in the Internet. One scenario is based on integrated services and RSVP and the other two are based on differentiated services. By performing a series of tests, we characterize the processing and memory requirements for a software implementation of our classifier. Evaluation is done using real data sets taken from two existing high-speed networks. Results from the IntServ/RSVP tests on a Pentium 200 MHz show that it takes about 10.5 μs per packet and requires 2000 KBytes of memory to classify among 11000 entries. Classification for a virtual leased line service based on DiffServ with the same number of entries takes about 9 μs per packet and uses less than 250 KBytes of memory. With an average packet size of 2000 bits, our classifier can manage data rates of about 200 Mbit/s on a 200 MHz Pentium. We conclude that multi-field classification is feasible in software and that high-performance classifiers can run on low-cost hardware.