(Georgia Institute of Technology, 2001)
Merugu, Shashidhar; Gummalla, Ajay Chandra V.; Sala, Dolors; Zegura, Ellen W.
Multi-dimensional packet classification is increasingly important for
applications ranging from fire-walls to traffic accounting. Fast link
speeds, the desire to classify with fine granularity, and the need for
agility in a dynamic environment all pose significant challenges for packet
classification. We propose an approach that is capable of handling a
changing set of classification rules that span multiple
fields. Our approach is based on extracting a relatively small set of bits
that uniquely identify the packets satisfying each rule. Changes to the rule
set are handled in-line via a fast update mode that adds to the rule table,
while a background process performs reoptimization of the full rule table
less frequently. The classification process can be efficiently implemented
using pipelined hardware and supports
high packet arrival rate.