Prior TCAM-based classifier compression schemes [Draves et al(1999)Draves, King, Venkatachary, and Zill,Suri et al(2003)Suri, Sandholm, and Warkhede,Applegate et al(2007)Applegate, Calinescu, Johnson, Karloff, Ligett, and Wang, Dong et al(2006)Dong, Banerjee, Wang, Agrawal, and Shukla, Liu and Gouda(to appear), Meiners et al(2007)Meiners, Liu, and Torng] suffer from one fundamental limitation: they only produce prefix classifiers, which means they all miss some opportunities for compression. A prefix classifier is a classifier in which every rule is a prefix rule. In a prefix rule, each field is specified as a prefix bit string (e.g.,, 01**) where *s all appear at the end. In a ternary rule, each field is a ternary bit string (e.g.,, 0**1) where * can appear at any position. Every prefix rule is a ternary rule, but not vice versa. Because all previous compression schemes can only produce prefix rules, they miss the compression opportunities created by non-prefix ternary rules.
Unable to display preview. Download preview PDF.