A Study & Comparative Evaluation Of Packet Classification Algorithms
In a field of incessantly expanding Internet connectivity, along with intensifying computer security threats,
protection-conscious network applications technology is growing in prominence. Packet classifiers are exhaustively engaged
for various network applications in distinctive types of network devices such as Firewalls and Router, besides many more.
Comprehending the actual accomplishment of endorsed packet classifiers is essential for both algorithm designers as well as
clients. Nonetheless, this is sometimes tedious to undertake. Each pioneering algorithm put out is evaluated from different
views and is initiated on diverse theories. Bereft of a common establishment, it is fundamentally impossible to compare
different algorithms in a direct fashion. This, in a way, helps the system facilitators to smoothly select the most appropriate
algorithm for their actual applications.
Selecting a feeble algorithm for an application can lead to major expenses, particularly in the case of packet
classification in network routers, as packet classification is essentially a tough problem and all contemporary algorithms are
created on explicit heuristics and filter set features. The performance of the packet classification subsystem is of paramount
importance for the collective success of the network routers.
In this study, we have conducted an advanced investigation of the existing algorithms to offer a comparative evaluation
of a number of known classification algorithms that have been deliberated for both software and hardware applications. We
have also explained our previously proposed DimCut packet classification algorithm, and associated it with the TSS, BV,
HiCuts, HyperCuts and Woo packet classification algorithms with the comparative evaluation analysis. This comparison has
been implemented on applications created on the matching principles and design selections from diverse sources.
Performance measurements have been obtained by feeding the implemented classifiers with a large number of random Rules
and Packets in an identical test scenario.