KDB: a fast update and high speed packet classifier in SDN

Abstract

Packet classification is a fundamental function to support several services of software defined networking (SDN). Increasing complexity of the flow tables in SDN leads to challenges for packet classification on update and classification time. In this paper, we propose KDB, a hybrid decision tree classifier, to achieve fast update and high speed packet classification. Experimental results show that KDB is faster in update time compared with SmartSplit and PartitionSort, two state-of-the-art decision tree classifiers, and achieves comparable classification time. Compared with Tuple Search Space (TSS), a classifier used in Open vSwitch, KDB is faster in classification time and achieves comparable update time.