Affiliations: [a] Department of Computer Engineering and Information Technology, Razi University, Kermanshah, Iran | [b] Department of Information Network, Information Technology, University of Babylon, Babylon, Iraq
Abstract: Packet classification is a network kernel function that has been widely investigated over the past decade. New networking paradigms, such as software-defined networking and server virtualization, have led to renewed interest in packet classification and its upgrade from classical five-field to many-field classification. With the increasing size of the rule sets and demands for higher throughput, performing many-field packet classification at wire-speed has become challenging. In this paper, we propose an approach to classification by integrating a probabilistic data structure called the Cuckoo filter for approximate membership queries into an R-tree data structure for high-speed, many-field packet classification. Experimental results show that the proposed classifier obtains high throughput of up to 1.5 M packets per second, and requires little memory to support large rule sets (up to 1 million rules).
