Development of improved packet classification system in software defined network

Abstract

The software-defined network is a new paradigm to simplify the network management by integrating network control with a centralized control platform. In a control platform, the network devices execute various applications to perform different tasks like packet forwarding, load balancing, firewall, and monitoring. The main objective of the research is to perform packet forwarding function in the network system and it depends on the speed of the network. The routers process the packets using a single field as well as a multi-field of IPv4 packet header information. The single field packet classification has been achieved by applying the longest prefix matching and range matching techniques on the destination IP address of an incoming packet to find the next-hop address over the autonomous system dataset. Multi-field packet classification has been produced by applying supervised Machine Learning (ML) algorithms over the firewall dataset. These algorithms examine the packet header information and classify the packets based on the predefined actions (permit/deny). In single field packet classification, the routers have the problems of storage space and search time complexity, due to the increase of entries in the forwarding table. To overcome these problems, high-speed IP lookup algorithms namely, multi-bit trie and parallel Bloom filters are used to improve the space and search time complexity. An improved Huffman encoding with a leaf pushing compression algorithm has been applied to compress the IP prefixes and an average of 40.88% of space reduction has been obtained in the forwarding table. When compared to Huffman, adaptive Huffman, and canonical Huffman compression techniques, the proposed compression algorithm has achieved an improvement of 31.9%, 35.36% and 23.64% space reduction, respectively. For searching the nexthop address in the compressed table, a Multi-bittrie is used. In this trie, several bits are compared at a time based on the stride value and it shows 61% of search time improvement than the binary