Network classification algorithms and packet switching

The modern network edge must perform a wide spectrum of tasks: deep packet inspection, firewalling, VPN services, intrusion detection and many others. These tasks may have radically different computational requirements but traditional architectures of network processors are not suitable for processing packets with heterogeneous processing requirements. Algorithms for these models have only started to be developed, and we plan to obtain new results in this field.

Efficient implementation of packet classification is one of the main problems solved in network services. However, existing traditional approaches, both software-based and based on TCAM, have conceptual limitations and cannot always both achieve fast execution of queries (lookups) and save memory. We plan to develop new approaches to the optimization of classifiers located in TCAM memory and develop a new approach to classifier optimization which will greatly extend the capabilities of query parallelization in network classification.

Preliminary results of this project have been presented at a leading networking conference SIGCOMM2014.