Internet Packet Classification with Associative Memories at SiPS’16

Professor Warren Gross recently presented our new work on internet packet classification hardware:

Sparse-clustered network with selective decoding for internet traffic classification

Led by McGill graduate Scott Dickson Dagondon (now with Motorleaf), this work employs sparsely clustered networks (SCN) to learn the relationship between packet headers and application classes to facilitate QoS, identify malicious traffic, or detect network intrusion. Taking advantage of parallel lookup in the associative structure of the SCN, our implementation is orders of magnitude faster than the state of the art.  This work was presented at the International Workshop on Signal Processing Systems (SiPS).