A low-cost classification method for high-speed network processors

Network processors utilize a special unit for the classification of the received packets according to their characteristics. The classification of packets is currently the most processor intensive portion of network processing which has to be performed at wire speed. Currently, network processors use Content Addressable Memories (CAM), for classification of thousands of flow identifiers in ATM and IP, with very long keys. This paper proposes an alternative method for searching wide keys, using RAM chips as storage medium instead of the expensive and complex CAMs. The technique relies on the open-addressing hashing methodology to provide high speed look-ups close to the CAM's performance. It also handles efficiently the limitations imposed by the hashing algorithms by selecting properly the parameters of the system.