Design and Implementation of Low-Power Bloom Filters for Deep Packet Inspection

Network intrusion detection systems must be able to perform deep packet inspection at increasingly high throughput as both network speeds and potential threats to networks increase. Bloom filters have recently re-emerged as efficient architectures for pattern matching network data with Snort signatures. In this paper, we present design and implementation details for a low power pipelined Bloom filter architecture. The pipelined architecture is intended for deep packet inspection applications and exploits the virus free nature of the network traffic most of the time; however, it can be utilized for other applications which can benefit from its low-power aspects. The implementation results confirm that the pipelining technique can yield an order of magnitude power savings compared with a regular architecture. In addition, the effects of using different hashing families in the pipeline stages have been investigated for further improvements. Due to the Inter-hash function collisions, a new measure we introduce, we find that using lower-complexity hash functions in the first stage achieves further reduction in power.