Hardware acceleration for power efficient deep packet inspection

The rapid growth of the Internet leads to a massive spread of malicious attacks like viruses and malwares, making the safety of online activity a major concern. The use of Network Intrusion Detection Systems (NIDS) is an effective method to safeguard the Internet. One key procedure in NIDS is Deep Packet Inspection (DPI). DPI can examine the contents of a packet and take actions on the packets based on predefined rules. In this thesis, DPI is mainly discussed in the context of security applications. However, DPI can also be used for bandwidth management and network surveillance.\ud \ud DPI inspects the whole packet payload, and due to this and the complexity of the inspection rules, DPI algorithms consume significant amounts of resources including time, memory and energy. The aim of this thesis is to design hardware accelerated methods for memory and energy efficient high-speed DPI.\ud \ud The patterns in packet payloads, especially complex patterns, can be efficiently represented by regular expressions, which can be translated by the use of Deterministic Finite Automata (DFA). DFA algorithms are fast but consume very large amounts of memory with certain kinds of regular expressions. In this thesis, memory efficient algorithms are proposed based on the transition compressions of the DFAs.\ud \ud In this work, Bloom filters are used to implement DPI on an FPGA for hardware acceleration with the design of a parallel architecture. Furthermore, devoted at a balance of power and performance, an energy efficient adaptive Bloom filter is designed with the capability of adjusting the number of active hash functions according to current workload. In addition, a method is given for implementation on both two-stage and multi-stage platforms. Nevertheless, false positive rates still prevents the Bloom filter from extensive utilization; a cache-based counting Bloom filter is presented in this work to get rid of the false positives for fast and precise matching.\ud \ud Finally, in future work, in order to estimate the effect of power savings, models will be built for routers and DPI, which will also analyze the latency impact of dynamic frequency adaption to current traffic. Besides, a low power DPI system will be designed with a single or multiple DPI engines. Results and evaluation of the low power DPI model and system will be produced in future