Publication Details
Hardware Acceleration in Computer Networks
The speed of network traffic processing is a crucial parameter for most devices and systems, because any packet drop can cause lower quality of network services, affect precise monitoring or disallow detection of security threats. General purpose processors are not able to process all data on high-speed network links. For 100\,Gbps links, every packet has to be processed in less than 5 ns. Network devices widely use hardware acceleration to speed up time-critical operations. Therefore, this thesis deals with five widely used time-critical operations together with corresponding hardware architectures, which are able to achieve wire speed 10, 40 or even 100 Gbps throughput. In particular, the thesis deals with packet parsing and header fields extraction, longest prefix matching (IP look-up), packet classification, pattern matching and deep packet inspection. All these operations are widely used in precise network monitoring systems, network security devices and also in the infrastructure of data centres. The thesis introduces how deep pipelines, perfect hashing and pipelined automata can help to achieve 100\,Gbps throughput and decrease hardware resources. A novel concept is introduced to accelerate deep packet inspection. The concept provides software flexibility together with high performance, because fast packet processing is controlled at the level of flows by software modules. Moreover, proposed concept and hardware architectures are used in hardware accelerated network security and monitoring devices, which has been transferred to successful commercial products and used to monitor and protect CESNET2 academic network.