A Fast CRC Update Implementation

In networking environments, the cyclic redundancy check (CRC) is widely utilized to determine whether errors have been introduced during transmissions over physical links. In this paper, we focus on the CRC calculation that is performed during the routing of the Ethernet packets by encapsulating the packets into Ethernet frames, adding a frame header and adding a frame trailer. The CRC code is placed within the frame trailer. In our investigation, we observed that only specific fields located at the beginning of a frame are changed when it is passing through interconnecting devices. Based on this observation, we propose a novel implementation of the CRC update and we present the proof of correctness of our implementation. Our approach calculates the intermediate results of the changed fields based on the parallel CRC calculation and performs a single step update afterwards. Consequently, the number of cycles utilized to recalculate the CRC codes is dramatically reduced. Furthermore, an estimation on the maximum throughput is made based on synthesis results of our implementation and under the assumption that the CRC operation is the only bottleneck. In this case, we have estimated that the theoretical throughput can reach about 56 Gbps assuming realistic frame size distributions. Keywords— Cyclic redundancy check, parallel CRC calculation, CRC update.