Parallelizing Deep Packet Inspection on GPU

Network security has been very important for the software industry. Deep packet inspection (DPI) is one of the widely used approaches to enforce network security. Due to the high volume of network traffic, it is challenging to provide high performance for DPI in real time. Thus, a few studies used various accelerators for DPI such as FPGAs, ASICs and GPUs for better performance. Though existing studies successfully demonstrated outperformance of DPI on these accelerators, not many of them compared the performance of various signature algorithms, which are core function of DPI. In this study, we design and evaluate a GPU-based DPI framework that includes packet header checking and payload inspection. We implemented header checking in CUDA by using two types of parallelism; block-level and thread-level. Payload inspection was developed by using three widely used signature matching algorithms. We also provide performance comparison of the three signature matching algorithms and the two types of parallelism. Our evaluation demonstrates that GPU version DPI outperforms sequential CPU version significantly.