high performance computing on graphics processing units: hgpu.org

As the fiber propagation velocity grows and the routing scale expands, IP lookup speed becomes the major bottleneck of high-performance network. Its efficiency directly determines the throughput of the entire routing channel. Recently, Graphics Processing Units (GPUs), highly parallel, flexibility for program and low price, is widely adopted in different areas including software router. In this paper, we propose an architecture named GPU-Based Parallel Multiple Hash IP Lookup/Update Architecture (GPMHIA) to perform high-performance IP address lookup/update. We design a high-speed IP address lookup/update architecture based on IPv4, and this architecture can be easily extended to IPv6. Meanwh-ile, parallel multiple hash functions have been used for accelerating the matching speed and we propose an Adaptive Optimal XOR Hash (AOXH) to construct hash functions. Simulation results using 5 real IPv4 forwarding tables present that GPMHIA is a high-performance IP forwarding architecture.