Towards a High-Performance Multi-core Architecture for Network Processing

: It is difficult to adapt high performance routers based on ASIC to ever-changing network protocols. In other hands, software routers with high flexibility and programmability achieve low throughput. Hardware implementations based on FPGA not only are programmable solutions but also can offer acceptable throughput. Also multi- core and many-core architectures outperform throughput of software solutions with parallel computing model. In this paper, we propose a high-performance multi-core architecture for network processing. In this architecture, Bloom filter as a new and popular application is designed that called Bloom filter core. In addition, exploiting inherent parallelism in the querying of packets using Bloom filter is a motivation to apply GPU as many-core architecture in our work. The core is implemented on FPGA and its achieved throughput is compared with the results of implementation on the GPU. Experimental results show that both the FPGA-based and the GPU- based Bloom filter outperform considerably CPU-based results. Based on these results the maximum speedup of GPU and FPGA implementations are improved ۲۷۴ and ۱۴۱ times in compared to CPU, respectively.

کلیدواژه ها:

Bloom Filter ، FPGA ، GPU ، Multi/Many-coreArchitecture ، Parallel Computing.