Abstract:

The flourish of new applications, such as network function virtualization (NFV), has brought higher requirements on high performance and energy efficient Key-Value (KV) lookups. Traditionally, KV operations can be accelerated by software-based HASH tables or dedicated TCAM chips. Software-based solutions are cost efficient but can lead to much worse performance with the increase of data collisions. TCAM-based solutions, on the other hand, have sound performance but suffer high additional system cost and power consumption. Recently, FPGA-based heterogeneous computing becomes more and more popular, so it is quite reasonable to exploit the provided FPGA resources to accelerate the software-based KV operations. To this end, we discuss how to implement high scalable and energy efficient TCAM logics with RAMs on FPGA in this paper. Compared with the traditional TCAM architecture, the proposed FPGA-based TCAM is highly scalable and enables dynamical configurability of the range of lookups so that the power consumption can be reduced significantly. To validate the proposed architecture, we implemented it on Xilinx Virtex-7 FPGA. Experimental results show that the throughput can be as high as 234 Mpps and the latency as low as 25.56ns. Compared with traditional software-based solutions, the throughput is improved by 780 times  and the latency is improved by 240 times.

Key words: network function virtualization, Key-Value lookup, TCAM, FPGA