Communications of the ACM
Technical Perspective: hXDP: Light and Efficient Packet Processing Offload
Credit: Getty Images
For a very long time, the network was considered the slow part of a system. Even today, our CPU is running at clock speeds of several GHz, while our home network is likely to still run at tens to hundreds of megabits per second. This, however, is far from being the case in the wired systems world. Network switches process tens of terabits every second, and many network interface cards (NICs) have one or more 100G ports. Suddenly, the CPU is slow in comparison with the rate of data arriving from the network.
Unsurprisingly, solutions to bridging the performance gap between the CPU and the network are either software or hardware based. In the following paper, the authors offer an interesting solution: taking a software-based solution (Linux's eXpress Data Path—XDP) and offloading it to the hardware (Field Programmable Gate Arrays—FPGA). In that, it achieves the best of both worlds: ease of adoption and use, combined with performance benefits.
No entries found