The Case for a Network Fast Path to the CPU.

For the past two decades, the communication channel between the NIC and CPU has largely remained the same---issuing memory requests across a slow PCIe peripheral interconnect. Today, with application service times and network fabric delays measuring hundreds of nanoseconds, the NIC--CPU interface can account for most of the overhead when programming modern warehouse-scale computers. In this paper, we tackle this issue head-on by proposing a design for a fast path between the NIC and CPU, called Lightning NIC (L-NIC), which deviates from the established norms of offloading computation onto the NIC (inflating latency), or using centralized dispatcher cores for packet scheduling (limiting throughput). L-NIC adds support for a fast path from the network to the core of the CPU by writing and reading packets directly to/from the CPU register file. This approach minimizes network IO latency, providing significant performance improvements over traditional NIC--CPU interfaces.