Medium
Kernel-bypass packet I/O with DPDK and AF_XDP
A market-data ingest system needs sub-microsecond tick-to-trade latency and switches its NIC I/O from standard kernel sockets to a kernel-bypass framework such as DPDK or AF_XDP. What is the core mechanism this enables, and which specific costs of the ordinary recv() path does it eliminate?
23 of 45 in Networking