cppvalleyC++ · Systems · Low Latency · HFT
HFT Core Systems

HFT CORE SYSTEMS

Socket tuning and busy polling

Lesson 37 of 96Phase 04: Network datapaths
Video lesson coming soon

The written objectives and hands-on lab are available now.

cppvalley on YouTube ↗

Learning objectives

What you should understand

  1. Size socket buffers from traffic and burst assumptions
  2. Compare blocking, polling and busy-poll behavior
  3. Measure CPU cost alongside latency benefit

Lab

Hands-on exercise

Sweep receive-buffer and busy-poll settings under bursts.

Deliverable

Evidence of completion

A tuning table with loss, CPU and percentile trade-offs.

Prerequisite
EP 36

Phase 04

Network datapaths

Full curriculum
33Ethernet frames and the real wire budget34IPv4/IPv6, UDP, checksums, and fragmentation35Multicast joins, IGMP, and feed topology36The Linux receive path: DMA to userspace37Socket tuning and busy polling38RSS, flow steering, RPS/RFS, and XPS39AF_PACKET, PACKET_MMAP, and AF_XDP40DPDK architecture and memory ownership41AMD Solarflare ef_vi and raw layer-2 access42PCIe, DMA, IOMMU, DDIO, and NIC locality43Hardware and software packet timestamps44Phase gate: receiver bake-off