Can the Linux networking stack be used with very high speed applications?
summary
Experiments use two commodity Ryzen 9 Zen3 servers, VCU1525 FPGAs over PCIe Gen3 x16 and unmodified Linux kernels (5.11 → 5.13) to push line rates; a custom driver + FPGA trick injects payloads and avoids PCI limits, achieving 31.8 million pps and up to 782 Gbit single flow (previously 670 Gbit). Work includes a modified hyper3 to implement a TX zero‑copy API, heavy use of LRO and hugepages; results show it works but bottlenecks remain in socket APIs, buffer management, memory handling and priming the pipeline.
AI-written summary · about
chapters (6)
- 0:00Start
- 0:17Introduction
- 7:57Memory Management
- 18:58Congestion Control Algorithm
- 19:35Hardware Considerations - CPU
- 24:41Summary
related talks
14mHigh-throughput and Flexible Host Networking for Accelerated Computing
31mA Practical, Research‑Backed Introduction To Cloud‑Native Kernel‑Bypass Networking
31mHow to capture 100G Ethernet traffic at wire speed to local disk
47mThe World of 100G Networking
16mA High-Speed Stateful Packet Processing Approach for Tbps Programmable Switches
27mHigh speed load balancing from the Linux Kernel
17mRearchitecting the TCP Stack for I/O-Offloaded Content Delivery
30mZXDH DPU Adapter and It's Application