Explore XDP bulk packet processing techniques in this Linux Plumbers Conference talk. Dive into the potential performance improvements of batching in networking, particularly for XDP processing pipelines. Learn about a proof-of-concept implementation that yielded a 20% performance boost for packet dropping with header memory access. Discuss implementation details, including BPF verifier logic, generic NAPI clean Rx interrupt routines, and the potential benefits of batching XDP actions. Gain insights into optimizing network performance through efficient instruction cache utilization and reduced indirect function calls. Engage in a technical discussion on the future of XDP bulk processing and its implications for driver development and network stack optimization.
Overview
Syllabus
Introduction
Results
Agenda
Performance improvements
Two approaches
Driver side
EVP
BBF
EBPF registers
EBPF program layout
Responsibility
Subtract
Epilogue
Summary
QA for bulk processing
QA questions
Taught by
Linux Plumbers Conference