Receive CPU Selection Framework: Cross-Layer Optimization In Network Stack to Improve Server Scalability.

This paper explores what normal web requests look like in the Linux kernel network stack's perspective, summarizes the available hardware and software optimization techniques, and then presents a method of improving web server applications' performance and scalability by introducing the Receive CPU Selection framework. This framework bridges the information gap between network stack layers and makes it possible to process a web request always within one CPU core from the NIC (network interface card) to the application process, which avoid cache misses and CPU core switches. Experiments show that with this optimization, server applications perform and scale better. Unlike previous optimization, server applications can gain noticeable performance improvement while their source codes remain unmodified, which is a significant requirement in the production environment.