Network stack specialization for performance

Contemporary network stacks are masterpieces of general-ity, supporting a range of edge-node and middle-node func-tions. This generality comes at significant performance cost: current APIs, memory models, and implementations drasti-cally limit the effectiveness of increasingly powerful hard-ware. Generality has historically been required to allow in-dividual systems to perform many functions. However, as providers have scaled up services to support hundreds of mil-lions of users, they have transitioned toward many thousands (or even millions) of dedicated servers performing narrow ranges of functions. We argue that the overhead of gener-ality is now a key obstacle to effective scaling, making spe-cialization not only viable, but necessary. This paper presents Sandstorm, a clean-slate userspace network stack that exploits knowledge of web server seman-tics, improving throughput over current off-the-shelf designs while retaining use of conventional operating-system and programming frameworks. Based on Netmap, our novel ap-proach merges application and network-stack memory mod-els, aggressively amortizes stack-internal TCP costs based on application-layer knowledge, tightly couples with the NIC event model, and exploits low-latency hardware access. We compare our approach to the FreeBSD and Linux network stacks with nginx as the web server, demonstrating ~3.5x throughput improvement, while experiencing low CPU uti-lization, linear scaling on multicore systems, and saturating current NIC hardware