Network Stack Specialization For Performance外文学习材料.pdfVIP

Network Stack Specialization for Performance Ilias Marinos Robert N.M. Watson Mark Handley University of Cambridge University of Cambridge University College London ABSTRACT individual systems had to perform multiple diverse func- Contemporary network stacks are masterpieces of general- tions. In the last decade, the advent of cloud computing ity, supporting a range of edge-node and middle-node func- and the ubiquity of networking has changed this model; to- tions. This generality comes at significant performance cost: day big content providers serve hundreds of millions of cus- current APIs, memory models, and implementations drasti- tomers. To scale their systems, they are forced to employ cally limit the effectiveness of increasingly powerful hard- many thousands of servers, with each providing only a sin- ware. Generality has historically been required to allow in- gle network service. Yet most content is still served with dividual systems to perform many functions. However, as conventional general-purpose network stacks. providers have scaled up services to support hundreds of mil- These general-purpose stacks have not stood still, but to- lions of users, they have transitioned toward many thousands day’s stacks are the result of numerous incremental updates (or even millions) of dedicated servers performing narrow on top of codebases that were originally developed in the ranges of functions. We argue that the overhead of gener- early 90’s. Arguably these network stacks have proved to be ality is now a key obstacle to effe