Administering Very High Volume Internet Services

Providing WWW or FTP service on a small scale is already a well-solved problem. Scaling this to work at a site that accepts millions of connections per day, however, can easily push multiple machines and networks to the bleeding edge. In this paper, we give concrete configuration techniques that have helped us get the best possible performance out of server resources. Our analysis is mostly centered on WWW service, but much of the information applies equally well to FTP service. Additionally we discuss some of the tools that we use for day-to-day management. We don't have a lot of specific statistics about exactly how much each configuration change helped us. Rather, this paper represents our many iterations through the watch the load increase; see various failures; fix what's broken loop. The intent is to help the reader configure a high-performance, manageable server from the start, and then to supply ideas about what to look for when it becomes overloaded.