The tutorial will start with a broad overview about the fundamental issues to high availability and how HA for networked services can be achieved at the different layers. We will then focus on the approach covered by the Linux Virtual Server project, which combines many real servers into a single virtual service for TCP/UDP applications, and discuss the basics - how an incoming request is processed through the networking stack, scheduling strategies, dispatching methods and how to design the network around it, pointing out some of the pitfalls I and others have already run into. If time permits, some yet unresolved interesting design aspects will be discussed with the audience. The tutorial will be illustrated by configuration snippets from real running systems and pointers to useful software packages and other sources of information. Who Should Attend Network administrators trying to achieve more reliability and higher performance for their TCP/UDP based services. The target audience is familiar with the basic concepts of networking and sysadmin'ing, interested in a lively discussion and sharing their own experience and ideas with the others instead of just listening to a boring talk.

Lars Marowsky-Brée has spend a few years with an ISP startup, responsible for all aspects of network operations and learning about very interesting failure modes which can affect availability. Besides being empowered, he was also exposed to Linux at an early age, which he now made his profession, working for SuSE to work on Linux HA solutions. He is always happy to share his views of how things should be done with everyone who listens (or not).