Lightweight Autonomic Network Architecture

During the last decades the Internet architecture matured and gained more and more popularity. Networking devices of all kind ranging from high-end servers to personal computers, tablet devices, mobile phones or even embedded sensor networks make use of it. The Internet became ubiquitous for us. Even though we witness such a diversity of devices, the progress of having an efficient interconnection between all of them only advances in small steps. No matter if we consider a server, a mobile phone or a resource-constrained temperature sensor on an embedded device from this large spectrum, with high probability, all of them have the same, one-size-fits-all TCP/IP protocol stack, although there are more appropriate communication mechanisms than that. Recent research tries to challenge this stagnation by introducing dynamic adaption into computing systems. For instance, the EPiCS project aims at making devices more self-aware, so that they can dynamically adapt to their current needs. Imagine such a temperature sensor that only has a very minimal protocol stack, hence it has only protocols active, it is actually using in order to save resources. If such a sensor detects that it is loosing too much battery power, it could automatically switch to a less reliable but less power-intense protocol. Therefore it needs to dynamically adapt its communication stack during runtime. For the EPiCS networking part, the Autonomic Network Architecture (ANA) lays the foundation of such an architecture. This master’s thesis is situated in the ANA context. The first ANA prototype implementation has shown that such a dynamic adaption of the protocol stack is possible, but it heavily suffers from performance issues. While complying with the most basic ANA design principles, we have redesigned the architecture to a great extend and implemented it from scratch in the Linux kernel by utilizing design principles for efficient networking architectures. The outcome that we have called Lightweight ANA (LANA) has a competitive packet per second performance with the Linux networking subsystem and is about 21 times faster than the original ANA prototype. In this work, we have shown that it is possible to fulfill three partially conflicting goals with LANA: i) high flexibility for network programmers, ii) re-configuration of the network stack at runtime, and iii) high packet processing rates. With a voice-over-Ethernet example application, we have demonstrated that it is possible to adapt the underlying protocol stack during runtime without notable voice interruption. Hence, we claim that with LANA we have developed the base for a successful future Internet architecture.