Dynamic Software Reconfiguration in Programmable Networks

Programmable networks allow third parties to reprogram networking devices. By opening up the execution environment of routers, firewalls, gateways, etc., users and service providers can adapt the behavior of these devices to meet their own specific needs. Programmable networks are therefore an interesting technology to build adaptive networks and to support the increasing evolution of networking software. At the same time, it can be perceived that many distributed applications impose stringent availability and performance requirements on the employed network infrastructure, among others to meet increased user expectations. Interrupting network communication to update or to customize the network software on programmable network devices hence may have extensive consequences. This dissertation, therefore, is targeted at supporting the reconfiguration of network software dynamically – that is, without temporarily shutting down (parts of) the network. Whether or not such dynamic reconfigurations are beneficial depends very much on the effectiveness and efficiency of the reconfiguration process. Besides, implementing a correct reconfiguration that causes limited overhead can be very complex and errorprone (hence compromising the benefit of a dynamic reconfiguration). We argue that specific reconfiguration support is needed, therefore, which (1) conducts the effective and efficient reconfiguration of network software, and (2) conceals the complexity of these reconfigurations from users or service provides who initiate the actual reconfigurations. This dissertation proposes the NeCoMan (Network reConfiguration Management) middleware as reconfiguration support for programmable networks. In short, this middleware coordinates the runtime addition, replacement, and removal of both local and distributed network services among out-of-band active nodes. The novelty of this middleware is in its ability to tailor the reconfiguration process. To accomplish this, the NeCoMan middleware includes various reconfiguration algorithms as well as an extensive set of customizations to these algorithms. This enables NeCoMan to customize the reconfiguration process starting from (1) a declarative description of the recomposition that must be executed and (2) a specification of the network service characteristics and the reconfiguration semantics. To conclude, we summarize the key contributions of this dissertation. Besides proposing a middleware to reconfigure out-of-band active nodes and validating this middleware by a number of reconfigurations, we present an extensive analysis on how to coordinate local and distributed out-of-band compositional adaptations. In addition, this dissertation proposes to make change management support customizable. In contrast to existing change management support (which typically conforms to the black-box philosophy by encapsulating a single and fixed reconfiguration algorithm) NeCoMan tailors the employed reconfiguration algorithm to exploit the network service characteristics and the reconfiguration semantics.