Content-based routing systems such as publish/subscribe (pub/sub) have become an important model for distributed systems with loosely coupled participants. Usually publish/subscribe systems are realized by overlay networks where dissemination and filtering of information is done in the application layer, which causes significant delay. The emergence of software-defined networking (SDN), where switches with programmable TCAM memory allow dynamic configuration of networks, has opened new

more »

... ties in realizing dynamic logic inside the network. Current publications have presented realizations of pub/sub systems based on SDN. In these systems the information filtering is not done in the application layer but directly inside the network by switches. This allows event filtering with low delay and line-rate performance. However, SDN-based pub/sub systems are limited by the available resources. The TCAM memory of the switches, containing the forwarding rules, is very cost-intensive and hence the maximum number of rules and their complexity is limited. In order to provide bandwidthefficient content-based routing it is necessary to use a large number of complex forwarding rules. Therefore the limitation of resources causes a drop of the routing quality and less bandwidth-efficient routing. In this thesis, approaches to increase bandwidth-efficiency in the context of limited resources are proposed. To achieve efficient routing, the precision of in-network filtering must be high to avoid unnecessarily disseminated information, so-called false positives, which cause higher network utilization. This thesis proposes and evaluates two approaches to increase the efficiency of in-network filtering: Selection of more important information to be used for filtering and improvement of the filtering itself. Several algorithms to rate the importance of information are proposed and evaluated. Furthermore, ways to combine the selection of information and the improved filtering are shown. Our results show that the developed approaches [...]