Nowadays real-time networks are indispensable when it comes to modern industrial applications. Traffic has to be transmitted with hard or soft delay bounds and guaranteed throughput. Handling this traffic and maximizing the throughput in a network can be very challenging depending on the type of traffic. On the one hand, there is planned traffic in a network which is easy to handle because of its predictable nature. On the other hand, there is unplanned traffic, e.g. event-triggered traffic, which is much harder to manage. This event-triggered traffic is needed to allow applications to improve control loop quality or give them the possibility to handle alarm storms, where one event forces many reactions in the network. Thus, techniques are needed to handle unplanned traffic, such as event-triggered traffic. To do so, over provisioning a network is not an option because it would be to expensive and the resources of the network would not be optimally used. This means new options have to be investigated to handle unplanned traffic for time-sensitive applications.

This thesis introduces a new approach where traffic shaping is used to handle traffic of time-sensitive application like event-triggered traffic. The new approach helps to improve the overall utilization in the network while guaranteeing a certain bandwidth for very application. To do so shaping with application knowledge at the network edge is used to optimize the performance, and in-network shaping with limited knowledge about the network situation helps to cope with worst-case situations.

The approach has the benefit that it is almost stateless and does not have the requirement of any per-stream basis knowledge except of a metering algorithm at the edges of a network. In addition to that the shaping at the edges of a network allows a higher send rate at the application side. This gives applications the possibility to improve utilization in the network, while also limiting single applications from greedily spamming the network with to much data.

The improvements and the increasing overhead as well as necessary technologies of this new approach will be introduced in this thesis. The evaluation shows that it is possible to guarantee a certain bandwidth and delay for every stream while increasing the utilization for specific network scenarios and traffic characteristics such as event-triggered traffic. The thesis also provides the theoretical background on how to calculate a worst-case estimation for the delay, drop rate of messages and the throughput for this new approach. These estimations are done using the Network Calculus framework. Furthermore the fairness between streams and limiting factors for this approach are discussed and evaluated.