The Reliable Server Pooling (RSerPool) architecture currently under standardization by the IETF RSerPool Working Group is an overlay network framework to provide server replication and session failover capabilities to applications using it. These functionalities as such are not new, but their combination into one generic, application-independent framework is. Initial goal of this thesis is to gain insight into the complex RSerPool mechanisms by performing experimental and simulative proof-of-concept tests. The further goals are to systematically validate the RSerPool architecture and its protocols, provide improvements and optimizations where necessary and propose extensions if useful. Based on these evaluations, recommendations to implementers and users of RSerPool should be provided, giving guidelines for the tuning of system parameters and the appropriate configuration of application scenarios. In particular, it is also a goal to transfer insights, optimizations and extensions of the RSerPool protocols from simulation to reality and also to bring the achievements from research into application by supporting and contributing relevant results to the IETF's ongoing RSerPool standardization process. To achieve the described goals, a prototype implementation as well as a simulation model are designed and realized at first. Using a generic application model and appropriate performance metrics, the performance of RSerPool systems in failure-free and server failure scenarios is systematically evaluated in order to identify critical parameter ranges and problematic protocol behaviour. Improvements developed as result of these performance analyses are evaluated and finally contributed into the standardization process of RSerPool.

The Internet is developing with a very high speed. Just a few years ago, its main applications were file transfer and e-mail access, all based on the best effort TCP/IP service over quite slow links. With the quickly growing deployment of high-speed Internet accesses, there is a growing demand for also using the Internet in application scenarios like real-time multimedia streaming, particularly for audio and video on demand services. However, such multimedia applications do not only require sufficient bandwidth. They have stricter quality of service demands, like e.g. an upper bound on the transfer delay to ensure interactivity for their users. In the context of this thesis, a system for the cost-efficient transfer of variable bitrate multimedia streams over Differentiated Services (DiffServ) is presented and evaluated. This system supports layered transmission, i.e. streams may be partitioned into sub-streams denoted as layers (e.g. a video layer and an audio layer). Distinct layers may have different priorities as well as quality of service requirements and may therefore use different DiffServ classes for data transmission. Also, the streams may be scalable, i.e. the quality of the transferred media may be decreased in order to reduce the bandwidth requirements when bandwidth becomes scarce. Particularly, the system makes use of the fact that media files in e.g. a video on demand library can be analyzed a priorily. The information that is computed during the offline analysis is used to realize a dynamic and cost-efficient transport of layer data over different DiffServ classes. Furthermore, the system applies an adaptive buffering to smooth the flows – in order to save bandwidth – while also taking care of the delay constraints.