Nowadays video-on-demand servers have to fulfil high expectations to attract and keep clients. Typically, they work in Internet settings serving large numbers of requests. New concepts and architectures are being developed to provide better services than the competitors and to take full advantage of the available resources. One area in these new developments is investigating the inherent possibility of offensive adaptation, the aim being to increase the extent of available resources if they are currently insufficient to fulfil a task.
(The defensive technique, in contrast, tries to "reduce" the task to fit to the given circumstances.) Offensive adaptation techniques usually need to be proactive and to have flexible, self-organizing architecture.
This thesis provides deeper insight into the topics connected with offensive adaptation and proactive, self-organizing video servers. It identifies unconsidered issues connected to these topics and offers solutions for them. Above all the work allows the sophisticated development of such servers, scaling prior to installation and performance evaluation. Finally, it presents studies analysing the usefulness of the above-mentioned concepts.
As mentioned, servers applying offensive adaptation need to be proactive, which in turn means that they need to predict the circumstances of the near future. Forecasting the resources for self-organizing video servers has already been discussed in some other works. Here, the emphasis is on predicting the clients' behaviour and therefore a large number of predictor functions were investigated. This investigation includes the examination of similarity measures in order to compare the results provided by the predictors and the analyses of the predictors in artificial and real test scenarios.
A mathematical frame model was constructed to help scale video servers prior to installation and exploit the abilities of self-organizing ones.
The model's static part enables us to estimate the characteristics and the performance of a video server in a given infrastructure. Its second half, the dynamic part, allows us to calculate the utilization and balance of the system alongside different performance measures. Based on these functions, a control scheme aiming at a permanently advantageous state in a self-organizing server was introduced.
A simulation environment was then developed which can assist in the design of self-organizing video servers and their algorithms. It enables the mapping of video servers and their foreseen infrastructure onto the simulation model, thus facilitating their examination with different algorithms and settings under different circumstances and load. The simulation environment was put into practice with the Adaptive Distributed Multimedia Server (ADMS) architecture. This simulation provoked many new considerations regarding offensive adaptation and self-organization. Furthermore, the results of two small performance studies and one large-scale study are presented, the large study representing a possible method for performance analysis as well. Its conclusion is that the self-organizing behaviour can bring significant improvement in an environment with restricted resources. If the resources are too restricted or not restricted enough, the static behaviour shows a small advantage over self-organizing behaviour.