Architectures for application transparent proxies : A study of network enhancing software

Proxies, software deployed inside the network, play a fundamental role in the Internet by providing enhanced functionality to the network. Deployment of proxies is a flexible way of extending the Internet architecture with new services and to cope with problems that were not foreseen at the time the original Internet protocols were defined. The creation of the Internet is an enormous investment in time, effort and money, and proxies allow us to build on the existing infrastructure to enhance its functionality, rather than replace it. As the use of proxies increase, so does the problem of proxy configuration and deployment, especially with respect to interference between different proxies. With a limited number of different proxies, this problem can be dealt with manually, or by encoding knowledge of interfering proxies into each proxy. As the number of proxies grow, methods to automatically detect and cope with conflicts must be devised. Therefore, proxies need to coordinate with each other. Towards this end, a signalling protocol that can be used to establish and configure a sequence of one or more proxies along an end-to-end flow is proposed. The protocol is realized as an extension of IP, using an IP option, which simplifies its deployment in the Internet. In order to facilitate reasoning about coordination, conflicts and deployment of proxies, a model has been developed. The model is based on the concept of regions, also developed in the thesis. Regions are interconnected parts of the network that share a common property, e.g. administrative control or error characteristic. Along with the model, a classification of proxy architectures with respect to how they gather information and deploy proxies is presented. A method based on this model is also proposed. We also propose an algorithm for controlling compression to maximize perceived throughput in situations where available bandwidth and CPU power varies. Along with the algorithm, experimental results that show that the algorithm approximates the best non-adaptive choice in a number of situations are presented. This algorithm has been implemented as an end-to-end enhancement