eXperience Based Admission Control (XBAC)

The XBAC project aimed to create an overlay network architecture that provides a distance estimation service to user applications. Before an application opens a connection to a remote endpoint it can ask for a prediction of the quality of service to this destination. Depending of the prediction it can then decide to proceed with connection setup or to back off. Moreover, peer-to-peer networks can use the service to optimally configure their topology and also to adapt to later changes in quality of service.

The XBAC architecture is based on the idea of creating groups of endpoints that are close to each other in the physical network topology. While communicating these endpoints gather measurement data, which they distribute to the other endpoints in their cluster. Thereby they create a common pool of network measurements, termed the group's experience. Using this experience, the group can give robust answers to QoS prediction requests. In contrast to other approaches, XBAC can detect clusters of remote nodes and groups based on end-to-end measurements, which leads to better scalability and enables inclusion of non-cooperative nodes into the system.

Based on the architecture defined in the XBAC project we have designed and developed a prototype for experimental deployment on PlanetLab. We have also defined test cases to estimate the scalability of the implementation in terms of CPU, memory, and bandwidth usage. Furthermore, evaluation of several possible variants of the joining-node procedure has allowed us to select the optimal one.

Project Details