We have recently proposed a semi-decentralized collaborative hosting system Wikipedia. In this system, individuals willing to help host Wikipedia join their computer to an overlay network of nodes, which handles a significant part of Wikipedia's workload, resulting in reduced operational cost for Wikipedia.
A serious issue we have to deal with is the fact that nodes in the collaborative overlay can join and leave the system at any time, either voluntarily or because of failures. So far, we have been able to show that our proposed architecture is able to properly deal with churn if participating nodes exhibit exponentially distributed session and inter-arrival times. However, churn in practice might not follow these distributions.
The objective of this project is to evaluate how the proposed hosting system behaves under various churn models. This should give us more insight on how the proposed hosting system actually copes with churn and might lead to improvements of the design. The project involves studying and implementing various churn models and experimenting with them under diverse parameter settings.
References