In an ad-hoc network, client-server based application can encounter some problems due to the dy- namic topology. One of these problems is when the network splits, leading the server and the client to be physically unreachable from each other. Predicting those partitions could be a very useful feature that can be provided to applications. Indeed, being aware of a future disconnection in the network can help to ensure a better quality of service by adapting the client and/or the server behavior. Algorithms already exists to do this, but they are based on trajectory interpolation which need position information to be provided by a positioning system. This paper propose an original link robustness evaluation method based on the notion of disjoint paths set that allow efficient partition detection without using any kind of positioning system. After showing that the use of disjoint path is relevant for detecting network parti- tions, we propose a distributed algorithm that collects disjoint paths available between the server and the client and thus show that our partition detection metric can be used in a real network. Moreover, those set of disjoint paths could be used in multipath routing protocols.