In Fog Computing, FemtoClouds are emerging computing systems consisting of a set of heterogeneous mobile devices whose users allow to run tasks offloaded by other users. FemtoClouds are well suited to run Bag-of-Tasks (BoTs) applications, but they need effective scheduling algorithms that are able to deal with collections of independently-owned, heterogeneous devices that can suddenly leave the system. In this paper, we present UDFS, an online scheduling algorithm that, by combining knowledge-free task and device selection policies with suitable heterogeneity and volatility tolerance mechanisms, can effectively schedule a stream of BoT applications on FemtoClouds. We evaluate the ability of UDFS to achieve its design goals and to perform better than existing scheduling alternatives, by carrying out a thorough simulation study for a large set of realistic scenarios. Our results indeed show that UDFS can effectively schedule a stream of BoT applications on FemtoClouds, and it can do so more effectively than existing scheduling alternatives.