Efficient and Transparent Use of personal device storage in opportunistic data forwarding

We consider a growing research trend of using personal mobile devices for forwarding opportunistic network data. Because personal device storage is meant to support user applications, opportunistic networks must use it in a manner that remains completely transparent to the user. One way to make a device's storage use transparent is to allow priority access to the storage to user applications, even if the storage is currently occupied by network data yet to be forwarded. This means that data given to a device waiting to be forwarded can be overwritten by application data and may, thus, be lost. In this paper we consider random access memory (RAM) as the primary storage location in a mobile device. We propose three algorithms of different sophistications to answer the question of how much data should be moved when a contact opportunity arises between two devices in such a way to first maximise the data transferred while minimising the probability that this data will be overwritten when applications claim priority access. We collect 33 h of high-resolution RAM usage traces of two real smartphones over a 3-day period under a variety of usage scenarios to evaluate and compare the performances of the proposed algorithms. Surprisingly, we find that autoregression forecasting of RAM usage cannot outperform the simplest algorithm that greedily occupies all of the RAM that is found unused at the time of contact. We show that Bayesian inference is very effective in minimising the risk of data loss in such uncertain environments and significantly outperforms the greedy approach as well as autoregression forecasting.