History based techniques for device management and congestion control in mobile networks

We consider two problems in mobile computing that we approach through using historical data to develop relevant predictors.;We first present machine learning based algorithms by which a cell phone can discern that it may be lost, and take steps to enhance its chances of being recovered. We use data collected from the Reality Mining project to create a suite of test cases that model lost cell phone behavior. On these data sets our best algorithms can identify cases of a lost mobile device, based on its behavior over the previous 3 hours, with close to 100% accuracy.;We then study the problem of congestion control in DTN. A strategy consists of a drop policy and an algorithm that attempts to avoid congestion events. We introduce a simple drop policy, REJECT, that has not been considered in prior work, and a congestion avoidance algorithm: TISS.;In simulation experiments on popular archived real-life data sets, we show that REJECT can substantially improve delivery rates over prior techniques that focus on receiving nodes rejecting either incoming messages or messages held in their buffers. We refer to such strategies as EJECT. For two of the most important DTN routing algorithms, REJECT increases the delivery rate by 28.96% over EJECT. REJECT works exceptionally well when the DTN interaction graph is well connected, which causes earlier policies to eject large numbers of messages. In that case it outperforms EJECT by 55.78% in delivery rate.;We conduct an extensive experimental evaluation of our algorithm. We consider TISS when applied to the two DTN routing algorithms. In addition we modified other approaches to Congestion Control, Storage Routing (SR) and Fair Routing (FR), so as to be able to compare fairly (for those algorithms) to their core ideas. We show that the combination of [REJECT, TISS] is the best overall strategy to improve the delivery rate. It improves over [EJECT, No Congestion Avoidance] by 31.89%. In comparison, our (advantageously) modified versions of two other approaches to congestion control [REJECT, SR] and [REJECT, FR] improve by 24.16% and 27.57%, respectively.