Abstract:
Real-time monitoring applications have networks of Internet-of-Things (IoT) devices
sense and communicate information from a scene of interest to a monitor, for example,
a server in the cloud. Often the sensed information is first communicated by
the sensors, over an access network, to an IoT gateway that aggregates and sends
the sensed information to the monitor. The monitor would like to maximize the
freshness of sensed information at its end. In this work, we use the metric of age of
information to quantify freshness at the monitor. Earlier work has considered a set of polling policies, named Poll-s, that have the gateway poll one sensor for fresh information at a time and have the gateway send the polled information to the monitor after a certain a priori fixed number s of sensors have been polled. In our work, we (a) analyze a simpler randomized policy that unlike Poll-s doesn't require knowledge of the vector of ages of information at the monitor at every decision instant, (b) derive a lower bound on the achievable avrage age at the monitor, (c) propose a heuristic policy that at every decision instant chooses to poll a sensor or transmit to the monitor such that the average drift in age at the monitor is minimized, and (d) propose improvements via optimal ordering of polling of sensors over the best earlier proposed Poll-s policies for when the random times taken to poll sensors are not identically distributed. We show empirically over a wide range of assumed distributions of polling times of sensors,
and the time taken to transmit to the monitor, that the randomized policy achieves
an age that is within 2:5 * the lower bound. We also provide a detailed comparison
of the different policies.
