Finding mobile data under delay constraints with searching costs

A token is hidden in one of several boxes and then the boxes are locked. The probability of placing the token in each of the boxes is known. A searcher is looking for the token by unlocking boxes where each box is associated with an unlocking cost. The searcher conducts its search in rounds and must find the token in a predetermined number of rounds. In each round, the searcher may unlock any set of locked boxes concurrently. The optimization goal is to minimize the expected cost of unlocking boxes until the token is found. The motivation and main application of this game is the task of paging a mobile user (token) who is roaming in a zone of cells (boxes) in a cellular network system. Here, the unlocking costs reflect cell congestions and the placing probabilities represent the likelihood of the user residing in particular cells. Another application is the task of finding some data (token) that may be known to one of the sensors (boxes) of a sensor network. Here, the unlocking costs reflect the energy consumption of querying sensors and the placing probabilities represent the likelihood of the data being found in particular sensors. In general, we call mobile data any entity that has to be searched for.

The special case, in which all the boxes have equal unlocking costs has been well studied in recent years and several optimal polynomial time solutions exist. To the best of our knowledge, this paper is the first to study the general problem in which each box may be associated with a different unlocking cost. We first present three special interesting and important cases for which optimal polynomial time algorithms exist: (i) There is no a priori knowledge about the location of the token and therefore all the placing probabilities are the same. (ii) There are no delay constraints so in each round only one box is unlocked. (iii) The token is atypical in the sense that it is more likely to be placed in boxes whose unlocking cost is low. Next, we consider the case of a typical token for which the unlocking cost of any box is proportional to the probability of placing the token in this box. We show that computing the optimal strategy is strongly NP-Hard for any number of unlocking rounds, we provide a PTAS algorithm, and analyze a greedy solution. We propose a natural dynamic programming heuristic that unlocks the boxes in a non-increasing order of the ratio probability over cost. For two rounds, we prove that this strategy is a 1.143-approximation solution for an arbitrary token and a 1.108-approximation for a typical token and that both bounds are tight. For an arbitrary token, we provide a more complicated PTAS