eng
Schloss Dagstuhl – Leibniz-Zentrum für Informatik
Leibniz International Proceedings in Informatics
1868-8969
2018-03-28
14:1
14:16
10.4230/LIPIcs.OPODIS.2017.14
article
Treasure Hunt with Barely Communicating Agents
Dobrev, Stefan
Královic, Rastislav
Pardubská, Dana
We consider the problem of fault-tolerant parallel exhaustive search, a.k.a. “Treasure Hunt”, introduced by Fraigniaud, Korman and Rodeh in [13]: Imagine an infinite list of “boxes”, one of which contains a “treasure”. The ordering of the boxes reflects the importance of finding the treasure in a given box. There are k agents, whose goal is to locate the treasure in the least amount of time. The system is synchronous; at every step, an agent can ”open” a box and see whether the treasure is there. The hunt finishes when the first agent locates the treasure.
The original paper [13] considers non-cooperating randomized agents, out of which at most f can fail, with the failure pattern determined by an adversary. In this paper, we consider deterministic agents and investigate two failure models: The failing-agents model from [13] and a “black hole” model: At most f boxes contain “black holes”, placed by the adversary. When an agent opens a box containing a black hole, the agent disappears without an observable trace.
The crucial distinction, however, is that we consider “barely communicating” or “indirectly weakly communicating” agents: When an agent opens a box, it can tell whether the box has been previously opened. There are no other means of direct or indirect communication between the agents.
We show that adding even such weak means of communication has very strong impact on the solvability and complexity of the Treasure Hunt problem. In particular, in the failing agents model it allows the agents to be 1-competitive w.r.t. an optimal algorithm which does not know the location of the treasure, but is instantly notified of agent failures. In the black holes model
(where there is no deterministic solution for non-communicating agents even in the presence of a single black hole) we show a lower bound of 2f + 1 and an upper bound of 4f + 1 for the number of agents needed to solve Treasure Hunt in presence of up to f black holes, as well as partial results about the hunt time in the presence of few black holes.
https://drops.dagstuhl.de/storage/00lipics/lipics-vol095-opodis2017/LIPIcs.OPODIS.2017.14/LIPIcs.OPODIS.2017.14.pdf
parallel exhaustive search
treasure hunt
fault-tolerant search
weak coordination
black holes