Distributed computing with mobile agents: Solving rendezvous and related problems

• Main Author: Das, Shantanu
• Format: Others
• Language: en
• Published: University of Ottawa (Canada) 2013
• Subjects: Computer Science.
• Online Access: http://hdl.handle.net/10393/29643; http://dx.doi.org/10.20381/ruor-19839

In this thesis we study mobile agent systems consisting of a network of nodes and a set of autonomous agents travelling through the network. The mobile agents can traverse the edges (channels) of the network to move from one node to another and they can communicate with each other using public whiteboards available at each node of the network. We focus on the Rendezvous problem which is the central problem in such systems. The objective of the Rendezvous problem is to gather all the agents at a single node of the network. A procedure for Rendezvous is useful while performing many collaborative task by distributed agents---for instance, the agents may need to gather together for planning and coordinating their next operation. The problem of Rendezvous is in fact, related to several other problems in this setting, such as leader election, network exploration, distributed spanning-tree construction and graph-labelling or enumeration. Compared to most existing solutions, we study the Rendezvous problem in a weaker (and computationally difficult) model where both the agents and the network nodes are anonymous, and the network is asynchronous. Our focus in this thesis is not only determining when the problem is solvable in this setting but also designing efficient algorithms for the solvable cases. The main measure of efficiency in our algorithms is the number of moves (edge traversals) made by the agents in total. A secondary goal is to minimize the memory required for the whiteboard at each node of the network. We present solutions for the Rendezvous problem, both in the whiteboard model and the more restrictive token-model. We also consider situations where some network components may fail or the tokens used by the agents may disappear. We present fault-tolerant algorithms for these cases. We also show how to simulate any distributed computation in mobile agent systems when some of agents crash unexpectedly. Our results show that any problem that can be solved in message-passing systems can be solved in mobile agent systems while tolerating any number of crash faults, short of a total system collapse.