Antnet

from Wikipedia, the free encyclopedia

AntNet is a routing concept based on ant algorithms developed by the Italian mathematicians Gianni Di Caro and Marco Dorigo . It was first presented in 1997 . AntNet works with agents who use stigmergy to exchange information.

functionality

Network nodes send an agent ("Forward Ant") to any known destination node at regular intervals. On its way, the agent collects the IDs of the visited nodes and the time it took until then on its stack . When it arrives at the destination node, a second agent ("Backward Ant") is generated, which takes over the stack from the Forward Ant and returns to the starting node the same way. The routing information on the intermediate node is updated with the values ​​from the agent's stack.

Data structure on the node

The data structure on a node k consists of a quality table and a cost table. The quality table consists of entries of the form ( i ; n ; P ) for each pair of target nodes i and neighboring nodes n of node k . P denotes the quality of n with respect to i . If a backward ant comes from the destination node i via the neighboring node n , the quality of this neighboring node is increased and the quality of all other neighboring nodes is reduced.

The cost table contains mean values and variances of the latencies to the target nodes and are used to stabilize the quality table in the case of strongly fluctuating network conditions.

rating

The algorithm was tested by Dorigo and Di Caro using a discrete event simulator with several network topologies such as NTTnet (Japan Backbone) and NSFNet (USA Backbone) and compared with other algorithms (including OSPF , BF ) for throughput and latency. In these tests, AntNet showed a particularly efficient and robust behavior compared to other routing methods.