Learning models of macrobehavior in complex adaptive systems
AAAI'06 proceedings of the 21st national conference on Artificial intelligence - Volume 2
| Hi-index | 0.00 |
Distributed computing systems are increasingly being created as self-organizing collections of many autonomous (human or software) agents cooperating as peers. Peer-to-peer coordination introduces, however, unique and potentially serious challenges. When there is no one "in charge", dysfunctions can emerge as the collective effect of locally reasonable decisions. In this paper, we consider the dysfunction wherein inefficient resource use oscillations occur due to delayed status information, and describe novel approaches, based on the selective use of misinformation, for dealing with this problem. A model of several servers offering equivalent service to independent clients is presented and studied numerically and analytically; the spreading of misinformation about the queue status is found to dampen oscillations and improve system performance for a wide range of parameters.