LIP6 2011/002
- Reports
Une architecture du réseau autonomique basée sur l’apprentissage par paquet - Z. Movahedi, R. Langar, G. Pujolle
- 7 pages - 09/02/2011- document en - http://www.lip6.fr/lip6/reports/2011/lip6-2011-002.pdf - 415 Ko
- Contact : zeinab.movahedi (at) nulllip6.fr
- Team : Phare
- Keywords : Autonomic architecture, cognitive networking, learning mechanisms.
- Publisher : Aziza.Lounis (at) nulllip6.fr
Autonomic computing paradigm represents an emerging solution to deal with the ever-increasing size and complexity of managing IT systems. When applied to the networking, it relates to the capability of the network to operate and serve its objectives optimally by managing its own self without external intervention. We distinguish between two levels of management: Intra-application level which consists in managing different parameters within an application or service in order to optimize it; and inter-application level which is in charge of optimizing between several different applications and services. In this paper, we propose a generic approach applicable to both area. Our proposed architecture is based on random neural networks and uses the reinforcement learning. This enables the network to continuously converge to optimal configurations when network conditions change. In addition, the random neural network decision-making process is fed by normal transiting packets in the network, which significantly reduce the amount of control traffic. To show the effectiveness or our architecture, a case study consisting in optimizing the performance of mobile ad hoc network routing protocols is used. We proposed a dynamic routing protocol which interacts continuously with the architecture in order to enhance the network operation. Simulations show that the architecture improves significantly the QoS performance of ad hoc routing protocols.