doi:10.1016/S0140-3664(02)00043-9
Copyright © 2002 Elsevier Science B.V. All rights reserved.
Network resource brokerage by means of distributed agent-based systems encompassing reinforcement learning schemes
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
M. Saltouros
, 
, a, b, A. Taskarisa, P. Demestichasa, M. Theologoua and A. Vasilakosc
a Telecommunication Laboratory, Department of Electrical and Computer Engineering, National Technical University of Athens, 9 Heroon Polytechniou Street, Zographou, GR-157 73, Athens, Greece
b Department of Engineering, Hellenic Air Force Academy Greece, Greece
c Hellenic Aerospace Industry S.A., P.O. Box 23, GR-32009, Schimatari, Greece
Received 28 January 2002;
accepted 28 January 2002.
Available online 4 March 2002.
Abstract
The success of a network provider in a competitive communications market depends on the cost-effective provision of the appropriate QoS levels, and the ability to promote the network infrastructure, using dynamic attracting service providers. Advanced management tools encompassing the appropriate intelligence are enabling concepts in this direction. The aim of this paper is to present a part of a Network and Service Management System that acts as a distributed Bandwidth Brokerage System (BBS) over an Internet segment. Aspects addressed are the overall role of the BBS, its functional decoupling, the component deployment pattern in complex (hierarchical) network structures, and the computational intelligence encompassed in the components. The BBS components' logic, which is based on the Stochastic Estimator Learning Algorithm (SELA) concept, solves a version of the hierarchical routing and bandwidth management problem. Mathematical descriptions of the SELA concept and the corresponding routing and bandwidth management schemes are provided. Finally, the paper provides results on the efficiency of the BBS in managing network segments of commercial size and connectivity degree.
Author Keywords: Service provider; Routing; Learning algorithms; Foundation for intelligent physical agents
Fig. 1. Business model in open communication environments.
Fig. 2. Hierarchical structure of networks. Three logical nodes and the corresponding peer groups are depicted in the arbitrary (sample) structure above. In peer group 1 node v11 is an EBN. In peer group 2 nodes v21 and v22 are EBNs, while v21 is also a PGBN. Finally, in peer group 3 nodes v31 and v34 are PGBNs.
Fig. 3. General presentation of NSMS architecture.
Fig. 4. Pattern for distributing the BBS components (in the example of Fig. 2). Node v11 that has been assumed an EBN is controlled by a BBS-RA. All other nodes in peer group 1 are controlled by BBS-NAs. In the other peer groups the EBNs and/or PGBNs, namely nodes v21, v22, v31 and v34, are controlled by BBS-RAs.
Fig. 5. Sample BBS operation.
Fig. 6. Network topologies used in the experiments. (a) The NSFnet-like network consists of 14 nodes, 22 bi-directional links, average degree 3, and is divided into four PGs for a two-level hierarchical routing. (b) The vBNS-like network topology that runs on 622.08 Mbps links and interconnects 12 points of presence (POPs).
Table 1. Network parameters
Table 2. Parameter values of SELA
Table 3. Performance of the different versions of the BBS logic in terms of the earned network revenue ratios (ENRR) under various, uniform loading conditions for the vBNS-like network topology
Table 4. Performance of the different versions of the BBS logic in terms of the bandwidth rejection ratios (BRR) under various uniform loading conditions for the vBNS-like network topology
Table 5. Performance of the different versions of the BBS logic in terms of the achieved network revenue ratios (ENRR) under time-varying loading conditions for the vBNS-like network topology
Table 6. Performance of the different versions of the BBS logic in terms of the bandwidth rejection ratios (BRR) under time-varying loading conditions for the vBNS-like network topology
Table 7. Performance of the different versions of the BBS logic in terms of the average packet transfer delay under various uniform loading conditions for the NSFnet-backbone topology configured as a two level hierarchical network
Table 8. Performance of the different versions of the BBS logic in terms of the bandwidth acceptance ratio (BAR) under various uniform loading conditions for the NSFnet-backbone topology configured as a two level hierarchical network
Table 9. Performance of the different versions of the BBS logic in terms of the average packet transfer delay (i.e. the average delay required to deliver a packet from origin to destination) under time-varying and skewed loading conditions for the NSFnet-backbone topology configured as a two level hierarchical network
Table 10. Performance of the routing algorithms in terms of the bandwidth rejection ratio under time-varying and skewed loading conditions for the NSFnet-backbone topology configured as a two level hierarchical network
Corresponding author. Address: Telecommunication Laboratory, Department of Electrical and Computer Engineering, National Technical University of Athens, 9 Heroon Polytechniou Street, Zographou, GR-157 73 Athens, Greece. Tel.: +30-1-772-1478; fax: +30-1-772-2534; email: pdemest@telecom.ntua.gr
Abstract
Purchase PDF (365 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (2)
Purchase PDF (578 K)
