Elsevier

Computer Communications

Volume 22, Issue 12, 25 July 1999, Pages 1127-1135
Computer Communications

Bandwidth-demand prediction in virtual path in ATM networks using genetic algorithms

https://doi.org/10.1016/S0140-3664(99)00075-4Get rights and content

Abstract

The concept of Virtual Paths (VP) is a powerful technique to improve the transmission efficiency in ATM networks. Transmission efficiency can be improved by dynamically changing the bandwidths of the VPs, based on the demand. Intelligent controllers, which predict bandwidth-demand patterns to enable better VP management, have the potential to revolutionize ATM network performance. We present a scheme based on the Evolutionary Genetic Approach to predict the bandwidth-demand patterns in VPs. The efficiency of this approach, quantified in terms of the Degree of Learning (DoL), is evaluated through simulation and the results are presented.

Introduction

The Virtual Path (VP) concept is considered an effective means to manage an ATM network [1], [2]. The benefits of using VPs are many: reduced node processing per virtual circuit, reduced call setup time, simplified node structure, and proper control over routing and bandwidth management. VP bandwidth control improves transmission efficiency for a given call blocking probability (and vice versa) [1]. Improvement in efficiency is achieved through the multiplexing of the physical link bandwidth among the VPs that share the same physical link. The VP bandwidth is dynamically altered in accordance to the traffic demand. There are many schemes suggested in the literature for VP bandwidth control, some centralized [3], [4], [5], [6], [7], [8], [9], [10] and others distributed [11], [12], [13]. For efficient management of the VP bandwidth, an accurate estimate of the bandwidth-demand of the traffic flowing through the VPs is required. In order to make an accurate estimate of the bandwidth-demand, a proper understanding of the behavior of the traffic is required.

The behavior of the traffic is usually expressed in terms of its past statistical properties. For short term (less than one hour) estimate of traffic behavior the statistical properties are assumed to vary slowly and hence daily, weekly and seasonal cycles are not considered, instead the peak demand during the previous observation period is considered as the current demand [13]. Characterizing the traffic by on-line measurement of the large deviation rate function (entropy) is studied in [14], [15]. In this paper we characterize the traffic using bandwidth-demand patterns within a VP. The bandwidth-demand patterns have to be learned to predict the future behavior of the traffic.

Learning algorithms have been suggested for adaptive routing in ISDN [16] and multicast routing [17] based on the network load. We propose an evolutionary-genetic approach to learn the dynamic behavior of traffic on the basis of bandwidth-demand patterns in VPs. We also describe as to how to use the bandwidth-demand patterns, in order to make short term bandwidth-demand predictions in VPs that is subsequently used for VP bandwidth management. From our studies, it is observed that the genetic approach naturally captures the short-term trend and variation in bandwidth-demand for both Poisson and Self-Similar call arrivals.

The remaining part of the paper is divided as follows. Section 2 describes the proposed approach by developing the framework for using genetic algorithms in such class of problems. Simulation studies and the results are discussed in Section 3. Section 4 concludes the paper by exploring the scope for further research.

Section snippets

The evolutionary-genetic algorithm

A genetic algorithm is a heuristic approach that applies the natural genetic ideas of natural selection, mutation and survival of the fittest. A rigorous mathematical formalism was introduced by Holland and his collaborators, and is till date the basis for genetic algorithms. A comprehensive review of this field can be found in the book by Goldberg [18].

The algorithm uses a set of offered solutions called a “population”. Each solution called an “individual”, can be any solution in the solution

Simulation experiments

In this section we verify the effectiveness of the proposed bandwidth-demand predictor through simulation. First we describe the simulation environment and then explain the various experiments conducted and the inferences derived.

Conclusion

Virtual Path (VP) bandwidth control improves transmission efficiency in an ATM network. An accurate estimate of the bandwidth-demand within a VP leads to efficient VP bandwidth control. So far the statistical methods were employed to predict the bandwidth-demand. In this paper we have presented an Evolutionary-Genetic Approach (EGA) to predict bandwidth-demand patterns within a VP. We have quantified the efficiency of this scheme in terms of the Degree of Learning. Factors affecting the

References (22)

  • S. Ohta et al.

    Dynamic bandwidth control of the virtual path in an asynchronous transfer mode network

    IEEE Trans. Commun.

    (1992)
  • J. Burgin et al.

    Broadband ISDN resource management: the role of virtual paths

    IEEE Commun. Mag.

    (1991)
  • R. Siebenhaar

    Optimized ATM virtual path bandwidth management under fairness constraints

    Proc. IEEE Globecom

    (1994)
  • K.T. Cheng et al.

    On the joint virtual path assignment and virtual circuit routing problem in ATM networks

    Proc. IEEE Globecom

    (1994)
  • G. Gordos et al.

    A new degree of freedom in ATM network dimensioning: optimizing the logical configuration

    IEEE J. Selected Areas Comm.

    (1995)
  • M. Logothetis et al.

    Medium-term centralized virtual-path bandwidth control based on traffic measurements

    IEEE Trans. Comm.

    (1995)
  • Ake Arvidsson, High level B-ISDN/ATM traffic management in real time, Proc. of Second IFIP Workshop on Performance...
  • Saito Hiroshi

    Dynamic resource allocation in ATM networks

    IEEE Commun. Mag.

    (1997)
  • M. Pioro, P. Gajowniczek, Stochastic allocation of virtual paths to ATM links, Proc. of Second IFIP Workshop on...
  • N. Anerousis et al.

    Virtual path control for ATM networks with call level quality of service guarantees

    IEEE Trans. Networking

    (1998)
  • S. Shioda et al.

    Virtual path bandwidth control method for ATM networks: successive modification method

    IEICE Trans.

    (1992)
  • Cited by (14)

    • Intrusion detection through learning behavior model

      2001, Computer Communications
      Citation Excerpt :

      Genetic algorithms are used in various applications for searching, optimization, and machine learning as described in Ref. [6]. Recently, it could be observed that genetic algorithms are widely being used for machine learning [7,9,11]. In the behavior learning process each command of user forms a gene.

    View all citing articles on Scopus
    View full text