Computational intelligence in management of ATM networks

Soft Computing Journal manuscript No.(will be inserted by the editor)

Computational Intelligence in Management of ATMNetworks

A Survey of Current State of Research

Y. Ahmet Sekercioglu1, Andreas Pitsillides2, Athanasios Vasilakos3

1 Centre for Telecommunications and Information Engineering, Monash University, Aus-tralia

2 Department of Computer Science, University of Cyprus, Cyprus3 Computer Science Institute, Foundation for Research and Technology (CSI-FORTH),

Greece

Received: / Revised version:

Abstract Designing effective control strategies for Asynchronous Transfer Mode(ATM) networks is known to be difficult because of the complexity of the structureof networks, nature of the services supported, and variety of dynamic parametersinvolved. Additionally, the uncertainties involved in identification of the networkparameters cause analytical modeling of ATM networks to be almost impossible.This renders the application of classical control system design methods (whichrely on the availability of these models) to the problem even harder.

Consequently, a number of researchers are looking at alternative non-analyticalcontrol system design and modeling techniques that have the ability to cope withthese difficulties to devise effective, robust ATM network management schemes.Those schemes employ artificial neural networks, fuzzy systems and design meth-ods based on evolutionary computation.

In this survey, the current state of ATM network management research em-ploying these techniques as reported in the technical literature is summarized. Thesalient features of the methods employed are reviewed.

Key words Computational intelligence, ATM networks, fuzzy systems, neuralnetworks, evolutionary computation

1 Introduction

Asynchronous Transfer Mode (ATM) based networks are designed to be scalable,high-bandwidth, manageable, and have the flexibility of supporting various classes

2 Y. Ahmet Sekercioglu et al.

of multimedia traffic with varying bit rates and Quality of Service (QoS) require-ments. Thus, they have the potential to create a unified communications infras-tructure that can transport services with widely different demands on the network(such services include real-time video and voice with no tolerance to delays, butsome tolerance to loss, and data with some tolerance to delay, but no tolerance toloss).

An important difficulty of exploiting the potential of ATM optimally is themanagement and control complexity of the scheme itself (the basic concept is sim-ple). Since ATM simultaneously attempts to support voice, data and video appli-cations which all have differing performance and QoS requirements, optimal uti-lization of the network resources requires complex, nonlinear, distributed controlstructures. In order to achieve its potential, ATM networks will need to accommo-date several interacting control mechanisms, such as call admission control, flowand congestion control, input rate regulation, routing, bandwidth allocation, queuescheduling, and buffer management.

The complexity of the ATM networks and multidimensionality of the controlproblems dictate that traffic control in ATM networks be structured. The controlstructure is most likely to be implemented in a multilevel architecture which parti-tions the solution into different levels of control with varying temporal decompo-sition in network, call, and cell levels (Figure 1).

Due to the complex nature of the above mentioned control issues, some re-searchers are looking for solutions by application of Computational Intelligence(CI) techniques to design intelligent control systems to various aspects of ATMnetwork management, often supplementing the existing control techniques [PV00].Their motivation arises from the reported success of those techniques in variouspreviously unsolvable or difficult control problems in many diverse fields.

The focus of this paper is CI applications in ATM network control. It seeksto update, merge and (inevitably) summarize the previous reviews of the literature[Hab96,DD97,GRSC98].

2 Computational Intelligence

Computational Intelligence (CI) [Bez92,Bez94,Ped98] is an area of fundamen-tal and applied research involving numerical information processing (in contrastto the symbolic information processing techniques of Artificial Intelligence (AI)).Nowadays, CI research is very active and consequently its applications are appear-ing in some end user products.

The definition of CI can be given indirectly by observing the exhibited proper-ties of a system that employs CI components [Bez94]:

A system iscomputationally intelligentwhen it: deals only with numerical(low-level) data, has a pattern recognition component, and does not useknowledge in the AI sense; and additionally, when it (begins to) exhibit– computational adaptivity;– computational fault tolerance;– speed approaching human-like turnaround;

Computational Intelligence in Management of ATM Networks 3

Fig. 1 Multilevel traffic control in ATM networks.

– error rates that approximate human performance.

The major building blocks of CI are artificial neural networks, fuzzy logic, andevolutionary computation.

3 Applications of CI Techniques in Management of ATM Networks

A possible implementation of multilevel ATM control architecture partitions thecontrol domains into network, call and cell levels of varying temporal decomposi-tion (Figure 1) [Pit93, Chapter 2][NCGA95]. Time constants involved are: in thecell level in the order of microseconds; in the call level in the order of seconds tominutes; and in the network level minutes to hours.

This paper attempts to classify the research work on ATM network controlmethods along this partitioning and, presents an overview of the reported researchwhich employ fuzzy logic, artificial neural networks and evolutionary computa-tion.

4 Network Level Control

The main objective of network level control in an ATM network is to enable thecompletion of the maximum possible number of successful B-ISDN service calls[Yon90]. The network attempts to achieve this objective by implementing twomain control functions at the network level: fault management and resource man-agement.


4.1 Fault Management

Fault management is concerned with the detection, isolation, and correction ofacute failures that interrupt the availability of network resources. Besides acutefailures, some failures may appear intermittently, or malfunctions may graduallydegrade the network performance while network resources remain available (forexample, corruption of virtual path identifier (VPI)/virtual channel identifier (VCI)translation tables may cause misrouting for certain connections). It is the role offault management to continually monitor the network facilities to detect degra-dations in performance caused by such conditions, and respond with appropriateactions in order to minimize the effect on offered services [CL94]. This responsi-bility is implemented as part of the Operations and Management (OAM) in ATMnetworks.

In the area of fault management in communication networks, including ATMnetworks, reported research on applications of CI is very limited in scope. Appli-cations of AI in network management have been surveyed and reported by Smithand Fry [SF95].

4.2 Traffic Control and Resource Management

In the absence of faults and malfunctions, efficient utilization of network resourcesis maintained by the traffic control and resource management functions which in-volve routing and link capacity assignment. At the network level, route selectionand link capacity assignment to virtual paths are performed by using informationrelated to offered call traffic and tolerated call blocking probability.

First major function at the network level control in ATM networks is routeselection. In B-ISDN networks, links have to be described in terms of multiplemetrics, including QoS and policy constraints, which makes routing with multiplerequirements a difficult problem to solve (see [AD97] for an example).

A number of researchers have proposed fuzzy logic and artificial neural net-work based schemes for the route selection in ATM networks, such as Aboelelaand Douligeris’s [AD97] study of fuzzy logic based route selection, and Park andLee’s [PL95b,PL95a] work on optimized routing using recurrent ANNs.

Second major function of network level control in ATM networks is the op-timal allocation of bandwidth to virtual paths [CL94,PPS¸V97]. Effective imple-mentation of this function is very important for a number of reasons:

– By reserving capacity in anticipation of the virtual channels which will belongto a virtual path, the processing effort required to establish individual virtualchannels can be minimized.

– Virtual paths may be used as a means of logically separating traffic types hav-ing different QoS requirements while allowing virtual channels to be statisti-cally multiplexed.

– Virtual paths allow groups of virtual channels to be managed and policed moreeasily.


– Dynamic routing control at virtual path level should lead to designing effectivenetwork reconfiguration mechanisms.

– Effective allocation of bandwidth to virtual paths would lead to a more efficientand effective network.

Bandwidth allocation to virtual paths is a typical multiobjective optimization prob-lem. The application of combination of evolutionary programming with fuzzylogic could be very beneficial for this type of problems. Heuristic methods ex-ploiting these benefits have been proposed to reach near-optimal solutions [VAP97,VRA+98,PPS+00]. In the study presented in [VRA+98] the researchers have usedevolutionary-fuzzy prediction in inter domain routing of broadband network con-nections with QoS requirements in the case of an integrated ATM and SDH net-working infrastructure. In their method, a subset-interactive autoregressive modelis used to predict link utilization levels, based on experience from both static trafficobservations as well as dynamic knowledge, acquired during the network’s opera-tion. Based on these, the shadow cost of allowing the connection through each fea-sible path is calculated, which is then used to select the “best” path. The shadowcost is calculated in such a way as to lead the network to states which exhibitthe lowest expected blocking probabilities in regard to information about user’sdemand - thus aiming to match network state with demand at all times.

Also, early results of a study for VP bandwidth allocation using evolutionaryprogramming techniques has been published by Pitsillides et al [PPS¸V97].

5 Call Level Control

In an ATM network, the main control tasks performed at the call level is con-nection admission control (CAC) and service rate allocated to each virtual pathconnections. Most of the studies reported in the literature have focused on solvingthe CAC problem.

5.1 Connection Admission

Connection Admission Control (CAC) is defined as a set of actions, performedat connection set-up phase, to determine whether or not the virtual path (VP) orvirtual channel (VC) requesting the connection can be accepted. The decision isbased on the connection’s anticipated traffic characteristics, the requested QoS,and the current state of the network. The anticipated traffic characteristics of theconnection are determined by a source traffic descriptor, and the user terminaldeclares these source traffic descriptor values to the network when the connectionset up is requested. If the request is accepted, network resources are implicitlyallocated to the connection.

To attain high utilization of VPs while meeting their QoS requirements, CACmust determine whether to accept a new VC by considering its anticipated trafficcharacteristics, the QoS requirements of existing VCs, the availability of the net-work resources, and current utilization of the links. There are many demanding


problems waiting to be solved for the development of an effective CAC algorithm,especially

– The statistical behavior of several sources of different types multiplexed on anATM link is difficult to predict. Therefore, deciding how to allocate resourcesfor multiple QoS requirements is hard to solve.

– Developing accurate analytical models to evaluate QoS services could be verydifficult.

Application of CI techniques appears to be appropriate, and not surprisingly, sev-eral researchers have attempted to solve the problem by using ANNs and fuzzylogic control techniques. Especially, learning and generalization capabilities ofANNs make them suitable for solving the CAC problem quite successfully.

Hiramatsu is one of the earliest researchers who have realized the potential ofANNs for solving ATM CAC problem [Hir90] and has also published his workon training techniques for ANN applications in ATM [Hir95]. Similarly, Ramalhoand Scharf [RS94] have used an ANN based method for learning the behavior ofthe traffic in an ATM link. Park and Lee [PL95b,PL95a] also have published theirwork on adaptive call admission control using feedforward ANNs.

Some other researchers concentrated on fuzzy logic based solutions. Ueharaand Hirota [UH97] have proposed a method based on estimation of the possibilitydistribution of cell loss ratio (CLR). These researchers have used fuzzy inferencefor the estimation of CLR of new connections. The mechanism operates this way:Each call request is placed into a transmission rate class depending on its declaredparameters. Then, by using the number of active connections for each class, aCLR estimation is made by the fuzzy inference engine. If the estimate exceeds therequired CLR, the connection request is rejected. The fuzzy sets representing thevalues of the fuzzy numbers of the rule base are shaped by a learning mechanismand observed CLR data which gives the scheme its adaptation capability.

An accurate model of traffic source requesting a connection is very importantfor estimating its future behavior. If the behavior of the traffic source can be pre-dicted precisely, connections can be admitted to the network with tighter safetymargins, leading to maximizing the resource utilization and consequently increas-ing the revenue. Scheffer and Kunicki have studied methods for accurate model-ing of voice and video sources by applying of fuzzy logic techniques, and basedon these models, for prediction of their behavior [SS93,SS94,SBK94]. They haveproposed a CAC scheme which uses a fuzzy logic based traffic prediction algo-rithm [SK96].

Early results of a study whose aim is the application of CI techniques to ATMCAC problem and development of a simulation testbed has been published byCzezowski [Cze98].

6 Cell Level Control

In an ATM network, cell level controls involve flow and congestion control, en-forcement of agreed traffic parameters, cell servicing discipline, traffic shaping


and control algorithms for cell switching and multiplexing. Some of these issueshave been studied quite extensively (such as traffic enforcement and congestioncontrol), while others virtually ignored by the CI research community.

6.1 Usage Parameter Control

Usage parameter control (UPC), or in other words, traffic enforcement or polic-ing, is a very important function in ATM networks. Its task is to ensure that trafficsources stay within the limits of the negotiated traffic parameters which are de-clared during the call setup phase. Traffic enforcement functions are performedby the network provider at the virtual circuit or virtual path level and correctivemeasures are taken if a traffic source does not stay within the declared limits. Themeasures could be as drastic as blocking the traffic source or could be less severesuch as selectively discarding the violating cells or tagging violating cells thatcould be discarded in downstream nodes if necessary.

The ideal UPC mechanism should have these desirable characteristics: accu-rate detection of any traffic situation violating the negotiated values, and separat-ing those connections from the ones that stay within the negotiated limits; fastresponse to violations; implementation simplicity and cost effectiveness. Design-ing a UPC mechanism encompassing these features could be a daunting task. Forexample, well studied mechanisms such as leaky bucket and window mechanismscannot achieve the ideal UPC characteristics but only provide a trade-off betweenthe above requirements.

Catania et al. [CFPP95,CFPP96a,CFPP96b] and Ascia et al. [ACF+97] haveproposed a UPC mechanism based on fuzzy logic control which displays charac-teristics close to ideal UPC, and have also implemented the algorithm as a VLSIchip. The mechanism they propose ensures that a bursty source conforms to itsagreed average cell rate. It is a window based control mechanism. It allows shortterm fluctuations of the source cell rate around a negotiated average value, as longas the source respects this value over the long term, and at the same time it is capa-ble of recognizing a violation immediately. The maximum number of cells whichare considered to be non-violating in a fixed period is dynamically updated by aset of fuzzy inference rules. The set of rules is shaped to guarantee transparency toa compliant source by assigning a credit value of allowable cells which is higherthan the negotiated value agreed at call set up phase. This credit value representsthe number of cells that the source can send during a particular transmit window.If a source has a high flow of traffic, the UPC intervenes to enforce a reductionof the bit rate of the source. To do so, it reduces the assigned credit to lower theallowable cell rate threshold and identifies any cells that exceed this threshold asviolating cells.

The parameters describing the behavior of the source consist of the averagenumber of cell arrivals per window since the start of the connection, the number ofcell arrivals in the last transmit window, and current value of the maximum allow-able cells that can be transmitted. By using these three parameters, the fuzzy UPCmechanism determines the value of the threshold to be used in the next transmis-sion window.


A UPC mechanism particularly designed for policing voice sources has beenproposed by Ndousse [Ndo94]. Since voice cells are characterized by a high degreeof burstiness, utilization of a classical control approach faces difficulties. Ndousseproposes an intelligent implementation of the leaky bucket cell rate control mech-anism which yields a lower cell drop rate than the leaky bucket algorithm undersimilar circumstances.

The fuzzy leaky bucket is implemented as a two-level fuzzy logic controller(FLC) by connecting three fuzzy associative memories (FAMs) [Kos92, pages299–338]. In the first level, there are two FAMs, each taking two input variablesand generating an output variable which is supplied to the second level FAM. Theoutput of the FLC determines the number of special tokens in the token buffer inthe next sampling interval. These special tokens are used to tag the violating cells.Therefore, depending on the availability of the network resources, violating cellsare not discarded straight away, but can have a chance of transmission. The FLCdynamically determines the number of special tokens allocated in the token bufferby monitoring the buffer occupancies and buffer growth rates in the token bufferand channel buffer allocated to the voice connection.

6.2 Flow, Congestion and Rate Control

It is generally accepted that the problem of network congestion control remainsa critical issue and a high priority. One could argue that network congestion is aproblem unlikely to disappear in the near future. Furthermore, congestion may be-come unmanageable unless effective, robust, and efficient methods for congestioncontrol are developed. This assertion is based on the fact that despite the vast re-search efforts, spanning a few decades, and the large number of different controlschemes proposed, there are still no universally acceptable congestion control so-lutions. As demand for multimedia (streaming) applications increases, it becomesincreasing important to ensure that these applications can co-exist with currentapplications.

In early stages of B-ISDN development, prevailing belief among the researchcommunity was, preventive (or, in other words open-loop) type congestion con-trol at the edge is necessary due to the large bandwidth delay product, and wouldbe sufficient for ATM networks. But, outcomes of subsequent studies have shownthat, because of the variety of the traffic to be supported in B-ISDN networks,open-loop congestion and flow control is rendered to be ineffective in ATM net-works. Today, the shift is towards closed-loop congestion controls (within the net-work).

In ATM networks, depending on the nature of the traffic sources, the closed-loop congestion control issue can be approached in two ways:

– For delay tolerant traffic, which basically comprises of TCP/IP type traffic,switches can send feedback signals to the sources leading them to reduce therate at which they release cells to the network. Then excess traffic is queued atthe traffic source and consequently delayed.


The ATM Forum has introduced a service category, called available bit rate(ABR), in order to allocate bandwidth dynamically within an ATM network,while simultaneously minimizing the cell losses, and has selected a feedbackcontrol framework to achieve these aims [ATM96]. The framework allowsdownstream nodes or intermediate ATM switches to periodically send infor-mation to the traffic sources relating to maximum cell rates that they can han-dle. The cell rate information is carried by a stream of resource management(RM) cells generated by the traffic sources and relayed back to the sourcesby the destination end systems. During their round-trip, while these cells passthrough the switching nodes, the cell rate information contents of these cellsare dynamically updated by these intermediate systems. For the calculation ofrate, several algorithms have been proposed.

– On the other hand, since delay tolerance ofvideo/voice trafficis very low, con-gestion control is performed by sending coding rate signals to these types ofsources. In the presence of congestion, the sources can vary their coding rate,and so reduce the frequency of cells generated by using this feedback informa-tion. Lower coding rate inevitably reduces the image/sound quality at the re-ceiver but network utilization and quality of offered service rate are maintainedat higher levels by minimizing the cell losses and delays due to congestion.

Following paragraphs summarize the research utilizing CI techniques for imple-mentation of congestion and rate control algorithms in ATM networks.

Tarraf, Habib and Saadawi [TH94,THS95,THS95b] have investigated exten-sively how ANNs can be used to solve many of the problems encountered in thedevelopment of coherent traffic control strategies in ATM networks. In [THS95b]they present congestion control schemes for ATM networks. Also, they investigatea reinforcement-learning based neural network for congestion control in ATM net-works [THS95a]. Liu and Douligeris have published the results of a comparisonstudy on the performance of static and adaptive feedback congestion controllerswhich uses ANNs [LD95].

Huang and Yan [HY96] use a recurrent neural network for the dynamic con-trol of communication systems, particularly dynamic congestion control in ATMnetworks. Mhrvar and Le-Ngoc [MLN95] apply a neural network scheme for con-gestion control in packet switch OBP satellite systems.

Cheng and Chang [CC94,CC96] have devised a fuzzy control system whichcombines CAC and a feedback mechanism for controling congestion. The mecha-nism sends back coding rate control signals to video sources, and congestion con-trol signals to data sources. In this scheme, fuzzy sets representing the linguisticvalues are selected by evolutionary techniques. The system has the ability to adjustthe cell transmission rate of the video sources, and subsequently traffic density atthe switches, it can still maintain QoS for the connections.

Pitsillides et al. [PS¸R95,PSR97], have proposed the Fuzzy Explicit Rate Mark-ing (FERM) algorithm, and analyzed its performance regarding fairness, respon-siveness, resource utilization and cell loss in LAN and WAN environments. FERMoperates on switching nodes and by periodically monitoring the buffer utilizationand queue growth rate, determines a cell rate which is used to update the max-


imum cell rate information carried by the RM cells passing through the virtualconnection.

Douligeris and Develekos [DD95] have studied a FLC which is based on theshort term observation of the network status to predict the near future cell dis-carding behavior of the switching nodes. This prediction is then fed back to trafficshapers in the sources to minimize cell losses.

Jensen [Jen94] has proposed a three-step FLC for controlling the transmissionrate of sources to protect links against overload in the case of connections exceed-ing their negotiated traffic parameters. The scheme operates as follows: At the calladmission stage a service dependent priority is assigned to each connection. Thispriority is kept as a fixed value for the whole life time of the connection. Also, inthe switching node, a certain buffer capacity is allocated to the connection. TheFLC generates the cell service rate control signals for each buffers. Input parame-ters of the FLC are: a) allocated priority level; b) current buffer occupancy level;c) bandwidth utilization at the output link of the node; and d) the difference be-tween the effective bandwidth at which the source is transmitting the cells and thedeclared bandwidth negotiated during the call set up stage.

Hu and Petr [HPB96] have studied an adaptive traffic controller based onSugeno’s self-tuning fuzzy control methods.

6.3 Cell Switching and Multiplexing

In an ATM network, cell queuing is required to alleviate congestion at switchingnodes. Congestion occurs when multiple cells simultaneously attempt to access anoutput link in a switch. Cell queuing can be arranged either by placing buffers atinput ports (called input queuing), or by placing cell buffers at the output ports(called output queuing). Output queuing yields better performance in terms of celldelay and throughput, but computationally more demanding to operate than inputqueuing. On the other hand, in input queuing, if the head-of-line blocking problemcan be solved, comparable performance can be achieved. One way of solving thisproblem is to employ a mechanism called bypass queuing. When bypass queuingis used, a controller module schedules the cells in an optimal fashion to enhancethe switch throughput. Additionally, cells can be dispatched optimally if they areassigned priorities, with higher priorities assigned to real-time traffic such as voiceand video (due to rigid delay requirements) and lower priorities assigned to datatraffic, by an intelligent scheduling mechanism.

Liu and Douligeris [LD96] have proposed a fuzzy scheduler to optimize thecell servicing sequences to reduce cell losses. In their mechanism, each traffic classin the switch has its own portion of the dedicated buffer and a fuzzy schedulingalgorithm manages the server. Park and Lee [PL95b,PL95a] have also worked onoptimal scheduling and published their study on application of recurrent ANNs tothis problem.


7 Discussion and Concluding Remarks

Research on applications of CI in telecommunication systems, particularly in ATMnetworks, is being pursued by an active research community, and methods arebeing developed simultaneously. However, unlike consumer applications, there areno commercially deployed applications as yet. The reasons could be

– the lack of comprehensive performance comparisons between the best tradi-tional techniques and the ones involving CI applications. The comparisonsperformed in the research studies usually have been undertaken in simplifiednetworking scenarios, and testing on real hardware has not been undertakenyet except for some partial implementations such as in [ACF+97]. Before theapplications of CI techniques to high speed communication networks becomesaccepted, it will be necessary to place a greater emphasis on rigorously demon-strating the advantages to be gained, and that is an area we strongly recom-mend.

– the reluctance to adopt new technologies by telecommunications companiesand equipment manufacturers. This issue is closely related to the lack of com-prehensive performance studies mentioned above.

As a final note, we would strongly encourage a thorough study of an integratedcontrol structure implementing a multilevel control strategy spanning network, calland cell levels. The integration can be achieved by appropriate design of eachindividual strategy in a new multilevel fuzzy logic structure, and/or the integrationof existing, or separately designed strategies, with their integration achieved via afuzzy logic based supervisor, taking care of the overall “goodness” of the networkand handling any interactions among the control functions, at the same or differentlevels.

References

[ACF+97] G. Ascia, V. Catania, G. Ficili, S. Palazzo, and D. Panno. A VLSI fuzzy expertsystem for real-time traffic control in ATM networks.IEEE Transactions on FuzzySystems, 5(1):20–31, February 1997.

[AD97] E. Aboelela and C. Douligeris. Routing in multimetric networks using a fuzzy linkcost. InProceedings of the International Symposium on Computers & Communi-cations ISCC’97, pages 397–401, Alexandria, Egypt, June 1997. IEEE ComputerSociety.

[ASV98] A. F. Atlasis, M. P. Saltouros, and A. Vasilakos. On the use of a stochastic learn-ing algorithm to the ATM routing problem: A methodology.Computer Commu-nications, (21):538–546, 1998.

[ATM96] Traffic management specification version 4.0. Technical Report AF-TM-0056.000, April 1996.

[AV98] A. F. Atlasis and A. Vasilakos. LB-SELA: Rate-based access control for ATMnetworks.Computer Networks and ISDN Systems, (30):963–980, 1998.

[AVL00] A. Atlasis, A. Vasilakos, and N. Loukas. The use of learning algorithms in ATMnetworks call admission control problem:a methodology.Computer Networks,2000. (to appear).


[Bez92] J. C. Bezdek. On the relationship between neural networks, pattern recognitionand intelligence. International Journal of Approximate Reasoning, 6:85–107,1992.

[Bez94] J. C. Bezdek. What is computational intelligence? In J. M.Zurada, R. J. MarksII, and C. J. Robinson, editors,Computational IntelligenceImitating Life, pages1–12. IEEE Press, 1994.

[CC94] C. Chang and R. Cheng. Traffic control in an ATM network using fuzzy settheory. InProceedings of the IEEE INFOCOM’94 Conference, pages 1200–1207,Toronto, Canada, June 1994. IEEE Communications Society.

[CC96] R-G. Cheng and C-J. Chang. Design of a fuzzy traffic controller for ATM net-works. IEEE/ACM Transactions on Networking, 4(3):460–469, June 1996.

[CFPP95] V. Catania, G. Ficili, S. Palazzo, and D. Panno. A fuzzy expert system for usageparameter control in ATM networks. In ProcGlobecom95 [Pro95b], pages 1338–1342.

[CFPP96a] V. Catania, G. Ficili, S. Palazzo, and D. Panno. A comparative analysis offuzzy versus conventional policing mechanisms for ATM networks.IEEE/ACMTransactions on Networking, 4(3):449–459, June 1996.

[CFPP96b] V. Catania, G. Ficili, S. Palazzo, and D. Panno. Using fuzzy logic in ATMsource traffic control: Lessons and perspectives.IEEE Communications Maga-zine, pages 70–81, November 1996.

[CL94] T. M. Chen and S. S. Liu. Management and control functions in ATM switchingsystems.IEEE Network, pages 27–40, July 1994.

[Cze98] P. Czezowski. Computational intelligence in ATM call admission control. InProceedings of the IEEE Ph.D. Students Conference (GRADCON’98), Winnipeg,Canada, 1998. IEEE.

[DD95] C. Douligeris and G. Develekos. A fuzzy logic approach to congestion control inATM networks. In ProcIcc95 [Pro95a], pages 1969–1973.

[DD97] C. Douligeris and G. Develekos. Neuro-fuzzy control in ATM networks.IEEECommunications Magazine, pages 154–162, May 1997.

[GRSC98] S. Ghosh, Q. Razouqi, H. J. Schumacher, and A. Celmins. A survey of re-cent advances in fuzzy logic in telecommunications networks and new challenges.IEEE Transactions on Fuzzy Systems, 6(3):443–447, August 1998.

[Hab96] I. W. Habib. Applications of neurocomputing in traffic management of ATMnetworks.Proceedings of the IEEE, 84(10):1430–1441, October 1996.

[Hir90] A. Hiramatsu. ATM communications network control by neural networks.IEEETransactions on Neural Networks, 1(1):122–130, March 1990.

[Hir95] A. Hiramatsu. Training techniques for neural network applications in ATM.IEEECommunications Magazine, 33(10):58, 63–67, October 1995.

[HPB96] Q. Hu, D. W. Petr, and C. Braun. Self-tuning fuzzy traffic rate control for ATMnetworks. In ProcIcc96 [Pro96], pages 424–428.

[HY96] Y. Huang and W. Yan. Dynamic control of communication systems based onsimple recurrent neural networks. InProceedings of the IEEE 1996 NationalAerospace and Electronics Conference, volume 1, pages 254–258. IEEE, 1996.

[Jen94] D. Jensen. B-ISDN network management by a fuzzy logic controller. InProceed-ings of the IEEE Global Telecommunications Conference GLOBECOM’94, pages799–804. IEEE Communications Society, 1994.

[KMV00] I. Kassotakis, A. Markaki, and A. Vasilakos. A hybrid genetic approach forchannel reuse in multi-access networks.IEEE Journal on Selected Areas in Com-munications, February 2000.

[Kos92] B. Kosko.Neural Networks and Fuzzy Systems: A Dynamical Systems Approachto Machine Intelligence. Prentice Hall, 1992.


[LD95] Y-C. Liu and C. Douligeris. Static vs. adaptive feedback congestion controller forATM networks. In ProcGlobecom95 [Pro95b].

[LD96] Y-C. Liu and C. Douligeris. Nested threshold cell discarding with dedicatedbuffers and fuzzy scheduling. In ProcIcc96 [Pro96], pages 429–433.

[MLN95] H. R. Mehrvar and T. Le-Ngoc. ANN approach for congestion control in packetswitch OBP satellite. In ProcIcc95 [Pro95a], pages 810–814.

[NCGA95] E. Nordstrom, J. Carlstr¨om, O. Gallmo, and L. Asplund. Neural networksfor adaptive traffic control in ATM networks.IEEE Communications Magazine,33(10):43–49, October 1995.

[Ndo94] T. D. Ndousse. Fuzzy neural control of voice cells in ATM networks.IEEE Jour-nal on Selected Areas in Communications, 12(9):1488–1494, December 1994.

[Ped98] W. Pedrycz.Computational IntelligenceAn Introduction. CRC Press, 1998.[Pit93] A. Pitsillides. Control Structures and Techniques for Broadband-ISDN Commu-

nication Systems. PhD thesis, School of Electrical Engineering, Swinburne Uni-versity of Technology, Melbourne, Australia, 1993.

[PL95a] Y-K. Park and G. Lee. Applications of neural networks in high-speed communi-cation networks.IEEE Communications Magazine, 33(10):68–74, October 1995.

[PL95b] Y-K. Park and G. Lee. Intelligent congestion control in ATM networks. InPro-

ceedings of the5th IEEE Computer Society Workshop on Future Trends of Dis-tributed Computing Systems, pages 369–375. IEEE, 1995.

[PPS+00] A. Pitsillides, C. Pattichis, G. Stylianou, A. S¸ekercioglu, and A. Vasilakos.Bandwidth allocation for virtual paths (BAVP): Investigation of the performanceof classical constrained and genetic based optimization techniques. InProceed-ings of the IEEE INFOCOM’00 Conference, Tel Aviv, Israel, March 2000. IEEECommunications Society.

[PPSV97] A. Pitsillides, C. Pattichis, Y. A. S¸ekercioglu, and A. Vasilakos. Bandwidthallocation for virtual paths using evolutionary programming (EP-BAVP). InPro-ceedings of the International Conference on Telecommunications ICT’97, pages1163–1168, Melbourne, Australia, April 1997. Monash University.

[Pro95a] IEEE Communications Society.Proceedings of the 1995 IEEE InternationalConference on Communications ICC’95, Washington, USA, 1995.

[Pro95b] IEEE Communications Society.Proceedings of the IEEE Global Telecommuni-cations Conference GLOBECOM’95, Singapore, 1995.

[Pro96] IEEE Communications Society.Proceedings of the 1996 IEEE International Con-ference on Communications ICC’96, Dallas, Texas, USA, 1996.

[PSR95] A. Pitsillides, Y. A. Sekercioglu, and G. Ramamurthy. Fuzzy Backward Con-gestion Notification (FBCN) congestion control in Asynchronous Transfer Mode(ATM). In ProcGlobecom95 [Pro95b], pages 280–285.

[PSR97] A. Pitsillides, Y. A. Sekercioglu, and G. Ramamurthy. Effective control of trafficflow in ATM networks using fuzzy logic based explicit rate marking (FERM).IEEE Journal on Selected Areas in Communications, 15(2):209–225, February1997.

[PV00] W. Pedrycz and A. Vasilakos, editors.Computational Intelligence in Telecommu-nications Networks. CRC Press, 2000.

[RS94] M. F. Ramalho and E. Scharf. Fuzzy logic based techniques for connection ad-mission control in ATM networks. InProceedings of the11th IEE TeletrafficSymposium, pages 12A/1–12A/8. IEE, March 1994.

[SBK94] M. F. Scheffer, J. J. P. Beneke, and J. S. Kunicki. Fuzzy modelling and predictionof network traffic fluctuations. InProceedings of COSIG’94, South Africa, 1994.

[SF95] J. A. Smith and M. Fry. Artificial intelligence in network management.AustralianJournal of Intelligent Information Processing Systems, Autumn:53–62, 1995.


[SK96] M. F. Scheffer and J. S Kunicki. Fuzzy adaptive traffic enforcement for ATMnetworks. InProceedings of the4th IEEE AFRICON Conference, pages 1047–1050. IEEE, 1996.

[SS93] M. F. Scheffer and I. S. Shaw. VLSI hardware realization of self-learning re-cursive fuzzy model for dynamic systems. InProceedings of IFAC12th WorldCongress, Sydney, Australia, 1993.

[SS94] M. F. Scheffer and I. S. Shaw. Application using VLSI hardware realization ofself-learning recursive fuzzy model. InProceedings of SICICA’94, Budapest,Hungary, 1994.

[TH94] A. A. Tarraf and I. W. Habib. A novel neural network traffic enforcement mech-anism for ATM networks.IEEE Journal on Selected Areas in Communications,12(6):1088–1096, August 1994.

[THS95] A. A. Tarraf, I. W. Habib, and T. N. Saadawi. Intelligent traffic control for ATMbroadband networks.IEEE Communications Magazine, 33(10):76–82, October195.

[THS95a] A. A. Tarraf, I. W. Habib, and T. N. Saadawi. Congestion control mechanismfor ATM networks using neural networks. In ProcIcc95 [Pro95a], pages 206–210.

[THS95b] A. A. Tarraf, I. W. Habib, and T. N. Saadawi. Reinforcement learning-basedneural network congestion controller. InProceedings of the Military Communi-cations Conference MILCOM’95, volume 2, pages 668–672. IEEE, 1995.

[UH97] K. Uehara and K. Hirota. Fuzzy connection admission control for ATM networksbased on possibility distribution of cell loss ratio.IEEE Journal on Selected Areasin Communications, 15(2):179–190, February 1997.

[VAP97] A. Vasilakos, K. Anagnostakis, and A. Pitsillides. An evolutionary fuzzy algo-rithm for QoS and policy based inter-domain routing in heterogeneous ATM andSDH/SONET networks. InProceedings of the5th European Congress on Intelli-gent Techniques and Soft Computing EUFIT’97, Aachen, Germany, 1997.

[VRA+98] A. Vasilakos, C. Ricudis, K. Anagnostakis, W. Pedrycz, and A. Pitsillides.Evolutionary-fuzzy prediction for strategic QoS routing in broadband networks.In Proceedings of the IEEE World Congress on Computational IntelligenceWCCI’98, pages 1488–1493, Anchorage, Alaska, USA, 1998.

[Yon90] S. Yoneda. Broadband-ISDN ATM layer management: Operations, administra-tion, and maintenance considerations.IEEE Network, pages 31–35, May 1990.

Documents

Computational intelligence in management of ATM networks