A fuzzy algorithm for combined control of traffic parameters: assessment and key issues

Abstract

One of the most critical functions in the management of high-speed networks using the ATM technique is that of “policing”. It has the task of ensuring that each traffic source complies with the traffic parameter negotiated in order to avoid network congestion. In this article we focus on a dual aim. The first is to analyse the performance of a policing mechanism based on Fuzzy Logic for the control of packetized voice sources. The results obtained show an excellent selectivity, close to that of an ideal policer, and a responsiveness, assessed by the combined measures of reaction time and rise time, which is decidedly better than that of the Leaky Bucket. The second aim addresses the capacity of the fuzzy policer in controlling not only the long-term Average Cell Rate but also the Sustainable Cell Rate. Analysis by simulation proves that fuzzy policer is extremely flexible in the combined control of these two traffic descriptors.

Introduction

The flexibility attained in Asynchronous Transfer Mode (ATM) networks to accommodate a wide variety of traffic sources raises challenging new issues on resource management [1], [2], [3]. Preventing the network from becoming congested, achieving network performance objectives and optimizing the use of network resources requires the following traffic controls: Connection Acceptance Control (CAC) and Usage Parameter Control (UPC), the latter is also referred to as flow enforcement or policing. The CAC decides, during the call set-up phase, to accept a new connection on a link and, consequently, to allocate a certain portion of bandwidth to it. This decision is based on the connection’s anticipated traffic characteristics, the required Quality of Service (QoS), and the current network load. To protect network resources from malicious as well as unintentional misbehaviour which can affect the QoS of already established connections, the UPC function, during the entire phase of the call, monitors and controls the offered traffic, by detecting violations of negotiated parameters. The action usually taken by a policer on cells detected as violating is to discard or mark them.

In order to define an efficient policing mechanism, a first issue is identifying the traffic parameters which best characterize the behaviour of a source. The difficulty lies in the fact that the sources to be characterized have different statistical properties as they range from video to data services, and it is necessary to define parameters that can be monitored during the call [4], [5], [6]. A traffic parameter contributing to a source traffic descriptor should be understandable by the user, of significant use in resource allocation, and enforceable by the network provider through the UPC. Ordinarily, UPC is performed for each traffic parameter in a source traffic descriptor. For example, if the source traffic descriptor consists of the peak bit rate and the mean bit rate, UPC is necessary for both of them.

Another key issue is defining a traffic enforcement mechanism which will be efficient in coping with the conflicting requirements of ideal flow enforcement: selectivity, that is, the capability of detecting any illegal traffic situation and transparency for connections that respect the parameter values negotiated, on whose cells no policing action need be taken; and responsiveness, that is, low response time to parameter violations.

In literature several mechanisms such as the Leaky Bucket (LB) and window mechanisms have been proposed to police Average Cell Rate (ACR) seen as the average calculated over the whole duration of the connection. An extensive review and assessment of these mechanisms can be found in Refs. [7], [8], [9]; for none of them has it been possible to achieve a satisfactory tradeoff between the aforementioned conflicting requirements. This is essentially as a result of two causes. The first is that enforcement of this parameter is very hard, since short-term statistical fluctuations of the source traffic are admissible as long as the source respects the average value negotiated, λ_n, in the long term. The second is related to the fact that the logic on which these mechanisms are based is of the traditional kind. This makes it difficult to compile the control know-how, probably because of nonlinearity, to the time variant behaviour of the system, or to the fact that the measurements available are of poor quality. In order to overcome the limits that traditional mechanisms seem to present several research groups have explored alternative solutions based on artificial intelligence techniques, namely artificial neural networks [10] and fuzzy logic [11], [12].

Fuzzy logic [13], [14], [15] is an important tool for formalizing processes of approximate reasoning in which the knowledge base can be acquired from a human expert. To reproduce the concepts expressed in natural language, fuzzy logic replaces true and false with continuous membership values ranging from zero to one. This allows the processing of linguistic concepts such as “small”, “big”, “low”, “high” or “approximately”, which can be expressed in the fuzzy inferential rules that describe the control algorithm. Thanks to these inherent features, fuzzy logic proves to be efficient in controlling real-time processes which are too complex to be represented by exact mathematical models.

Exploiting the advantages offered by fuzzy logic in [11] we formalized the control actions of the policer translating the know-how of an expert in the field into fuzzy rules. The fuzzy policer obtained was analysed using the same bursty sources as those examined in [8] and compared with conventional mechanisms such as the LB and the Exponential Weighted Moving Average (EWMA). The results obtained showed that fuzzy logic, when applied to policing, is extremely promising on account of both the performance obtainable and the simplicity of the control algorithm.

In this article, we focus on a dual aim. The first is to analyse extensively the behaviour of the Fuzzy Policer (FP) in controlling a real source, namely a packetized voice source, evaluating its selectivity and dynamic response. The latter is evaluated by means of a new combined measure of reaction time and rise time. The performance of the FP is compared with single and double [3] LBs. The second aim is to see whether the FP is capable of controlling not only the ACR but also the Sustainable Cell Rate (SCR) defined by ITU in [16]. Some of the main issues linked to use of the ACR as a traffic descriptor parameter are discussed and the need is shown for combined control of ACR and SCR. We then present a simulation analysis from which it emerges that the FP, unlike conventional mechanisms, is extremely flexible in the combined control of ACR and SCR, guaranteeing an optimal level of performance.

The article is organized as follows: in Section. 2, after a brief overview on fuzzy logic, we outline the model of the fuzzy policing mechanism; in Section. 3 we assess the efficiency of the fuzzy policer in the control of packetized voice sources; in Section. 4 we address some of the issues involved in use of the ACR as a descriptor of source behaviour and propose the FP for the combined control of ACR and SCR. Finally in Section. 5 some conclusions are drawn.