Efficient and stateless deployment of VoIP services

Abstract

A critical issue for the convergence of telephone networks towards an all IP environment is the provisioning of an adequate QoS level to the telephone traffic. Following the guidelines detailed in RFC 2990, we recently proposed an admission control function, named Gauge & Gate Reservation with Independent Probing (GRIP), through which we are able to provide QoS guarantees by means of stateless procedures compliant within the Differentiated Service (DiffServ) paradigm, which is the de facto standard for QoS provisioning in Internet. The main goal of this paper is to present a specific implementation of GRIP optimized for supporting voice over IP (VoIP) services within a DiffServ cloud. In particular, we propose and analyze a simple measurement-based call admission control procedure for carrying VoIP traffic in a DiffServ environment. We assume voice sources are leaky bucket controlled with known leaky bucket parameters. We analyze the leaky bucket to derive the distribution of the times over which voice sources generate traffic at peak and regulated rates. We use the results of that analysis, which provides complete knowledge of the traffic generation characteristics of a single source, to derive a Markov chain model that characterizes the voice packet generation process of a single voice source. We then derive a simple stateless threshold policy for admission control; that is, our policy is of the form if, at admission request time, $x⩽x_{\text{threshold}}$, admit the call; else reject the call. Finally, we show, via simulation, that the QoS achieved using the threshold-based admission control policy is very close to that achieved by a stateful admission control policy even at very high traffic utilization.

Introduction

The convergence of telephone networks towards an all IP environment would bring about many well-known advantages including more efficient support of actual telephone services [6] and the potential to offer a richer collection of services such as sophisticated DNS functionality, customized mobility support, personalized, and graphically rich services. In fact, a pervasive deployment of VoIP could serve as a catalyst to develop a much larger class of services that use the full functionality of protocols such as Session Initiation Protocol (SIP) [1], [2]. On the other hand, providing high-quality VoIP call involves many issues and requires some engineering trade-offs, as thoroughly discussed in [3], [6], [5]. One of the most challenging is to provide the required QoS to voice calls [5], [7] by means of efficient schemes. In this regard, IETF pursued two well-known general approaches to provide QoS over the Internet [10], [14]: Integrated Services (IntServ, [12], [11]) and Differentiated Services (DiffServ, [13], [8]). The stateful IntServ has a greater level of accuracy and a finer level of granularity; however, there are several reasons for not using IntServ with RSVP for IP telephony [3]. The stateless DiffServ possesses excellent scaling properties; however, it lacks a standardized admission control scheme and, upon overload in a given service class, degradation of service can occur. In fact, DiffServ has no topology-aware admission control mechanism, and the IETF DiffServ Working Group has not recommended a mechanism for limiting the volume of VoIP calls to control the quality of service.

Some have argued that QoS issues are best handled via over-provisioning [7] of network resources, rather than by means of (possibly) complex control schemes. But, not only this seems a short term solution [43], but also with over-provisioning, if calls are accepted without limit, the QoS for calls cannot be guaranteed because the total bandwidth required could exceed the network capacity [35]. A trade-off between complex admission control procedures (as in IntServ) and plain over-provisioning is in order. The focus of this paper is to propose a solution for viable, practical, feasible admission control mechanisms over DiffServ for VoIP traffic.

Quite surprisingly, existing, standard per-hop behaviors (PHBs) classes in DiffServ [13], suitably coupled with the functionality of existing routers, are semantically and implicitly capable of supporting per-flow admission control. This is the approach that we followed to define an “admission control function which can determine whether to admit a service differentiated flow along the nominated network path [10]”. This solution, named GRIP (Gauge & Gate Reservation with Independent Probing, [21]), can provide strict QoS guarantees by means of stateless, measurement-based, DiffServ-compliant procedures. Our type of approach is not in contrast with IETF guidelines; on the contrary, it is suggested as a viable option in [10].

In order to assure performance, we assume that, as in both IntServ and DiffServ, individual traffic flows are regulated at the edge of the network by using, for example, dual leaky buckets (DLBs) [15], [16]. It has been noted in the literature that a regulator provides a standard interface and simplifies the definition of standardized admission control criteria [18], [19], especially when traffic models are analytically intractable (see, e.g., large Markovian models in [17]). In addition, a regulator is a common and low-cost way to specify service level agreements between service provider and user.

Our goal in this work is to optimize the GRIP admission control function for voice traffic. Our working plan is the following:

• We select a specific coding scheme (namely, the G.726 [3], although our approach is applicable to whatever codec), since, in general, different application developers and carriers define or choose different voice codecs.¹

• We analytically model the output of a DLB fed by our selected codec through markovian models. We propose a number of alternatives and then discuss about their suitability and performance in representing the regulated voice traffic. This is the first contribution of our work.

• We derive a novel measurement-based criterion to admit VoIP calls by using the GRIP scheme, optimized for the selected voice codec and exploiting the developed traffic model. Specifically, each admission decision point maintains a measure of the traffic currently carried, say x. The admission criterion is then of the form if, at admission request time, $x⩽x_{\text{threshold}}$, admit the call; else reject the call. This is the second contribution of this paper.

• Finally, we demonstrate via simulation the effectiveness of our simple technique in delivering high-quality service.

The basic idea of GRIP has already been presented in [21]. In that previous paper, it is assumed that the only knowledge about traffic sources consists of their traffic descriptors (T-SPEC, see also [12]). Such a “generic” description implies that the technique is not optimized for a specific application. The novelty of this paper is to show that, when the traffic model is known, it is possible to increase the resource utilization coefficient, while guaranteeing the same minimum performance level delivered to end users.

We point out that, although we focus on VoIP, our approach is applicable to other real time services. We chose VoIP not only for the reasons discussed above (importance and interest) but also because this case is may be the simplest one to work with. In fact, performance requirements and traffic models of telephone services are very well know and understood.

As for the organization of the paper, the next section provides a brief introduction to the GRIP mechanism and related work on admission control. In Section 3, we propose a model for a DLB-regulated voice traffic. In Section 4, we define a decision criterion to be used to admit a new voice call that makes use of the previously defined model. In Section 5, we present a performance evaluation of the proposed scheme, performed through simulations, and finally, in Section 6, we draw our conclusions.