An All IP DiffServ Architecture for 3G/4G Wireless Communications 

Context in source publication

Context 1

... IETF has introduced two main QoS frameworks IntServ [1] and DiffServ [ 2] to provide predictable and controllable behavior of IP networks. IntServ focuses on supporting individual applications by providing an architecture requiring per-flow traffic at every hop along an application's end-to-end path. The Resource Reservation Protocol (RSVP) is used to reserve resources in routers within an IS domain to provide particular QoS levels to different flows. As a counterpoint to the relative complexity and end-to-end nature of IntServ, the DiffServ domain does not reserve network resources on a per-flow basis, traffic is instead classified into a number of traffic groups. Each group is labeled appropriately by a particular value called Differentiated Services Code Point (DSCP) and based on the DSCP value, each group is then treated independently by the DiffServ domain. The dramatic increase in demand for wireless Internet access has lead to the introduction of new wireless architectures and systems including 3G, Wi-Fi and WiMAX. It is a likely development that mobile terminals will increasingly having the capability to access many of these wireless networks types. Because of the scalable class-based traffic management mechanism without using per-flow resource reservations, an all-IP DiffServ platform seems to be the most promising architecture to interwork the heterogeneous wireless access networks and the Internet to provide seamless global roaming and broadband access to the end-user [3- 4]. The domain-based resource management feature of DiffServ makes it the most suitable platform for interconnecting heterogeneous wireless access networks because each domain can freely choose whatever policies are proper for internal resource management as long as its Service Level Agreements (SLAs) are met with neighboring domains. In this paper, we investigate an all-IP DiffServ architecture (shown in Fig. 1 in the Appendix B) for interconnecting heterogeneous wireless access networks as given in [5]. In this all-IP DiffServ architecture, a number of nearby radio access networks (RANs) having the same interface are grouped into a wireless DiffServ domain, and all the domains are connected through a DiffServ Internet backbone to provide end-to-end Internet services to the mobile station (MS). It is assumed that in each DiffServ wireless domain, all network elements of RAN are enhanced to fulfill the functionality of a DiffServ IP router. The gateway and base stations are edge routers of the domain and connected through core routers [5]. Further, the gateway is the interface to the Internet backbone. For example, GPRS support node (GGSN) is the gateway of the UMTS domain to the external DiffServ Internet; similarly, PDSN is the gateway of the domain to the external DiffServ Internet in CDMA2000. In the gateway, SLAs are negotiated to specify the resources allocated by Internet Service Provider flowing from/into the domain. The gateway conditions the aggregate traffic for each service class according to SLA resource commitments. All DiffServ routers use different queueing and scheduling algorithms, to provide differentiated classes of services. Moreover, it has been shown that wireless data traffic exhibits self- similarity and long-range dependency [6, 7, 8, 9]. While taking into account a self-similar nature of multiservice traffic, it is not mundane to build tight bound SLAs between heterogeneous QoS domains. To offer realistic SLAs based on tight bound QoS parameters, between a wireless DiffServ domain gateway (for example GGSN) and the DiffServ Internet backbone, the present study focuses on the performance evaluation of two different queueing schemes (PQ and LLQ) in a DiffServ domain with multiple classes of self- similar input traffic. The derived QoS parameters in terms of expected queue length, packet delay and packet loss rate forms the basis on which to build realistic SLAs and ultimately provide support to interwork heterogeneous wireless access networks. This in turn is necessary to provide seamless global roaming, fast handoff and end-to-end QoS to the end-user. 3GPP has defined four QoS classes for UMTS; (1) Conversational, (2) Interactive, (3) Streaming and (4) Background. Traffic is classified and ordered on the basis of relative delay sensitivity [10]. On the other hand, DiffServ defines Expedited Forwarding (EF) per-hop behavior for premium service and the Assured Forwarding (AF) PHB for assured service in addition to the classical best effort class [11]. To extend IP services to the wireless domain, the UMTS QoS classes must be mapped to the DiffServ classes. According to 3GPP, UMTS-to-IP QoS mapping is performed by a translation function in the GGSN that classifies each UMTS packet flow and maps it to a suitable IP QoS [12]. Normally, the conversational class can be mapped to EF PHB for very low-delay and low-loss service, streaming and interactive traffic to the AF PHB and background traffic to best- effort service [5]. Under such mapping, we present a model based on a G/M/1 queueing system by considering three different classes of self-similar traffic input and analyze it on the basis of PQ and LLQ and bring the following contributions to the wireless traffic modeling; • We present closed form expressions of packet delay and PLR for different classes under PQ and LLQ disciplines. • We build the Markov chain for both systems. • We develop a comprehensive discrete-event simulator for a G/M/1 queueing system in order to understand and evaluate the QoS behavior of self-similar traffic. The simulation study produces performance evaluation results of multiple classes of input traffic in terms of expected queue length, packet delay and packet loss rate. • A traffic generator is also developed to realize our self- similar traffic model. • We implement a Cisco-router based test bed, which serves to experimentally validate the simulation results. The results obtained from the two different queuing schemes (PQ and LLQ) provide a foundation for better resource allocation to different traffic classes based on different QoS parameters e.g. delay, queue length and packet loss rate. The rest of the paper is organized as follows. Section 3 and 4 are devoted to explaining the self-similar traffic model with multiple classes and the formulation of Embedded Markov Chain along with the derivation of packet delays and PLR. In Section 5, we present simulation and test-bed results. Section 6 gives an overview of the related work. Finally, conclusion and future work is given in Section 7. We consider the self-similar traffic model studied in [13] which captures the dynamics of packet generation while accounting for the scaling properties of the traffic in telecommunication networks. It has similarities to on/off processes, in particular to its variation N-Burst model studied in [14] where packets are incorporated. The advantage of our traffic model is that we have already computed the induced packet interarrival distribution both for single and multiple classes of input traffic. In general, such analytical computations are not possible for other self-similar models for which only approximate or asymptotical queueing results are available. In [13], the traffic is found by aggregating the number of packets generated by several sources. Each source initiates a session with a heavy-tailed distribution, in particular a Pareto distribution whose density is given by g ( r ) = δ b δ r − δ − 1 , r > b , where δ is related to the Hurst parameter by H = ( 3 − δ ) / 2 . The sessions arrive according to a Poisson process with rate λ and the packets arrive according to a Poisson process with rate α , locally, over each session. In the framework of a Poisson point process, the model represents an infinite number of potential sources. For each class, the traffic Y ( t ) measured as the total number of packets injected in [0 , t ] is found by where i denotes the local Poisson process over session i , and R i , S i denote the duration and the arrival time of session i , respectively. Hence, Y ( t ) corresponds to the sum of packets generated by all sessions initiated in [0, t ] until the session expires if that happens before t , and until t if it does not. The increments of traffic input are stationary which is ensured by an indefinitely ongoing arrival process. The traffic model Y is long-range dependent and almost second-order self-similar; the auto- covariance function of its increments is that of fractional Gaussian noise. Three different heavy traffic limits are possible depending on the rate of increase in the traffic parameters [13, 15]. Two of these limits are well known self-similar processes, fractional Brownian motion and Levy process, which do not account for packet dynamics in particular. In the present study, we exploit the traffic model to represent different classes of traffic streams. Each stream has its own parameters and is independent from the other. The packet sizes are assumed to be fixed because each queue corresponds to a certain type of application where the packets have fixed size or at least fixed service time distribution. The interarrival time distribution for a single class has been stated recently in [16, 17] and will be derived elsewhere. It is characterized through the complementary distribution function of the time T until the next arrival ...