Dimensioning of UTRAN Iub Links for Elastic Internet Traffic ? Â· Dimensioning of UTRAN Iub Links for Elastic Internet Traffic Xi LiA, Richard SchelbB, Carmelita GörgA and Andreas Timm-GielA A Communication Networks, University of

Download Dimensioning of UTRAN Iub Links for Elastic Internet Traffic ? Â· Dimensioning of UTRAN Iub Links for Elastic Internet Traffic Xi LiA, Richard SchelbB, Carmelita GörgA and Andreas Timm-GielA A Communication Networks, University of

Post on 02-Feb-2018

214 views

Category:

Documents

0 download

Embed Size (px)

TRANSCRIPT

  • Dimensioning of UTRAN Iub Links for Elastic Internet Traffic Xi LiA, Richard SchelbB, Carmelita GrgA and Andreas Timm-GielA A Communication Networks, University of Bremen, FB1 Otto-Hahn-Allee NW1, 28359 Bremen, Germany {xili, cg, atg}@comnets.uni-bremen.de B Siemens AG Siemensdamm 62, 13627 Berlin, Germany richard.schelb@siemens.com Abstract: This paper studies a comprehensive dimensioning approach to design the UMTS access networks for elastic Internet traffic carried by the TCP protocol. The dimensioning approach is based on the M/G/R Processor Sharing (M/G/R-PS) queuing model for application performance. With considering all TCP connection aspects, an adjustment on the extended M/G/R-PS model is proposed in this paper. And specifically, a new expression for the mean sojourn times is presented for advocating admission control as a necessary means to maintain goodput in case of traffic overload, which behaves as an M/G/R/N processor sharing queuing model. In addition, the frame protocol (FP) PDU delay is addressed as another critical dimensioning factor due to the strict delay requirements on the Iub link. But an analytical solution considering all lower layer aspects could be intractably complex, thus the dimensioning for the FP layer is taken from a simulation approach. The M/G/R-PS model and related notable results are validated by comparison with simulation results. Keywords: UMTS, UTRAN, Iub, Dimensioning, M/G/R-PS, Call Admission Control, M/G/R/N-PS, FP PDU Delay 1. INTRODUCTION

    In a public multi-service UMTS network, it is important to provide an appropriate quality

    of service (QoS) for a great variety of services as well as effective resource utilization. Each of these services asks for specific quality of service demands with respect to performance parameters such as: end-to-end delay, jitter, loss, blocking ratio, etc. This poses new challenges in the planning of the UMTS network, especially within the radio access network (RAN) where the transmission resources are scarce and expensive. With the extension of the UTRAN to cover more and more suburban and rural areas, the narrow-bandwidth links between Radio Network Controller (RNC) and NodeB (base station) become considerably costly. For a cost-efficient design of the UMTS access network, it is essential for the network operator to dimension the Iub links carefully: over-dimensioning wastes precious bandwidth resources, whereas under-dimensioning generally leads to less satisfactory quality of service perceived by subscribers. This paper discusses a comprehensive dimensioning approach for

    ITC19/ Performance Challenges for Efficient Next Generation NetworksLIANG X.J. and XIN Z.H.(Editors)V.B. IVERSEN and KUO G.S.(Editors)Beijing University of Posts and Telecommunications Press

    415-424

    mailto:@comnets.uni-bremen.demailto:richard.schelb@siemens.com

  • the Iub link carrying the elastic Internet traffic which resides on top of the TCP/IP protocol suite, where the rate of TCP flow adjusts itself to fill the available bandwidth according to the network traffic condition by using the TCP flow control. This dimensioning approach is based on the M/G/R Processor Sharing (M/G/R-PS) queuing model which characterizes TCP traffic at the flow level, where mobile users represent flows generated by downloading Internet objects; the sojourn times represent the object transfer times. The transmission bandwidth of UMTS user connections is limited by their assigned radio access bearer (RAB) type, e.g. 64 kbps or 384 kbps. In order to guarantee a minimum throughput one must have a CAC (Call Admission Control) [6] scheme employed in the network to avoid instability and low goodput in case of overload. The admission control for elastic traffic is based on the maximum number of allowed flows. Whenever a per-flow access control is employed a blocking model is used to dimension the network. Thus the original M/G/R-PS method is extended to an M/G/R/N-PS model, which allows a maximum number of N active user connections in the system sharing R servers simultaneously. In addition, since the Iub interface has to fulfill strict delay requirements posed by UMTS specific layers in order to guarantee each radio frame delivered on time according to the air interface, the FP PDU delay is also a very critical dimensioning issue to be specifically considered. Due to this AAL2 was chosen as the adaptation layer which basically is focused on real-time traffic whereas a major portion of UMTS upper layer traffic represents non real-time characteristics.

    The remaining parts of this paper are organized in 4 sections. The following section summarizes analytical approaches for UTRAN dimensioning with focus on application performance. In the third section a description of the simulation model is given and the simulation results obtained from this model are presented. In the last section, key conclusions on the dimensioning problems of UMTS and an outlook of future work are given. 2. DIMENSIONING METHODS FOR ELASTIC TRAFFIC

    Figure 2-1 shows an overview of a UMTS network sketching its main components.

    Populations of mobile subscribers are connected to the NodeB via the radio interface (Uu-Interface), and then go through the UTRAN and Core Network to the remote Internet servers. The link between the Base Station (NodeB) and the Radio Network Controller (RNC), the Iub interface, is a potential bottleneck within the UTRAN. End users generate service requests to the application servers to download Internet objects such as Web pages, emails over TCP connections, which are established between the end users and the application servers. In this way, each request generates one or more TCP flows over the Iub link.

    It is assumed in this analysis that every elastic traffic flow is generated by one file transfer and all users are assigned to the same maximum radio access bit rate, e.g. 128kbps. The file transmission rates are controlled by the TCP feedback algorithm as a network congestion control function. If TCP works ideally (i.e. instantaneous feedback), all elastic traffic flows going over the same link will share the bandwidth resources equally, and thus the system only carrying elastic traffic flows is essentially behaving as an M/G/R-Processor Sharing queue on the Iub link. Each application is assigned a specific radio access bearer with a certain peak data rate. This is modeled by assuming R servers inside the system where each application can (at maximum) utilize the full capacity of a single server. In this paper the focus of the analysis is placed on a single RAB type, i.e. each application receives the same

    416

  • maximum data rate.

    Figure 2-1: Iub link connecting the access network to the core network

    The M/G/R-PS model can be imagined in such a way that several files that need to be transmitted over one link are broken into little pieces, i.e. individual IP packets of the different traffic flows, and are processed by the link quasi-simultaneously. In this way, large files do not delay small ones too much when compared with FIFO scheduling. Recently, a number of studies ([1], [2] and [3]) show that the M/G/R-PS model provides a simple and accurate characterization of elastic IP traffic at flow level. The performance criteria of dimensioning for elastic IP traffic is the average transfer delay and throughput, both of which are related to a delay factor, in addition other parameters like blocking probability have to be considered. 2.1 M/G/R-PS Model

    In this case, the attainable transfer rate for one individual subscriber is limited to a certain peak rate rp (i.e. RAB rate in our context) and C represents the trunk line capacity. Thereby the link appears like a Processor Sharing queue system with R servers (here server means TCP flows with maximum available bandwidth) where R = C/rp, hence the name M/G/R-PS queue. That means, only up to R flows can be served at the same time without rate reduction imposed by the system. But if individual flows are not subject to any bandwidth restrictions, i.e. a single elastic flow is able to fully utilize the total link capacity on its own, and then the M/G/R-PS model reduces to the simple M/G/1-PS model. However, this is not a common case in the real system. An important property of M/G/R-PS queues mentioned in [4] is that the average sojourn time (average time in system) is insensitive to file size distributions.

    If E{x} is the average file size and is the average flow arrival rate, then the traffic load (or utilization) is: = (E{x})/C. It has been shown in [1] that the expected sojourn time (or file transfer delay) E{T(x)} for a file of size x is given by:

    { } Rpp

    R/G/M frx

    )(R)R,R(E

    rx)x(TE =

    +=

    11 2 (2-1)

    Where E2 represents Erlangs second formula (Erlang C formula) and fR is defined as the delay factor. The delay factor fR represents the increase of the average file transfer time (and

    Core

    Network

    UTRAN

    Internet

    Iub Interface

    Radio Interface

    (Uu)

    User Equipment

    RAB Rate

    417

  • decrease of the average throughput) due to link congestion. It is a quantitative measure of how link congestion affects the transaction times, taking into account the economy of scale effect. An example of calculating the file transfer delay using formula 2-1 can be seen in Figure 3-1.

    2.2 Extended M/G/R-PS Model

    However, the M/G/R-PS model introduced above assumes ideal capacity sharing among active flows. But in practice the TCP flows are not always able to utilize their fair share of available bandwidth. TCPs effectiveness of capacity sharing is determined by the TCP slow start and congestion avoidance mechanisms which are affected by network conditions such as round-trip times and p