Evaluating the GPRS Radio Interface for Different Quality of Service Profiles
Abstract. This paper presents a discrete-event simulator for the General Packet Radio Service (GPRS) on the IP level. GPRS is a standard on packet data in GSM systems that will become commercially available by the end of this year. The simulator focuses on the communication over the radio interface, because it is one of the central aspects of GPRS. We study the correlation of GSM andGPRS users by a static and dynamic channel allocation scheme. In contrast to previous work, our approach represents the mobility of users through arrival rates of new GSM and GPRS users as well as handover rates of GSM and GPRS users from neighboring cells. Furthermore, we consider users with different QoS profiles modeled by a weighted fair queueing scheme. The simulator considers a cell cluster comprising seven hexagonal cells. We provide curves for average carried traffic and packet loss probabilities for differentchannel allocation schemes and packet priorities as well as curves for average throughput per GPRS user. A detailed comparison between static and dynamic channel allocation schemes is provided.
The General Packet Radio Service (GPRS) is a standard from the European Telecommunications Standards Institute (ETSI) on packet data in GSM systems , . By adding GPRS functionality to the existing GSM network, operators can givetheir subscribers resource-efficient wireless access to external Internet protocol-bases networks, such as the Internet and corporate intranets. The basic idea of GPRS is to provide a packet-switched bearer service in a GSM network. As impressively demonstrated by the Internet, packet-switched networks make more efficient use of the resources for bursty data applications and provide more flexibility in general. In previous work, several analytical models have been developed to study data services in a GSM network. Ajmone Marsan et al. studied multimedia services in a GSM network by providing more than one channel for data services . Boucherie and Litjens developed an analytical model based on Markov chain analysis to study the performance of GPRS under a given GSM call characteristic . For analytical tractability, they assumed exponentially distributed arrival times for packets and exponential packet transfer times, respectively. On the other hand, discrete-event simulation based studies of GPRS were conducted. Meyer et al. focused on the performance of TCP over GPRS under several carrier to interference conditions and coding schemes of data . Furthermore, they provided a detailed implementation of the GPRS protocol stack . Malomsoky et al. developed a simulation based GPRS network dimensioning tool . Stuckmann et al. studied the correlation of GSM and GPRS users with the simulator GPRSim . This paper describes a discrete-event simulator for GPRS on the IP level. The simulator is developed using the simulation package CSIM  and considers a cellcluster comprising of seven hexagonal cells. The presented performance studies were conducted for the innermost cell of the seven cell cluster. The simulator focuses on the communication over the radio interface, because this is one of the central aspects of GPRS. In fact, the air interface mainly determines the performance of GPRS. We studied the correlation of GSM and GPRS users by a static and dynamic channel allocation scheme. A first approach of modeling dynamic channel allocation was introduced by Bianchi et al. and is known as Dynamic Channel Stealing (DCS) .
The basic DCS concept is to temporarily assign the traffic channels dedicated to circuit-switched connections but unused because statistical traffic fluctuations. This can be done at no expense in terms of radio resource, and with no impact on circuitswitched services performance if the channel allocation to packet-switched services is
permitted only for idle traffic channels, and the stolen channels are immediately released when requested by the circuit-switched service. In contrast to the models developed in , , , and , our approach additionally represents the mobility of users through arrival rates of new GSM and GPRS users as well as handover rates of GSM and GPRS users from neighboring cells. Furthermore, we consider users with different QoS profiles modeled by a weighted fair queueing scheme according to . The remainder of the paper is organized as follows. Section 2 describes the basic GPRS network architecture, the radio interface, and different QoS profiles, which will be considered in the simulator. In Section 3 we describe the software architecture of the GPRS simulator, details about the mobility of GSM and GPRS users, the way we modeled quality of service profiles, and the workload model we used. Results of the simulation studies are presented in Section 4. We provide curves for average carried traffic and packet loss probabilities for different channel allocation schemes and packet priorities as well as curves for average throughput per GPRS user.
3 The Simulation Model
We consider a cluster comprising of sever hexadiagonal cells in an integrated GSM/GPRS network, serving circuit-switched voice and packet-switched data calls. The performance studies presented in Section 4 were conducted for the innermost cell of the seven cell cluster. We assume that GSM and GPRS calls arrive in each cell according to two mutually independent Poisson processes, with arrival rates ëGSM and ëGPRS, respectively. GSM calls are handled circuit-switched, so that one physical channel is exclusively dedicated to the corresponding mobile station. After the arrival of a GPRS call, a GPRS session begins. During this time a GPRS user allocates no physical channel exclusively. Instead the radio interface is scheduled among different GPRS users by the Base Station Controller (BSC) . Every GPRS user receives packets according to a specified workload model. The amount of time that a mobile station with an ongoing call remains within the area covered by the same BSC is called dwell time . If the call is still active after the dwell time, a handover toward an adjacent cell takes place. The call duration is defined as the amount of time that the call will be active, assuming it completes without being forced to terminate due to handover
failure. We assume the dwell time to be an exponentially distributed random variable with mean 1/µh,GSM for GSM calls and 1/µh,GPRS for GPRS calls. The call durations are
also exponentially distributed with mean values 1/µGSM and 1/µGPRS for GSM and
GPRS calls, respectively. To exactly model the user behavior in the seven cell cluster, we have to consider the handover flow of GSM and GPRS users from adjacent cells. At the boundary cells of the seven cell cluster, the intensity of the incoming handover flow cannot be
specified in advance. This is due to the handover rate out of a cell depends on the
number of active customers within the cell. On the other hand, the handover rate into
the cell depends on the number of customers in the neighboring cells. Thus, the
iterative procedure introduced in  is used to balance the incoming and outgoing
handover rates, assuming that the incoming handover rate ëh GSM
in i ,
( ) ( ) −1 computed at step i-1.
Since in the end-to-end path, the wireless link is typically the bottleneck, and given
the anticipated traffic asymmetry, the simulator focuses on resource contention in the
downlink (i.e., the path BSC → BTS → MS) of the radio interface. Because of the anticipated traffic asymmetry the amount of uplink traffic, e.g. induced by
acknowledgments, is assumed to be negligible. In the study we focus on the radio
interface. The functionality of the GPRS core network is not included. The arrival
stream of packets is modeled at the IP layer. Let N be the number of physical channels available in the cell. We evaluate the performance of two types of radio resource sharing schemes, which specify how the cell capacity is shared by GSM and GPRS users:
the static scheme ; that is the cell capacity of N physical channels is split into
NGPRS channels reserved for GPRS data transfer and NGSM = N - NGPRS channels
reserved for GSM circuit-switched connections.
the dynamic scheme ; that is the N physical channels are shared by GSM and
GPRS services, with priority for GSM calls; given n voice calls, the remaining
N-n channels are fairly shared by all packets in transfer.
In both schemes, the PDCHs are fairly shared by all packets in transfer up to a
maximum of 8 PDCHs per IP packet ("multislot mode") and a maximum of 8 packets
per PDCH .
The software architecture of the simulator follows the network architecture of the
GPRS Network . To accurately model the communication over the radio
interface, we include the functionality of a BSC and a BTS. IP packets that arrive at
the BSC are logically organized in two distinct queues. The transfer queue can hold
up to Q n = ⋅ 8 packets that are served according to a processor sharing service
discipline, with n the number of physical channels that are potentially available for
data transfer, i.e. n = NGPRS under the static scheme and n = N under the dynamic
scheme. The processor sharing service discipline fairly shares the available channel
capacity over the packets in the transfer queue. An arriving IP packet that cannot enter
the transfer queue immediately is held in a first-come first-served (in case of one
priority) access queue that can store up to K packets. The access queue models the
BSC buffer in the GPRS network. Upon termination of a packet transfer, the IP
packet at the head of the access queue is polled into the transfer queue, where it
immediately shares in the assignment of available PDCHs. For this study, we fix the
modulation and coding scheme to CS-2 . It allows a data transfer rate of 13,4
kbit/sec on one PDCH. Figure 1 depicts the software architecture of the simulator.
Figure 1. Software Architecture of GSM/GPRS Simulator
To model the different quality of service profiles GPRS provides, the simulator
implemented a Weighted Fair Queueing (WFQ) strategy. The WFQ scheduling
algorithm can easily be adopted to provide multiple data service classes by assigning
each traffic source a weight determined by its class. The weight controls the amount
of traffic a source may deliver relative to other active sources during some period of
time. From the scheduling algorithm's point of view, a source is considered to be
active if it has data queued at the BSC. For an active packet transfer with weight wi
the portion of the bandwidth Âi(t) allocated at time t to this transfer should be
( ) ( ) = ⋅ ∑
where the sum over all active packet transfers at time t. The overall bandwidth at time
t is denoted by B(t) which is independent of t in the static channel allocation scheme.
The workload model used in the GPRS simulator is a Markov-modulated Poisson
Process (MMPP) . It is used to generate the IP traffic for each individual user in
the system. The MMPP has been extensively used for modeling arrival processes,
because it qualitatively models the time-varying arrival rate and captures some of the
important correlations between the interarrival times. It is shown to be an accurate
model for Internet traffic which usually shows self-similarity among different time
scales. For our purpose the MMPP is parameterized by the two-state continuous-time
Markov chain with infinitesimal generator matrix Q and rate matrix Ë:
The two states represent bursty mode and non-bursty mode of the arrival process.
The process resides in bursty mode for a mean time of 1/á and in non-bursty mode for
a mean time of 1/â respectively. Such an MMPP is characterized by the average
arrival rate of packets, ëavg and the degree of burstiness, B. The former is given by:
The degree of burstiness is computed by the ratio between the bursty arrival rate and
the average arrival rate, i.e., B = ë1/ëavg.
4 Simulation Results
Table 1 summarizes the parameter settings underlying the performance experiments.
We group the parameters into three classes: network model, mobility model, and
traffic model. The overall number of physical channels in a cell is set to N = 20
among which at least one channel is reserved for GPRS. The overall number of GPRS
users that can be managed by a cell is set to M = 20. As base value, we assume that
5% of the arriving calls correspond to GPRS users and the remaining 95% are GSM
calls. GSM call duration is set to 120 seconds and call dwell time to 60 seconds, so
that users make 1-2 handovers on average. For GPRS sessions the average session
duration is set to 5 minutes and the dwell time is 120 seconds. Thus, we assume
longer “online times” and slower movement of GPRS users than for GSM users. The
average arrival rate of data is set to 6 Kbit/sec per GPRS user corresponding to 0.73
IP packets per second of size 1 Kbyte.
Figure 2 presents curves for carried data traffic and packet loss probabilities due to
buffer overflow in the BSC for the static channel allocation scheme and one packet
priority. For GPRS 1, 2, and 4 PDCHs are reserved, respectively. The remaining
channels can be used by GSM calls. With 4 PDCHs the system overloads at an arrival
rate of 0.8 GSM/GPRS users per second. This corresponds to an average of 12 GPRS
users in the cell (see Figure 7). In Figure 3 we present corresponding curves for the
dynamic channel allocation scheme. For GPRS 1, 2, and 4 PDCHs are reserved,
respectively but more PDCHs can be reserved "on demand". That means that
additional PDCHs can be reserved if they are not used for GSM voice service. From
Figure 3 we observe that for low traffic in the considered cell GPRS makes
effectively use of the on demand PDCHs. For example if 1 PDCH is reserved GPRS
utilizes up to 2 PDCHs at an arrival rate of 0.4 GSM/GPRS users per second. But
with increasing load the overall performance of GPRS decreases because of
concurrency among GPRS users, and more important, priority of GSM users over the
radio interface. In comparison with the static channel allocation scheme we conclude
that the combination of reserved PDCHs and on demand PDCH leads to a better
utilization of the scarce radio frequencies. The only advantage of the static channel
allocation scheme is that it can be realized more easily.
Figure 4 presents a comparison of overall channel utilization and average
throughput per GPRS user for the static and dynamic channel allocation scheme. For
the static scheme we reserved 2 and 4 PDCHs respectively and for the dynamic
scheme only 1 PDCH. We observe a higher overall utilization of physical channels by
the dynamic scheme. Comparing the dynamic with the static scheme for 2 PDCHs we
detect a slightly higher throughput for low traffic load for dynamic channel allocation.
This results from the high radio channel capacity available to GPRS users in this case.
They can utilize up to 8 PDCHs for their transfer (in contrast to 2 PDCHs in the static
scheme). When load increases, GSM calls allocate most of the physical channels.
Thus, throughput for GPRS users decreases very fast. In the static scheme (4 PDCHs)
the decrease in throughput is not so fast, because GSM calls do not effect the PDCHs.
In an additional experiment, we study the performance loss in the GSM voice
service due to the introduction of GPRS. Figure 5 plots the carried voice traffic and
voice blocking probability for different numbers of reserved PDCHs. The results are
valid for both channel allocation schemes because of the priority of GSM voice
service over GPRS. The presented curves indicate that the decrease in channel
capacity and, thus, the increase of the blocking probability of the GSM voice service
is negligible compared to the benefit of reserving additional PDCHs for GPRS users.
Figure 6 shows carried data traffic and packet loss probabilities for the dynamic
channel allocation scheme and different packet priorities. For GPRS 1 PDCH is
reserved. Weights for packets with priority 1 (high), 2 (medium), and 3 (low) and
percentages of GPRS users utilizing these priorities are given in Table 1. We observe
that for low traffic in the considered cell most channels are covered by packets of low
priority. This is due to the high portion of low priority packets (60%) among all
packets sharing the radio interface. With increasing load medium priority packets and
at last high priority packets suppress packets of lower priority and therefore the
utilization of PDCHs for low and medium priority packets decreases. For a call arrival
rate of up to 2 calls per second the loss probability of high priority packets is still less
than 10-5 and therefore the corresponding curve is omitted in Figure 6.