Analysis of a Traffic Model for GSM/GPRS - Instituto de ...
Analysis of a Traffic Model for GSM/GPRS Hugo Araújo, José Costa and Luis M. Correia Instituto de Telecomunicações/Instituto Superior Técnico, Technical University of Lisbon. Av. Rovisco Pais, 1049-001 Lisboa, Portugal; Ph: +351 218418478; Fax: +351 218418422 [email protected], [email protected], [email protected]
Abstract A traffic model for GSM/GPRS, the hybrid radio resource allocation (HRRA) algorithm is evaluated. A dedicated number of GPRS channels plus idle periods between voice calls are used for GPRS data packet transfers. A simulator was developed in order to evaluate the HRRA algorithm, which provides a reasonable forecast on the voice blocking probability and on packet delay for a single cell system. Since the major issue is the correct resource allocation, results are shown for the influence of some choices and assumptions on the overall system performance. As expected, blocking probability can reach very high values if the number of dedicated channels increases too much. For the specific case of 4 carriers and traffic of 20 Erl, 4 channels dedicated to GPRS still enable an affordable blocking probability, leading to a mean packet delay of 15 s. The results can be used to illustrate the fundamental options that need to be taken by an operator, when implementing GPRS. I. INTRODUCTION In today’s mobile communications world, the 2nd generation Global System for Mobile Communications (GSM) is clearly a winning system, used by millions around the globe. Due to its limited ability to grow and satisfy packet data communication needs, its future as a communications system is shadowed by the upcoming 3rd generation one, the Universal Mobile Telecommunications System (UMTS). That is mainly due to the fact that GSM creators thought of it as primarily a voice system, hence, lacking the ability to deal with large amounts of data, the kind of data that is overwhelmingly taking over all communication networks. It is widely foreseen that in the near future mobile data traffic will overtake voice as the primary service provided by mobile operators. Hence, operators need to evolve from current GSM networks, so that they can provide the necessary packet switched multi-service communications. In order to achieve that, General Packet Radio Service (GPRS) is the solution that is being implemented. The main reason for the limitation in dealing with large amounts of data is the circuit switched based transmission used in GSM. Due to the bursty nature of packet data, and to the scarcity of available resources, a packet switched transmission is be better suited. The GPRS concept was developed from the need to evolve GSM, in order to
accomplish an efficient way of delivering data packets, including the ones from other networks, with the minimum disturbance in the existing network. With the implementation of GPRS in GSM networks, and the foreseen growth of data transfer in mobile networks, it is necessary to achieve a compromise between voice and data services. The objective of this work is to study the behaviour of an integrated voice/data mobile communications system, by using simulations. The simulator was fully built around the Hybrid Radio Resource Allocation (HRRA) algorithm [1]. In the next section a brief description of the algorithm is done, which allows a better understanding of the HRRA implementation described in Section III. In Section IV, results from that implementation are presented for some cases of interest, leaving general conclusions to be drawn in Section V. II. HYBRID RADIO RESOURCE ALLOCATION ALGORITHM According to the HRRA algorithm, either voice or packet data traffic can be carried over a traffic channel, with GPRS dedicated channels being available exclusively for data transfers. If no resources are available voice calls are blocked and lost, while packet transfers are placed on a waiting queue. In this algorithm, voice calls have priority over all ongoing data calls, with the exception of those assigned to the packet switched dedicated traffic channels. In our study the case of a single cell GSM/GPRS system is considered. Considering N as the number of carries in use on that particular cell, it is well known that the number of physical channels available is n ≡ 8 × N.
(1)
Of these n channels, n control are assigned to control and signalling functions, the remaining ones, n traffic , being used for traffic. In the latter, both voice calls and data packets can be transmitted; however, it can happen that some of the n traffic channels are reserved for either voice or data only, in what is known as a prioritised channel sharing scheme [2]. In our study, one will use data dedicated channels, CGPRS , these being permanently allocated to GPRS data transfer. The remaining traffic channels, Cshared , are considered to be shared ones, and they will be allocated on a demand basis. The channel sharing scheme is represented in Fig. 1.
Fig. 1: Channel sharing scheme for a GPRS network. III. IMPLEMENTATION OF THE SIMULATOR As a first step, it was necessary to provide the voice and data traffic profile input parameters, so that the corresponding call generators were developed. A Poisson process arrival rate and voice calls’ duration with exponential distribution [3] were assumed. To simplify and reduce simulation time, a fixed length for generated data traffic of 536 byte was assumed, which is considered to be a typical IP packet size [1]. An example of the outputs produced by the voice and packet generators is displayed in Fig. 2, as well as the result of adding these two together. Another important assumption was that a fixed time unit was considered throughout the whole simulation process, in order to achieve a considerable simplification, with a
consequent reduction of processing time. The time unit was defined as being the time that is necessary to receive a complete data packet. With this assumption, a packet is treated as a unit, it never being divided or processed in more than one time unit. Considering the use of Coding Scheme-1, CS-1 [4], which can achieve a net bit rate of 9.05 kbit/s, one can calculate the duration of this time unit as the number of bits per packet multiplied by the amount of time it takes to send a bit with CS-1, which leads to a packet duration of 0.474 s. Since packets cannot be divided, once a transmission starts the corresponding channel is dedicated for that time unit to that single packet. Since no information was available for real GPRS data transfers, it was necessary to study a wide range of data loads, these being defined as the percentage of the system potential data transfer capability. For each data load, the packet arrival rate was calculated from the expected number of packets that would arrive in each simulated unit of time. Due to the asymmetric nature of packet data networks, upand downlinks do not carry the same amount of traffic. Since it is expected that the latter will be the one with a higher volume of traffic, it was the one considered in the simulator. This implementation of the HRRA algorithm, based on [1], is presented in Fig. 3. Several parameters are evaluated throughout the whole simulation process, the most important ones being the blocking probability and the packet delay, as they influence the grade and quality of services intended for the mobile network. These parameters are directly connected with the number of dedicated GPRS channels defined by the network operator, and the data load requested by the users.
Fig. 2: Voice calls and data generators output for a 10% data load.
Fig. 3: Hybrid Radio Resource Allocation (HRRA) algorithm as implemented in the simulator.
30
30
25
25
20
20
15
15
10
10
5
5
0
0 0
4 8 # GPRS dedicated channels
12
Blocking probability, T=50s Blocking probability, T=25s Mean packet delay, T=50 s Mean packet delay, T=25 s
Fig. 4: Voice blocking probability and mean packet access delay versus the number of GPRS dedicated channels . It is obvious from Fig. 4 that, for large values of the number of GPRS dedicated channels, the voice blocking probability grows more than the gain in performance of data transfer. One should note that the reference value for the voice blocking probability is 1 %. This fact shows that there
8
8
7
7
6
Voice blocking probability
6
5
Data packet delay
5
4
4
3
3
2
2
1
1
0
Mean delay [s]
35 Packet access delay [s]
.
35
Voice blocking probalility [%]
The case considered for analysis was the one of 4 carriers, 30 traffic channels, 30 % of data load, and an average in-cell voice call duration, T, of either 25 or 50 s, following a traffic profile made available by Telecel Vodafone. Voice blocking probability and mean packet access delay are shown in Fig. 4, where 20 Erl was considered for voice load; more detailed results can be found in [5].
is a compromise to be made by network operators when establishing certain grades of service to their packet switched calls. Since to many of today’s network operators voice calls traffic is still the key business, an interesting analysis is to see how the network is going to respond to an increasing number of voice traffic, in the case where some resources are allocated exclusively to GPRS. In Fig. 5 one can see the growing blocking probability and mean packet delay with the increase of voice calls’ traffic, while data transfers remain constant with an arrival rate of 0.4 s-1 and a 30% system data load; of the 30 traffic channels available, 8 were dedicated to GPRS.
Pb [%]
III. RESULTS
0 5,6
14,4
20
Voice Traffic [Erl]
Fig. 5: Voice blocking probability and packet delay versus voice traffic load. The most important assessment taken from Fig. 5 is the large growth in the voice blocking probability when the voice traffic load is high, clearly above 1%.
A useful tool for foreseeing network congestion is the delay cumulative probability distribution for the data packets transmitted over a certain period of time. In Fig. 6, a 3 hours period is represented, corresponding to the situation previously mentioned. 1,E+00
Probability of occurrence
5,6 Erl 14,4 Erl
1,E-01
20 Erl
1,E-02
1,E-03
1,E-04 0
3
5
8
10
12
15
17
19
22
Mean packet delay [s]
Fig. 6: Probability of packet delay for 3 different voice traffic loads in a 3 hours period. One can observe the evolution of the maximum delay a packet may suffer. This is very important to a network operator, so that he can design the system’s buffer to be used for packets queuing. For example, one has verified that near a system’s congestion situation the delay grows constantly, leading to an unstable system, since an infinite queue would be necessary to avoid losing data packets. In Fig. 7 the evolution of the maximum amount of delay encountered in a 3 hours simulation is presented, for the same scenario as in the previous cases.
V. ACKNOWLEDGMENTS The authors would like to thank the precious help provided by Telecel Vodafone network planning and development department regarding real networks traffic profiles. VI. REFERENCES
25
Maximum packet delay [s]
In the case of no channels dedicated to GPRS, voice calls are unaffected by GPRS traffic. As the number of GPRS dedicated channels increase, so does the voice calls’ blocking probability. This increase of blocking probability is more visible for a higher voice calls’ arrival rate when considering an equal traffic load. In the case of 4 carriers, the blocking probability reaches values higher than 2 % for more than 4 dedicated channels. It has been verified that voice calls ’ blocking probability increases with the number of GPRS dedicated channels, following the Erlang-B theoretical curve. Data transfers are unaffected by the number of GPRS dedicated channels for low data load situations. As data load increases so mean access delay does, as well as the number of packets delayed, overloading the system’s queue. Also the deviation from the expected packet delay is larger. In a system where the load is low and the voice traffic is constant, the number of voice calls handled when few GPRS dedicated channels are in use influences packet’s delay results. It can be concluded that dedicated GPRS channels play an important role only on heavily loaded voice and/or data systems. In the cases where the system is congested, using a small number of GPRS dedicated channels can increase packet transfer performance. This also ensures a minimum quality to the packet switch service, with a small sacrifice of the voice blocking probability.
20
15
10
5
0 0
2
4
6
8
10
12
14
16
18
20
Voice traffic [Erl]
Fig. 7: Maximum packet delay in a 3 hours simulation for different values of voice traffic. As it would be expected, the packet delay grows constantly with an increasing voice traffic load. V. CONCLUSIONS The case of mixed voice and data traffic in a GSM/GPRS system is dealt with in this paper. The HHRA algorithm has been implemented, so that system performance can be evaluated for different conditions of operation.
[1] K. Kennedy and R. Litjens, “Performance Evaluation of a Hybrid Radio Resource Allocation Algorithm in a GSM/GPRS Network”, in Proc. of PIMRC'99 –10th IEEE International Symposium on Personal Indoor and Mobile Radio Communications, Osaka, Japan, Sep. 1999. [2] D. Hong and T.S. Rappaport, “Traffic Model and Performance Analysis for Cellular Mobile Radio Telephone Systems with Prioritized and Non-prioritized Hand-off Procedures,” IEEE Transactions on Vehicular Technology, Vol. VT-35, No. 3, Aug. 1986, pp.77-92. [3] M.D. Yacoub, Foundations of Mobile Radio Engineering, CRC Press, Boca Raton, FL, USA, 1993. [4] ETSI Technical Specification GSM 03.64, Overall description of the GPRS radio interface – stage 2, V8.4.0, Apr. 2000, pp.20-37. [5] H. Araújo and J. Costa, Analysis of mixed voice and data traffic in GSM/GPRS (in Portuguese), Graduation Thesis, IST, Lisbon, Portugal, Nov. 2000.
Abstract A traffic model for GSM/GPRS, the hybrid radio resource allocation (HRRA) algorithm is evaluated. A dedicated number of GPRS channels plus idle periods between voice calls are used for GPRS data packet transfers. A simulator was developed in order to evaluate the HRRA algorithm, which provides a reasonable forecast on the voice blocking probability and on packet delay for a single cell system. Since the major issue is the correct resource allocation, results are shown for the influence of some choices and assumptions on the overall system performance. As expected, blocking probability can reach very high values if the number of dedicated channels increases too much. For the specific case of 4 carriers and traffic of 20 Erl, 4 channels dedicated to GPRS still enable an affordable blocking probability, leading to a mean packet delay of 15 s. The results can be used to illustrate the fundamental options that need to be taken by an operator, when implementing GPRS. I. INTRODUCTION In today’s mobile communications world, the 2nd generation Global System for Mobile Communications (GSM) is clearly a winning system, used by millions around the globe. Due to its limited ability to grow and satisfy packet data communication needs, its future as a communications system is shadowed by the upcoming 3rd generation one, the Universal Mobile Telecommunications System (UMTS). That is mainly due to the fact that GSM creators thought of it as primarily a voice system, hence, lacking the ability to deal with large amounts of data, the kind of data that is overwhelmingly taking over all communication networks. It is widely foreseen that in the near future mobile data traffic will overtake voice as the primary service provided by mobile operators. Hence, operators need to evolve from current GSM networks, so that they can provide the necessary packet switched multi-service communications. In order to achieve that, General Packet Radio Service (GPRS) is the solution that is being implemented. The main reason for the limitation in dealing with large amounts of data is the circuit switched based transmission used in GSM. Due to the bursty nature of packet data, and to the scarcity of available resources, a packet switched transmission is be better suited. The GPRS concept was developed from the need to evolve GSM, in order to
accomplish an efficient way of delivering data packets, including the ones from other networks, with the minimum disturbance in the existing network. With the implementation of GPRS in GSM networks, and the foreseen growth of data transfer in mobile networks, it is necessary to achieve a compromise between voice and data services. The objective of this work is to study the behaviour of an integrated voice/data mobile communications system, by using simulations. The simulator was fully built around the Hybrid Radio Resource Allocation (HRRA) algorithm [1]. In the next section a brief description of the algorithm is done, which allows a better understanding of the HRRA implementation described in Section III. In Section IV, results from that implementation are presented for some cases of interest, leaving general conclusions to be drawn in Section V. II. HYBRID RADIO RESOURCE ALLOCATION ALGORITHM According to the HRRA algorithm, either voice or packet data traffic can be carried over a traffic channel, with GPRS dedicated channels being available exclusively for data transfers. If no resources are available voice calls are blocked and lost, while packet transfers are placed on a waiting queue. In this algorithm, voice calls have priority over all ongoing data calls, with the exception of those assigned to the packet switched dedicated traffic channels. In our study the case of a single cell GSM/GPRS system is considered. Considering N as the number of carries in use on that particular cell, it is well known that the number of physical channels available is n ≡ 8 × N.
(1)
Of these n channels, n control are assigned to control and signalling functions, the remaining ones, n traffic , being used for traffic. In the latter, both voice calls and data packets can be transmitted; however, it can happen that some of the n traffic channels are reserved for either voice or data only, in what is known as a prioritised channel sharing scheme [2]. In our study, one will use data dedicated channels, CGPRS , these being permanently allocated to GPRS data transfer. The remaining traffic channels, Cshared , are considered to be shared ones, and they will be allocated on a demand basis. The channel sharing scheme is represented in Fig. 1.
Fig. 1: Channel sharing scheme for a GPRS network. III. IMPLEMENTATION OF THE SIMULATOR As a first step, it was necessary to provide the voice and data traffic profile input parameters, so that the corresponding call generators were developed. A Poisson process arrival rate and voice calls’ duration with exponential distribution [3] were assumed. To simplify and reduce simulation time, a fixed length for generated data traffic of 536 byte was assumed, which is considered to be a typical IP packet size [1]. An example of the outputs produced by the voice and packet generators is displayed in Fig. 2, as well as the result of adding these two together. Another important assumption was that a fixed time unit was considered throughout the whole simulation process, in order to achieve a considerable simplification, with a
consequent reduction of processing time. The time unit was defined as being the time that is necessary to receive a complete data packet. With this assumption, a packet is treated as a unit, it never being divided or processed in more than one time unit. Considering the use of Coding Scheme-1, CS-1 [4], which can achieve a net bit rate of 9.05 kbit/s, one can calculate the duration of this time unit as the number of bits per packet multiplied by the amount of time it takes to send a bit with CS-1, which leads to a packet duration of 0.474 s. Since packets cannot be divided, once a transmission starts the corresponding channel is dedicated for that time unit to that single packet. Since no information was available for real GPRS data transfers, it was necessary to study a wide range of data loads, these being defined as the percentage of the system potential data transfer capability. For each data load, the packet arrival rate was calculated from the expected number of packets that would arrive in each simulated unit of time. Due to the asymmetric nature of packet data networks, upand downlinks do not carry the same amount of traffic. Since it is expected that the latter will be the one with a higher volume of traffic, it was the one considered in the simulator. This implementation of the HRRA algorithm, based on [1], is presented in Fig. 3. Several parameters are evaluated throughout the whole simulation process, the most important ones being the blocking probability and the packet delay, as they influence the grade and quality of services intended for the mobile network. These parameters are directly connected with the number of dedicated GPRS channels defined by the network operator, and the data load requested by the users.
Fig. 2: Voice calls and data generators output for a 10% data load.
Fig. 3: Hybrid Radio Resource Allocation (HRRA) algorithm as implemented in the simulator.
30
30
25
25
20
20
15
15
10
10
5
5
0
0 0
4 8 # GPRS dedicated channels
12
Blocking probability, T=50s Blocking probability, T=25s Mean packet delay, T=50 s Mean packet delay, T=25 s
Fig. 4: Voice blocking probability and mean packet access delay versus the number of GPRS dedicated channels . It is obvious from Fig. 4 that, for large values of the number of GPRS dedicated channels, the voice blocking probability grows more than the gain in performance of data transfer. One should note that the reference value for the voice blocking probability is 1 %. This fact shows that there
8
8
7
7
6
Voice blocking probability
6
5
Data packet delay
5
4
4
3
3
2
2
1
1
0
Mean delay [s]
35 Packet access delay [s]
.
35
Voice blocking probalility [%]
The case considered for analysis was the one of 4 carriers, 30 traffic channels, 30 % of data load, and an average in-cell voice call duration, T, of either 25 or 50 s, following a traffic profile made available by Telecel Vodafone. Voice blocking probability and mean packet access delay are shown in Fig. 4, where 20 Erl was considered for voice load; more detailed results can be found in [5].
is a compromise to be made by network operators when establishing certain grades of service to their packet switched calls. Since to many of today’s network operators voice calls traffic is still the key business, an interesting analysis is to see how the network is going to respond to an increasing number of voice traffic, in the case where some resources are allocated exclusively to GPRS. In Fig. 5 one can see the growing blocking probability and mean packet delay with the increase of voice calls’ traffic, while data transfers remain constant with an arrival rate of 0.4 s-1 and a 30% system data load; of the 30 traffic channels available, 8 were dedicated to GPRS.
Pb [%]
III. RESULTS
0 5,6
14,4
20
Voice Traffic [Erl]
Fig. 5: Voice blocking probability and packet delay versus voice traffic load. The most important assessment taken from Fig. 5 is the large growth in the voice blocking probability when the voice traffic load is high, clearly above 1%.
A useful tool for foreseeing network congestion is the delay cumulative probability distribution for the data packets transmitted over a certain period of time. In Fig. 6, a 3 hours period is represented, corresponding to the situation previously mentioned. 1,E+00
Probability of occurrence
5,6 Erl 14,4 Erl
1,E-01
20 Erl
1,E-02
1,E-03
1,E-04 0
3
5
8
10
12
15
17
19
22
Mean packet delay [s]
Fig. 6: Probability of packet delay for 3 different voice traffic loads in a 3 hours period. One can observe the evolution of the maximum delay a packet may suffer. This is very important to a network operator, so that he can design the system’s buffer to be used for packets queuing. For example, one has verified that near a system’s congestion situation the delay grows constantly, leading to an unstable system, since an infinite queue would be necessary to avoid losing data packets. In Fig. 7 the evolution of the maximum amount of delay encountered in a 3 hours simulation is presented, for the same scenario as in the previous cases.
V. ACKNOWLEDGMENTS The authors would like to thank the precious help provided by Telecel Vodafone network planning and development department regarding real networks traffic profiles. VI. REFERENCES
25
Maximum packet delay [s]
In the case of no channels dedicated to GPRS, voice calls are unaffected by GPRS traffic. As the number of GPRS dedicated channels increase, so does the voice calls’ blocking probability. This increase of blocking probability is more visible for a higher voice calls’ arrival rate when considering an equal traffic load. In the case of 4 carriers, the blocking probability reaches values higher than 2 % for more than 4 dedicated channels. It has been verified that voice calls ’ blocking probability increases with the number of GPRS dedicated channels, following the Erlang-B theoretical curve. Data transfers are unaffected by the number of GPRS dedicated channels for low data load situations. As data load increases so mean access delay does, as well as the number of packets delayed, overloading the system’s queue. Also the deviation from the expected packet delay is larger. In a system where the load is low and the voice traffic is constant, the number of voice calls handled when few GPRS dedicated channels are in use influences packet’s delay results. It can be concluded that dedicated GPRS channels play an important role only on heavily loaded voice and/or data systems. In the cases where the system is congested, using a small number of GPRS dedicated channels can increase packet transfer performance. This also ensures a minimum quality to the packet switch service, with a small sacrifice of the voice blocking probability.
20
15
10
5
0 0
2
4
6
8
10
12
14
16
18
20
Voice traffic [Erl]
Fig. 7: Maximum packet delay in a 3 hours simulation for different values of voice traffic. As it would be expected, the packet delay grows constantly with an increasing voice traffic load. V. CONCLUSIONS The case of mixed voice and data traffic in a GSM/GPRS system is dealt with in this paper. The HHRA algorithm has been implemented, so that system performance can be evaluated for different conditions of operation.
[1] K. Kennedy and R. Litjens, “Performance Evaluation of a Hybrid Radio Resource Allocation Algorithm in a GSM/GPRS Network”, in Proc. of PIMRC'99 –10th IEEE International Symposium on Personal Indoor and Mobile Radio Communications, Osaka, Japan, Sep. 1999. [2] D. Hong and T.S. Rappaport, “Traffic Model and Performance Analysis for Cellular Mobile Radio Telephone Systems with Prioritized and Non-prioritized Hand-off Procedures,” IEEE Transactions on Vehicular Technology, Vol. VT-35, No. 3, Aug. 1986, pp.77-92. [3] M.D. Yacoub, Foundations of Mobile Radio Engineering, CRC Press, Boca Raton, FL, USA, 1993. [4] ETSI Technical Specification GSM 03.64, Overall description of the GPRS radio interface – stage 2, V8.4.0, Apr. 2000, pp.20-37. [5] H. Araújo and J. Costa, Analysis of mixed voice and data traffic in GSM/GPRS (in Portuguese), Graduation Thesis, IST, Lisbon, Portugal, Nov. 2000.
