Dynamic radio resource allocation for 3G and beyond mobile wireless ...

8 downloads 421110 Views 295KB Size Report
In this paper, theses strategies are extended for operation in 3G and beyond network. First ... Keywords: Wireless networks; Dynamic resource allocation; QoS; Queueing. 1. ...... cellular systems, IEEE Transactions on Vehicular Technology.
Computer Communications 30 (2006) 41–51 www.elsevier.com/locate/comcom

Dynamic radio resource allocation for 3G and beyond mobile wireless networks Salman A. AlQahtani *, Ashraf S. Mahmoud Computer Engineering Department, King Fahd University of Petroleum and Minerals, Dhahran 31261, Saudi Arabia Received 17 April 2006; received in revised form 13 July 2006; accepted 14 July 2006 Available online 10 August 2006

Abstract Next generation of wireless cellular networks aim at supporting a diverse range of multimedia services to mobile users with guaranteed quality of service (QoS). Resource allocation and call admission control (CAC) are key management functions in future 3G and 4G cellular networks, in order to provide multimedia applications to mobile users with QoS guarantees and efficient resource utilization. There are two main strategies for radio resource allocations in cellular wireless networks known as complete partitioning (CP) and complete sharing (CS). In this paper, theses strategies are extended for operation in 3G and beyond network. First, two CS-based call admission controls, referred to herein as queuing priority call admission control (QP-CAC) and hybrid priority call admission control (HP-CAC), and one CP-based call admission control referred to as complete partitioning call admission control (CP-CAC) are presented. Then, this study proposes a novel dynamic procedure, referred to as the dynamic prioritized uplink call admission control (DP-CAC) designed to overcome the shortcomings of CS and CP-based CACs. Results indicate the superiority of DP-CAC as it is able to achieve a better balance between system utilization, revenue, and quality of service provisioning. CS-based algorithms achieve the best system utilization and revenue at the expense of serious unfairness for the traffic classes with diverse QoS requirements. DP-CAC manages to attain equal system utilization and revenue to CS-based algorithms without the drawbacks in terms of fairness and service differentiation.  2006 Elsevier B.V. All rights reserved. Keywords: Wireless networks; Dynamic resource allocation; QoS; Queueing

1. Introduction Next generation of wireless cellular networks, including 3G and 4G wideband code-division multiple access (WCDMA), are expected to provide a diverse range of multimedia services for mobile users with different quality of service (QoS). The Universal Mobile Telecommunication System (UMTS) is required to support a wide range of applications each with its own specific QoS requirements. There are four QoS classes defined for UMTS network, namely, conversational, streaming, interactive, and background [3]. Each class has its own application-level QoS requirements in terms of delay, jitter, bit-error rate (BER), throughput, and burstiness, as well as call-level *

Corresponding author. Tel.: +966505225172. E-mail addresses: [email protected] (S.A. AlQahtani), [email protected] (A.S. Mahmoud). 0140-3664/$ - see front matter  2006 Elsevier B.V. All rights reserved. doi:10.1016/j.comcom.2006.07.007

QoS requirements in terms of call blocking/dropping probabilities. This requires medium access control (MAC) and call admission control (CAC) protocols to, respectively, enable application-level and call-level performance guarantees for the QoS classes. This paper focuses on CAC with call-level QoS guarantees. Call admission control (CAC) is one method to manage radio resources while optimizing the overall network performance. The existing resource sharing schemes can be broadly classified into two main categories: complete sharing (CS) and complete partitioning (CP) [1,2]. In CP, the available channels or resources are partitioned such that for each call class only a fixed partition of the resource is available. Therefore, the calls are accepted whenever there are available resources in their corresponding partition, otherwise they are blocked or queued. The CAC using this criterion is referred to as CP-CAC. In CS, a new user is always offered access to the network provided that there are

42

S.A. AlQahtani, A.S. Mahmoud / Computer Communications 30 (2006) 41–51

sufficient resources at the time of request, and all traffic classes share the total resources indiscriminately. Unlike second generation (2G) systems, the capacity of a WCDMA system is soft and does not correspond to a fixed number of channels. Several uplink CACs designed for 3G WCDMA have been proposed in the literature [4–14]. These CACs can be classified based on the admission criterion into the following four categories: powerbased CAC, throughput-based CAC, interference-based CAC, and signal-to-interference (SIR) based CAC. For power-based CAC algorithms the total received power is monitored, while throughput-based CACs monitor the system load. Interference-based CAC algorithms monitor the total received interference, and SIR-based CACs monitor the SIR figure experienced by every user. A reserved capacity for WCDMA is defined as a fraction of cell capacity in terms of the total interference, referred to herein as interference margin (IM) or in terms of the total load, referred to herein as load margin (LM). In [10,11] a CAC algorithm using multiple power-based thresholds for multiple services is proposed. By setting a higher threshold for voice traffic, the voice traffic is given a higher priority compared to data traffic. In [12,13] an interference-based admission control strategy with multiple IM is analyzed where only two classes of traffic are considered. In [14] a throughput-based admission control with multiple LM is proposed where four classes of traffic are considered. Recently, dynamic-threshold schemes have been discussed in the literature to improve the QoS guarantees for higher priority calls [15–17]. In [15], a throughput-based algorithm is proposed where it allows different adaptive LM for newly originating and handoff calls. The LM value is adapted using the arrival rates and the estimation of the blocking probability. In [16], IM needed for high priority calls is estimated by using the signal-tointerference ratio (SNR) and distance information of mobile users in neighboring cells. In [17], radio resource management (RRM) in each base station estimates the amount of IM by considering traffic load in its current cell as well as traffic conditions in neighboring cells. All these schemes introduce a large communication and processing overhead in order to keep up-to-date information about the state of the neighboring cells. Moreover the queuing techniques were not used. The above CAC techniques suffer form several drawbacks or disadvantageous. First, the focus is mostly on prioritization using different fixed or dynamic threshold values (i.e., IM or LM) without using buffering techniques. One major limitation of the fixed-threshold schemes is that the reserved capacity for higher traffic classes may remain unutilized while lower priority classes are being blocked. Moreover, most of the dynamic schemes rely on changing the threshold value based on periodic estimations in order to decrease the failure of higher priority handoff calls at the expense of lower priority new calls. This will introduce a large communication and processing overhead in order to keep up-to-date information about the state of neighboring

cells and therefore will limit the system scalability. In addition, these algorithms only increase the threshold when the estimate indicates high handoff traffic loads without giving a more balanced performance between new calls and handoff calls. Finally, these schemes do not provide detailed classification of calls based on traffic type (real-time and non-real-time) and request type (newly originating and handoff call) and no attempt is made to employ queuing for all classes of calls. The main contributions of this paper can be stated as follows. First, three uplink CAC algorithms, one based on CP and two based on CS, are extended for operation in 3G WCDMA networks. These adapted CAC schemes are the complete partitioning CAC (CP-CAC), the queuing priority CAC (QP-CAC), and the hybrid priority CAC (HP-CAC). Queuing, in addition to LM reservation strategies, for all traffic classes is extending here in order to increase the system utilization and to add other means of prioritization for higher priority classes. In CP-CAC, each call class has its own queue and resource partition whereas in QP-CAC, each call class has its own queue and all classes share the available resources. HP-CAC is a variant of QP-CAC, where resource utilization thresholds are assigned to each traffic class while allowing the higher priority class to share the resources of the lower class in addition to allowing each call class to be queued. CS-based CACs, such as QP-CAC and HP-CAC, achieve the highest resource utilization, but the individual QoS requirement of a certain traffic class cannot be guaranteed in these schemes. Therefore, while CS-based CACs produce the best utilization of the resource; it has been shown to be biased against calls that require high bandwidth. Consequently, QoS of each service cannot be easily controlled. On the other hand, CP-CAC can guarantee the resource commitment, and therefore the QoS, to each traffic class, but may underutilize the resources. If some underloaded traffic classes are not utilizing their allocations, the free capacity will be wasted as it cannot be used by other heavily loaded traffic classes. Based on the observations and evaluations of CP-CAC, QP-CAC, and HP-CAC, the second contribution of this paper is to design and evaluate a novel dynamically prioritized call admission control (DP-CAC) scheme. Resource utilization of the CP-CAC scheme and also the individual QoS requirement of a certain traffic class in case of CS can be improved by using this new scheme in which the resource allocation to a traffic class can be dynamically adjusted according to the traffic load variations and QoS requirements. While most of the free capacity from under-loaded traffic classes can be utilized by the overloaded traffic classes, dynamic priority is implemented to protect under-loaded traffic classes from resource starvation. In this study we compare the performance of CP-CAC, QP-CAC and HP-CAC against that of DP-CAC in terms of system utilization and quality of service provisioning. To summarize, the objectives of this new dynamic prioritization scheme are:

S.A. AlQahtani, A.S. Mahmoud / Computer Communications 30 (2006) 41–51

• Insure best system utilization and revenue while satisfying the required QoS and fairness.  At low and moderate traffic load, it ensures the best system utilization while QoS is satisfied (as best as CS).  At high load, it ensures the fairness of resource usage amongst different class (as good as CP). • Provide a scalable and easy to implement RRM procedure. • Eliminate the requirement for traffic estimation and communication with neighboring cells. • Supports preferential treatment to higher priority calls by serving its queue first. Unlike previous work, our proposed DP-CAC scheme aims at fulfilling the objectives above simultaneously. The rest of this paper is organized as follows. Section 2 describes the system model. Section 3 explains the proposed schemes in details while Section 4 presents the traffic model and performance measures. Finally, Section 5 provides the simulation model, the obtained results, as well as the discussion. The paper is concluded in Section 6.

43

when available based on its calculated priority. Let gi, be the LM corresponding to the ith (i = 1, 2, 3, and 4) traffic class, and gmax be the maximum loading that can be tolerated. Depending on how these loading limits are set, we have the following CAC schemes: the HP-CAC with fixed LM for each class, the QP-CAC where g1 = g2 = g3 = g4 = gmax, the CP-CAC with isolated load partitions for each class, and the dynamic priority CAC (DP-CAC) scheme. Unless specified, we assume that g1 = gmax for all the schemes. These CAC schemes are depicted pictorially in Figs. 1, 2, 3, and 4, respectively. The operation of these CAC schemes is detailed in the next section.

K

Total Cell Load max

Highest Priority

h1

1

Handoff RT

1

L

h2 Handoff NRT

2

2

Time out 3

M

n1

2. System model

Time out

New Calls RT

The system under consideration is a 3G and beyond WCDMA cellular network supporting heterogeneous traffic. The study assumes two types of services: real-time service (RT), such as conversational and streaming traffic, and non-real-time service (NRT) such as interactive and background traffic. The priority classes of incoming call requests are divided into four types. These types are: (class1) RT service handoff requests; (class2) NRT service handoff requests; (class3) newly originating RT calls; and (class4) newly originating NRT calls. These four priority classes are illustrated in Table 1 in the order of descending priority. The capacity of WCDMA cell is defined in terms of the cell load [3] where the load factor, g, is the instantaneous resource utilization upper bounded by the maximum cell capacity, gmax. Instantaneous values for the cell load g range from 0 to 1. Using this load factor, we devise a QoS-aware CAC algorithm for WCDMA-based networks by using the concept of thresholds and queuing techniques. Each call class has its own queues: Q1, Q2, Q3, and Q4 with finite capacities K, L, M, and N, respectively. A call class request is placed in its corresponding queue if it cannot be serviced upon its arrival and assigned a resource

3

4

Time out

N n2 New Calls NRT

4

Time out Lowest Priority

Fig. 1. HP-CAC scheme.

K

Highest Priority

h1

Total Cell Load max

Handoff RT

1

Time out

L

h2 Handoff NRT

2

Time out

M

n1 New Calls RT

3

Time out

N n2 New Calls NRT

4

Time out

Lowest Priority

Fig. 2. QP-CAC scheme.

Table 1 Priority classes Class

Traffic type

Call class descriptions

1 2 3 4

RT NRT RT NRT

Conversational Interactive and Conversational Interactive and

and streaming – handoff calls background – handoff calls and streaming – new calls background – new calls

44

S.A. AlQahtani, A.S. Mahmoud / Computer Communications 30 (2006) 41–51 Total Cell Load max

K h1 Handoff RT

1

L1

Time out

L

h2 Handoff NRT

2

L2

Time out

gc ¼

M

n1

3

Time out

n2 New Calls NRT

Time out

4

Dgi ¼ ð1 þ f Þ

B X i¼1

mi 6 gmax 1 þ Gi =ei

ð2Þ

where f is the factor accounting for interference from other cells and is defined as the ratio of inter-cell interference to the total interference in the referenced cell, whereas mi is the average traffic activity factor of ith MS. Below each of the studied CAC algorithms is specified.

L4

N

B X i¼1

L3

New Calls RT

quality. Using the load factor increment definition, the current total load factor, gc, for such an interference system is the sum of the load factor increments brought by all B active MSs. Therefore,

Fig. 3. CP-CAC scheme.

3.1. HP-CAC and QP-CAC schemes Total Cell Load max

K h1 Handoff RT

1

L1

Time out

L

h2

L2

Handoff NRT

2

Time out

M

n1

DPCAC

New Calls RT

3

N n2

New Calls NRT

4

L3

Time out

L4

Time out

Fig. 4. DP-CAC scheme.

3. CAC schemes The developed algorithms attempt to manage resource allocations amongst the different call classes, and to efficiently utilize the resources while satisfying the QoS requirements. Similar to studies [10,13,15], only the uplink direction is considered this study where it is assumed that whenever the uplink channel is assigned the downlink is established. To implement the admission control for WCDMA systems, first an estimate of the total cell load must be computed and then be employed in the decision process of accepting or rejecting new connections. In addition, the analysis assumes perfect power control operation where a mobile station (MS) and its home base station (BS) use only the minimum needed power in order to achieve the required performance. Considering the interference on the uplink, the load factor increment Dgi for a new request i can be estimated as [3]; Dgi ¼

1 1 þ Gi =ei

The HP-CAC and QP-CAC schemes’ model are shown in Figs. 1 and 2, respectively. In QP-CAC, each call class has its own queue and all classes share the available resources. The prioritization is implemented using only the queuing techniques where the queued calls with higher priority are served first. HP-CAC is a variant of QP-CAC, where different resource utilization thresholds (i.e., LM) are assigned to each traffic class based on each traffic priority. The higher priority class is assigned higher LM and allowed to share the resources of the lower class. Therefore, the prioritization is implemented in this scheme using both LM and queuing techniques. In these two schemes, the arrived call is queued in its corresponding queue, i.e., depending on its class, if no sufficient resource is available upon its arrival. The algorithm procedure is determined as follows: 1. class i calls are admitted if and only if the following criteria is satisfied • For HP-CAC Dgi þ gc 6 gi

ð3Þ

• For QP-CAC, all call classes have the same LM where g1 = g2 = g3 = g4 = gmax. Therefore, the admission criteria is, Dgi þ gc 6 gmax ð4Þ 2. When all resources are occupied, then after releasing a resource the next call to be served is the one with highest priority non-empty queue, i.e., the lower class index. 3. Any call class is deleted from its queue if it exceeds the queuing time limit. As a summarization, the call admission decision procedures of QP-CAC and HP-CAC are illustrated in Table 2.

ð1Þ

where Gi = W/Ri is the processing gain for the ith MS, Ri is the bit rate associated with the ith MS, and W is the chip rate of the WCDMA system. ei is the bit-energy to noisedensity (Eb/No) figure corresponding to the desired link

3.2. CP-CAC scheme In CP-CAC, each call class has its own queue and resource partition as shown in Fig. 3. The CP-CAC can guarantee the resource commitment, and therefore the

S.A. AlQahtani, A.S. Mahmoud / Computer Communications 30 (2006) 41–51

45

Table 2 Priority classes QP-CAC and HP-CAC ARRIVAL and DEPARTURE procedure ARRIVAL event of class i call

DEPARTURE event of class i call connection

Admit_Flag = False

In case of HP-CAC, or QP-CAC If (There are non-empty queues) Select non-empty Qj with the lowest index value -In case of HP-CAC If (gc + $gj) < gj Admit class j call Else Release the resource -In case of QP-CAC If (gc + $gj) < gmax Admit class j call Else Release the resource Else Release the resource End

In case of HP-CAC: If (gc + $gi) 6 gi Admit_Flag = True In case of QP-CAC: If (gc + $gi) 6 gmax Admit_Flag = True If Admit_Flag Admit class i call request Else If (Qi is not full) Insert class i call Else Reject class i call request End

QoS, to each traffic class, but may underutilize the resources. This scheme is explained as follows. The cell load is divided into four non overlapping load partitions, Li (i = 1, 2, 3, and 4), such that 4 X Li 6 gmax ð5Þ i¼1

The capacity size or load partition, Li, of each division is selected based on the traffic characteristics and the predefined QoS requirements of each service class [2]. The total current usage load occupied by each connected call class i, Oi, is defined as, Bi X mi 6 Li ð6Þ Oi ¼ ð1 þ f Þ 1 þ Gi =qi i¼1 where Bi is the number of currently connected class i call. In CP-CAC, The algorithm procedure is as follows: 1. Class i calls are admitted if and only if the current carried load occupied by this class is less than Li; i.e., when Eq. (6) is satisfied. 2. When all resources of class i calls are occupied, then arrival of class i calls is queued in the corresponding buffer and served using the FIFO policy. 3. Any call is deleted from its queue if it exceeds the queuing time limit.

As a summarization, the call admission decision procedures of CP-CAC are illustrated in Table 3. 3.3. DP-CAC scheme The problem with the above schemes is that, for HPCAC the unutilized loading limits for higher classes can not be used by lower class traffic and therefore the capacity is wasted. Also, when the traffic load presented by a higher priority class is much larger than that for a lower priority class, calls belonging to the higher class overwhelm those belonging to the lower class. In CPCAC the QoS for each class is guaranteed but this is achieved at the expense of inefficient total system utilization. Our proposed DP-CACA takes these two problems into account. We utilize the predefined load partition and the current system load to dynamically admit the queued calls. The priority is dynamically adjusted such that the higher priority class remains higher in priority as long as the lower priority class is not adversely affected by the higher priority class. For low to moderate loads, the total loading (i.e., total cell capacity) is available for all arrived call classes to enhance the resource utilization. However, at congestion condition the dynamic priority is used to differentiate between the queued

Table 3 Priority classes CP-CAC ARRIVAL and DEPARTURE procedures On ARRIVAL event of class i call

On DEPARTURE event of class i call connection:

If (Oi 6 Li) Admit class i call request Else If (Qi is not full) Insert class i call request into its queue Else Reject class i call request End

If (Qi > 0) Serve Qi based on FIFO Else Release the resource End

46

S.A. AlQahtani, A.S. Mahmoud / Computer Communications 30 (2006) 41–51

calls. The dynamic admission procedure can be stated as follows: 1. The arrived class i call is served as long as the criterion Eq. (4) is satisfied. 2. When all resources are utilized, then the arrived call is inserted in its corresponding queue. 3. Any call is deleted from its queue if it exceeds the queuing time limit. 4. Upon capacity release, dynamic priority value is computed for all call classes with non-empty queues. Then, the queue with least priority value is served first based on FIFO policy. The dynamic priority, utilized in step 4 of the above procedure, is calculated using the total load currently occupied by class i calls as in Eq. (6) and the load partition predefined for class i calls as in Eq. (4). Therefore, the dynamic priority value for class i calls is given by Pi ¼

Oi Li

ð7Þ

The traffic class i with minimum Pi value will has the highest priority. Here, as the total current load of class i calls drops below the predefined partition Li, its priority value will decrease and hence it will receive a high priority (the highest priority class is the one with minimum priority value). When two or more call classes have the same priority value, then the call with lower class index (higher priority) will be served first. Using these rules, the unutilized load of one traffic class can be utilized by other traffic classes when needed. Also, at high system load the priority value will prevent the classes from adversely affecting each other. As a summarization, the call admission decision procedures of DP-CAC are illustrated in Table 4. 4. Traffic model and performance measures

system is Poisson with rate k = kh1 + kh2 + kn1 + kn2. The channel holding time for each class of calls is exponentially distributed with mean l1 while the queuing time limit of each handoff calls class is exponentially distributed with mean c1 j . Using the above definitions, the generated load of each traffic class i, denoted by qi is given by qi ¼ l1  ki . The total offered load for the system, denoted i by q, is given by q = l1(kh1 + kh2 + kn1 + kn2). 4.2. System performance measures The system performance metrics that are used to evaluate our algorithms are grade of service (GoS), total carried traffic (CT), system utilization, and system revenue and they are defined as follows: (1) The grade of service metric is defined as: GoSj ¼ aP hj þ P nj

where Phj is the handoff failure probability, and Pnj is the new call blocking probability of calls belonging to (j = 1, 2) RT and NRT respectively. a = 10 indicates penalty weight for dropping a handoff call relative to blocking a new call. Smaller GoS means better system performance. (2) The total carried traffic (CT) is defined as: CT ¼ kh1 ð1  P h1 Þ þ kh2 ð1  P h2 Þ þ kn1 ð1  P n1 Þ þ kn2; j ð1  P n2 Þ

ð9Þ

(3) System utilization (U): the average number of connections in each traffic class that the system can accept for given average traffic is defined by ni. The bandwidth (bi) of connection i is defined by its load increment, Dgi, such that bi = Dgi. The average system utilization is defined as: U¼

4 X

ðni Þ  ðbi Þ

ð10Þ

i¼1

4.1. Traffic model The traffic model established in the most recent researches [5–7,14,15] is utilized for the study in this paper. We assume the arrival processes of new and handoff calls, belonging to the each call class, are Poisson occurring with rates, kh1, kh2, kn1, kn2, for RT handoff, NRT handoff calls, RT new calls and NRT new calls, respectively. The total arrival rate of the

ð8Þ

(4) System revenue (r): assuming that revenue is given by the value of assigned bandwidth unit (bi = Dgi), then the total revenue rate in the system is equal to the system bandwidth utilization in the system as follows: r¼

4 X

ðni Þ  ðri ¼ bi Þ

ð11Þ

i¼1

Table 4 Priority classes DP-CAC ARRIVAL and DEPARTURE procedures ARRIVAL event of class i call request:

DEPARTURE event of class i call connection:

If (gc + $gi) 6 gmax Admit class i call request Else If (Qi is not full) Insert class i call request into its queue Else Reject class i call request End

If (There are non-empty queues) Set: Pj = Oj/Lj Select non-empty Qj with the lowest Pj value If (gc + $gj) < gmax Admit class j call Else Release the resource Else Release the resource End

S.A. AlQahtani, A.S. Mahmoud / Computer Communications 30 (2006) 41–51

5. Simulation results 5.1. Simulation description and parameters We have simulated the above schemes by developing a C+ discrete-event simulation. The simulation follows closely and carefully the guidelines given in [18]. Table 5 describes the basic simulation’s events and Fig. 5 presents the general flowchart. There are four main events: INITIALIZE, ARRIVAL, DEPART and DROP events. The ARRIVAL and DEPART events are already described for each schemes in Tables 2–4. The selection of the next event is based on the event’s time. The simulation has four buffers (queues) implemented as FIFO queues. These queues hold the value of the arrival time and drop time of each call. The simulation run ends when the total number of served calls reaches five millions. The parameters for the traffic types used in this simulation are as specified in Table 6 while the physical layer parameters for the WCDMA network are as specified in Table 7. 5.2. Results and discussions In this section we compare the performance of DP-CAC with that for QP-CAC, HP-CAC, and CP-CAC under the same conditions in terms of system utilization, revenue and GoS guaranteed. This results section can be divided into two main parts, the first part analyzes system utilization and GoS provisioning, while the second part uses the fairness criterion to compare between the different CAC schemes. The system utilization provided by each scheme versus the offered traffic load is depicted in Fig. 6. The percentage of each arrival rate out of the total offered traffic is as follows: kh1 = 0.2k, kn1 = 0.3k, kh2 = 0.2k, and kn2 = 0.3k. One can note that DP-CAC and QP-CAC have identical capacity utilization and that utilization is higher than that provided by HP-CAC and CP-CAC. On the other hand, as expected CP-CAC has the lowest system utilization. It should be mentioned that though DP-CAC and QP-CAC have similar system utilization figure, DP-CAC differs from QP-CAC by dynamically

47

controlling priority level of queued calls and thus preventing one traffic class from being adversely affecting other classes. Figs. 7 and 8 show how the GoS of each service is affected by the traffic load of the other for all the CAC schemes. To obtain the corresponding GoS curves, we fix the traffic load of NRT services at 12 Erlangs and vary the RT traffic. Examining the GoS for NRT services shown in Fig. 7, we observe that the GoS of NRT services is not affected by the load increase of RT services for the case of CP-CAC. In addition, when RT traffic load is low or moderate compared to that for the NRT traffic, all schemes provide better GoS compared to CP-CAC, however, when the RT traffic load increases further, CP-CAC provides the best GoS followed by DPCAC. However, CP-CAC has the worst system utilization (see Fig. 6). From the perspective of GoS for RT services which is depicted in Fig. 8, as expected CS based algorithms such QP-CAC and HP-CAC provide the best GoS for most load conditions. This is due to the fact that RT traffic is always allowed to use any available resources. In addition RT handoff requests are given the highest priority amongst all RT services and allocated a higher weight in the GoS formulation. However considering the other two performance measures, GoS for NRT and system utilization, HP-CAC provides worst performance while QP-CAC provides worse performance in terms of GoS for NRT traffic. On the other hand, DP-CAC provides a comparable GoS for RT traffic while still outperforming QP-CAC and HP-CAC in terms of GoS for NRT and supporting the best utilization. Finally, while CPCAC provides the best GoS for NRT services, it results in the worst GoS for RT services and system utilization. Based on these results we conclude that DP-CAC provides the best service differentiation while still achieving the best system utilization and hence the best revenue for operator. Now we turn to investigate, how DP-CAC provides better balance between QP-CAC and CP-CAC in term of system resources utilized by each of RT and NRT traffics. In this, we focus on how DP-CAC provides a good balance at congestion case. The resources that are utilized individually by RT calls and NRT traffics using

Table 5 Simulation events Event

Description

INITIALIZE

Occurs only once at the beginning of simulation. This includes selecting the CAC scheme type (CP-CAC, QP-CAC, HP-CAC, or DP-CAC), and initializing the simulation parameters. Occurs when a new call arrives to the system. This includes the generation of next arrival event and the calculation of the current available resources to see whether to admit in queue or reject the arrived call. If a call is queued then its next drop time is generated. DEPART indicates the completion of call connection. This processing includes the selection of the next queued call request if any for service and the generation of the next departure event time. Occurs whenever a queued call requests exceeds its waiting time limit before getting the resource.

ARRIVAL

DEPART DROP

48

S.A. AlQahtani, A.S. Mahmoud / Computer Communications 30 (2006) 41–51 Main Program: Start Simulation

Procedure initialize: Initialize The Model 1. Set CLOCK=0. 2. Set Cumulative statistics to 0 3. Generate Initial events and TimeOf Next Event 4. Define Initial System State. 5. select the CAC scheme Obtain Input Parameters: 1. Read Initial Arrival Rate Of Each Call Class. 2. Read Call Holding time, cell residence time and Queuing time. 3. Read Queuing Capacity of each call. 4. Read MaxIterationValue N

Advance Time Procedure: 1. Find Imminent Event e. 2. Advance CLOCK to its Time 3. delete this event from event list.

DEPART

Check The Type of event

ARRIVAL

Call Arrival ProcedureYES

DROP NO

Resource available

Delete this call from its waiting queue Check if its Q =Full NO

YES if any of the waiting queues not empty then the one with higher priority will be served and Generate its Next DEPART event. Else if all queues are empty then release resource.

Admit this call

Block this call

i) Insert Call in its waiting queue ii) Generate its next DROP event.

Generate next ARRIVAL event:

Collect Statistics

NO

Simulation Over

YES

Compute Summary Statistics

Fig. 5. Simulation flowchart.

QP-CAC, HP-CAC, CP-CAC, and DP-CAC are depicted in Figs. 9 and 10. We assume that the RT traffics constitute 70% of the total traffic and the NRT traffics consti-

tute 30% of the total traffic. At high system loads, the resources that are utilized by NRT in case of HP-CAC and QP-CAC are very small because they are over-

S.A. AlQahtani, A.S. Mahmoud / Computer Communications 30 (2006) 41–51

49

Table 6 Characteristic of classes’ services Service class

Bit rates (kbps)

Required Eb/No (dB)

Activity factor

RT-voice (class 1 and class 3) RT-video streaming (class 1 and class 3) NRT (class 2 and class 4)

12.2 128, 384 64, 128, 256, 384

5 2.5, 2.0 3.5, 2.5, 2.0, 2.0

0.4 1 1

Table 7 Simulation parameters for the networks Parameter

Value

Radio access mode Chip-rate Thermal noise Dedicated channel rates Max cell load class 1, 2, 3, and 4 load limit class 1, 2, 3, and 4 load Margin(L1  L4) Call duration, queuing time Handoff calls Queue size Call Arrival

WCDMA (FDD) uplink with perfect power control 3.84 Mbps 1.0e15 W 12.2, 64, 128, 256, 384 kbps 80% of pole capacity 80%, 70%, 50%, 50% 0.3, 0.2, 0.15, 0.15 Exponential (120, 100, 15 s) 5–10 Poisson

1

10

0.9

8

0.7 0.6

QP-CAC HP-CAC CP-CAC DP-CAC

0.5 0.4

6

RT GoS

Utilization

0.8

4

QP-CAC CP-CAC HP-CAC DP-CAC

2

0.3

0 0

20

40

60

80

100

0

10

20

30

40

50

60

RT Offered traffic (Erl.), (NRT= 12 Erl)

Offered traffic (Erl.)

Fig. 8. GoS of RT services vs. its traffic load at (12 Erl. Traffic load for NRT).

Fig. 6. Cell capacity utilization (revenue).

1

11

RT calls resource utilization

10

NRT GoS

9 8 7 QP-CAC CP-CAC HP-CAC DP-CAC

6 5

0.8

0.6

0.4 QP-CAC-RT HP-CAC-RT CP-CAC-RT DP-CAC-RT

0.2

4

0

3 0

10

20

30

40

50

60

RT Offered traffic (Erl.), (NRT= 12 Erl)

Fig. 7. GoS of NRT services as a function of RT traffic loads.

0

20

40

60

80

100

120

Offered traffic (Erl.)

Fig. 9. RT resource utilization versus offered load.

50

S.A. AlQahtani, A.S. Mahmoud / Computer Communications 30 (2006) 41–51

References

NRT calls resource utilization

0.7 QP-CAC-RT HP-CAC-RT CP-CAC-RT DP-CAC-RT

0.6 0.5 0.4 0.3 0.2 0.1 0 0

20

40

60

80

100

120

Offered traffic (Erl.)

Fig. 10. NRT resource utilization versus offered load.

whelmed by the high traffic of RT while in case of CPCAC there is no effect. DP-CAC provides better balance between these two schemes and ensures that each traffic class can utilize sufficient resource at congestion condition. 6. Conclusion The design of QoS-aware CAC is a critical issue for 3G and beyond systems supporting heterogeneous traffics. In this paper, we extended two resource sharing algorithms known as complete sharing and complete partitioning and utilized them in the design of call admission control for the WCDMA network. These algorithms are QP-CAC (and its variant HP-CAC) and CP-CAC. Furthermore, we designed a novel CAC algorithm, referred to by DP-CAC, to overcome the shortcomings of the previous algorithms. The novel algorithm dynamically prioritizes the different traffic classes based on the predefined sharing limit and the current load. As shown by the results, DP-CAC provides an acceptable QoS for each traffic class and prevents a higher traffic class from overwhelming lower traffic class to enhance the fairness. CS-based algorithms achieve the best system utilization at the expense of serious unfairness for the traffic classes with different QoS requirements. DP-CAC manages to attain equal system utilization to CS based algorithms but without the drawbacks in terms of fairness and service differentiation. In the limit, DP-CAC reduces to QP-CAC at light loads and to CP-CAC at high traffic load from all call classes. DP-CAC succeeds to provide a better balance between the CS and CP strategies. Acknowledgement The authors acknowledge King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia for support.

[1] F.S. Lai, J. Misic, S.T. Chanson, Complete sharing versus partitioning: quality of service management for wireless multimedia networks, in: Proceeding of 7th International conference on computer communications and networks, Lafayette, LA, October 1998, pp. 584–593. [2] B. Epstein, M. Schwartz, Reservation strategies for multimedia traffic in a wireless environment, in: Proceedings of IEEE Vehicular Technology Conference, July 1995, pp. 165–169. [3] H. Holma, A. Toskala (Eds.), WCDMA for UMTS: Radio Access for Third Generation Mobile Communications, John Wiley and Sons, Ltd, England, 2000. [4] H. Mohamed, Call admission control in wireless networks: a comprehensive survey, IEEE Communications Surveys & Tutorials, First Quarter 2005, pp. 50–69. [5] C.W. Leong, W. Zhuang, Y. Cheng, L. Wang, Optimal resource allocation and adaptive call admission control for voice/data integrated cellular networks, IEEE Transactions on Vehicular Technology 55 (2) (2006) 654–669. [6] X. Ma, Y. Cao, Y. Liu, K.S. Trivedi, Modeling and performance analysis for soft handoff schemes in CDMA cellular systems, IEEE Transactions on Vehicular Technology 55 (2) (2006) 670–680. [7] M. Ghaderi, R. Boutaba, Call admission control for voice/data integration in broadband wireless networks, IEEE Transactions on Mobile Computing 5 (3) (2006) 193–207. [8] S.M. Shin, C. Cho, D.K. Sung, Interference-based channel assignment for DS-CDMA cellular systems, IEEE Transactions on Vehicular Technology (1999) 233–239. [9] Z. Liu, M.E. Zarki, SIR-based call admission control for DS-CDMA cellular systems, IEEE Journal on Selected Area in Communications 12 (4) (1994) 638–644. [10] J. Kuri, P. Mermelstein, Call admission on the uplink of a CDMA system based on the total received power, in: Proceedings of ICC, vol. 3, Vancouver, BC, Canada, 1999, pp. 1431–1436. [11] R. Fantacci, G. Mennuti, D. Tarchi, A priority based admission control strategy for WCDMA systems, IEEE International Conference on Communications, 16–20 May 2005, vol. 5, 2005, pp. 3344–3348. [12] Kuenyoung Kim, Youngnam Han, A call admission control with thresholds for multi-rate traffic in CDMA systems, in: Vehicular Technology Conference Proceedings, 2000, VTC 2000-Spring Tokyo, 2000 IEEE 51st, vol. 2, May 2000, pp. 830–834. [13] Wang Ying, Wang Weidong, Zhang Jingmei, Zhang Ping, Admission control for multimedia traffic in CDMA wireless networks, Communication Technology Proceedings, International Conference, 9–11 April 2003, vol. 2, 2003, pp. 799–802. [14] O. Yu, E. Saric, Anfei Li, Adaptive prioritized admission over CDMA, IEEE Wireless Communications and Networking Conference 2 (2005) 1260–1265, 13–17. [15] O. Yu, E. Saric, A. Li, Fairly adjusted multimode dynamic guard bandwidth admission control over CDMA systems, IEEE Journal on Selected Areas in Communications 24 (3) (2006) 579–592. [16] Huan Chen, S. Kumar, C.-C.J. Kuo, Dynamic call admission control scheme for QoS priority handoff in multimedia cellular systems, IEEE Wireless Communications and Networking Conference (WCNC2002) 1 (2002) 114–118, 17–21. [17] C. Huan, K. Sunil, C.-C. Jay, QoS-aware radio resource management scheme for CDMA cellular networks based on dynamic interference margin (IGM), Computer Networks 46 (2004) 867–879. [18] M. Law, W. Kelton, Simulation Modeling and Analysis, third ed., McGraw-Hill, New York, 2000.

S.A. AlQahtani, A.S. Mahmoud / Computer Communications 30 (2006) 41–51 Salman Alqhtani is currently pursuing his Ph.D. studies in Department of Computer Engineering, KFUPM, Saudi Arabia. He obtained his Bachelor degree with 1st honor in computer engineering from KFUPM, and Masters degree in Computer Engineering with 1st honor from KSU, Riyadh. His major research interests include Computer Networks performance, Wireless Networks, and Radio Resource Management.

51

Ashraf S. Mahmoud received the B.Sc. degree in Electrical and Computer Engineering from Kuwait University in 1990, the M. Eng. in Engineering Physics (Computer Systems) from McMaster University, Hamilton, Canada in 1992. He received his Ph.D. in Systems and Computer Engineering from Carleton University, Ottawa, Canada in 1997. During 1997–2002, he was with Nortel Networks Research and Development where he focused on development and evaluation of radio resource management algorithms for broadband and 3G networks. Since 2002, he is with the Computer Engineering department at King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia as an assistant professor. His research interests include radio resource management for 3rd and 4th G networks, wireless local are networks and integration of heterogonous networks.