A Novel Radio Admission Control Scheme for Multiclass Services in LTE Systems Manli Qian1,2 , Yi Huang1,2 , Jinglin Shi1 , Yao Yuan1,2 , Lin Tian1 , Eryk Dutkiewicz3
1
(Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China, 100190) 2 (Graduate University of Chinese Academy of Sciences, Beijing, China, 100049) Email: {qianmanli, huangyi01, sjl, yuanyao, tianlindd}@ict.ac.cn 3 (Department of Physics and Electronic Engineering, Macquarie University, Sydney, Australia) Email:
[email protected]
Abstract—In this paper, a novel radio admission control (RAC) scheme is proposed for handling multiclass services in Long Term Evolution (LTE) systems. An objective function of maximizing the number of admitted users is proposed to evaluate the system capacity. To solve the optimization problem, we present a combined complete sharing (CS) and virtual partitioning (VP) resource allocation model and develop a service degradation scheme in case of resource limitations in the proposed RAC scheme. Call blocking probability, system resource utilization and system capacity are used as performance metrics and are evaluated by using a K-dimensional Markov Chain model. Numerical results show that an optimal proportion of resource deployment for different service groups can be identified to maximize system capacity while at the same time maintaining quality of service (QoS) constraints of all admitted users.
I. INTRODUCTION The requirement of anytime, anywhere access in wireless networks drives increasing demand for new applications that utilize the triple play of simultaneous voice, video and data sessions. To meet the ever-growing demand of consumers, the Long Term Evolution (LTE) started out as an evolutionary approach by the 3rd Generation Partnership Project (3GPP) in 2004 to deliver faster data rates, improved capacity and coverage by creating a new radio access technology optimized for IP-based traffic and a decentralized radio access network (RAN) architecture while providing users a simple upgrade path from 3G networks [1]. The radio resource management (RRM) in LTE aims at providing multiclass services such as data, voice, video, gaming, ftp, etc., which have differentiated quality of service (QoS) requirements. As a key component of RRM, the task of radio admission control (RAC) is to admit or reject a new connection request depending on whether the required QoS of the new connection request will be fulfilled while guaranteeing the required QoS of all in-progress sessions [2]. So the RAC scheme is extremely important in RRM, because if all connection requests are admitted, the QoS of each admitted user may otherwise not be satisfied. In LTE a bearer-level service architecture is introduced to provide end-to-end service and the goal of RAC is to determine whether a new radio bearer can be admitted into the system. The bearer level QoS parameters are QCI (QoS Class Identifier), ARP (Allocation and Retention Priority),
GBR (Guaranteed Bit Rate) and AMBR (Aggregate Maximum Bit Rate)[2][3]. QCI is a scalar that is used as a reference to access node-specific parameters that control bearer level packet forwarding treatment. ARP contains information about the priority level, the preemption capability and the preemption vulnerability and is used to decide whether a bearer can be accepted or needs to be rejected in case of resource limitations. GBR and AMBR identify the transfer rate of a bearer. Hence, the bearer-level parameters ARP, GBR and AMBR are primarily used in making admission control decisions. Many admission control (AC) schemes have been proposed with QoS constraints for multiclass services in wireless networks. According to the manner of resource allocation, they can be categorized into three basic classes: complete sharing (CS) scheme, complete partitioning (CP) scheme and virtual partitioning (VP) scheme. The main concept of CS is that many users share a common resource. If the required resource is satisfied, the call will be admitted into the system, otherwise it will be rejected. In [5], [6] and [7] the authors proposed a CS resource allocation scheme in making AC decisions. The drawback of this scheme is that it treats each call equally and does not consider fairness of different calls. A CP based AC scheme is also proposed in [5] and [7] to serve multiclass services. In this scheme the overall resources are partitioned into several parts according to the service type and the new call connection request will be rejected if the resource allocated to the corresponding service type is used up. The CP scheme ensures the fairness of different priority calls, but due to the fixed partition, the radio resource utilization would be decreased. The VP scheme proposed in [8], [9] and [10] manages to combine the advantages of CS and CP. It strikes a balance between unrestricted sharing in CS and unrestricted isolation in CP. The VP scheme divides multiclass traffic into two groups and each group is allocated a nominal amount of resources with the provision that underutilized resources can be used by the excess traffic of an overloaded class, subject to preemption. This scheme solved the problem in [5]. But the preemption scheme may lead to a lower capacity of the system. So the current AC schemes in literature may not be suitable for LTE systems. In this paper, we construct a combined CS and VP resource allocation model and propose a novel RAC scheme with
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
service degradation for handling multiclass services in LTE systems. The main contribution of the proposed RAC scheme is that it can achieve higher system capacity and at the same time maintain the QoS requirements of different service classes. The rest of the paper is organized as follows. Section II describes the system model for multiclass services and proposes an optimization problem. Section III gives a detailed description of the proposed RAC scheme. The system performance is evaluated by using a K-dimensional Markov Chain Model in Section IV and numerical results are presented in Section V. Section VI concludes the paper. II. SYSTEM MODEL For simplicity, we consider a generic cell with a total capacity of bandwidth C that can support K different classes of multimedia service. The system parameters and bearer level parameters we use throughout this paper are listed below: C α C1 C2 K
p nk rk rkg rkmax Bk B Uk U
total amount of bandwidth available in the system; the ratio of nominal bandwidth to system capacity for Group 1; nominal bandwidth for Group 1, where C1 = α ∗ C; nominal bandwidth for Group 2 and 3, where C1 + C2 = C; total number of service classes; Group 1 includes traffic classes from 1 to l, Group 2 includes traffic classes from l + 1 to m and Group 3 includes traffic classes from m + 1 to K; priority of each service class; instantaneous user number of service class k in the system; instantaneous bandwidth allocated to service class k user; guaranteed bit rate of service class k; maximum bit rate of service class k; blocking probability of service class k; total blocking probability of the system; system resource utilization of service class k; total system resource utilization.
According to the bearer level QoS parameter ARP described in [4], the preemption capability information defines whether a service data flow can get resources that were already assigned to another service data flow with a lower priority level. The preemption vulnerability information defines whether a service data flow can lose the resources assigned to it in order to admit a service data flow with a higher priority level. Thus, we categorize the multiclass services into three groups. Group 1 represents the services whose resources can be preempted. Group 2 and 3 are for services whose resources can not be preempted while group 2 can preempt the resources that are allocated to services in group 1. Furthermore, we consider group 3 to be traffic flows that need to provide a guaranteed bit rate while service groups 1 and 2 need to provide a variable bit rate. These are respectively referred to, as ”guaranteed bit rate” and ”maximum bit rate”.
We adopt the basic concept of the CS and VP schemes and set up the system model for resource allocation in our proposed radio admission control scheme as follows. According to the feature of each service group mentioned above, within service group 1, we have CS, that is, each service class shares the nominal bandwidth C1 . Within service groups 2 and 3, we also have CS and the nominal bandwidth C2 is completely shared. Within group 1 and 2, we have VP, that is, when the nominal bandwidth C2 is fully utilized, we can also admit some traffic of service group 2 subject to preemption by degradation of the services in group 1 through which the system capacity can be increased. The objective of our proposed RAC scheme is to find out an optimal proportion of resource deployment for different service groups to maximize system capacity while guaranteeing the QoS requirements of each service class. We use the concept of admitted user number (Na ) to denote the system capacity. As a result, the objective function is given by f = max (Na ) 0 ≤ α ≤ 1 s.t. B ≤ BT
(1a) (1b)
where BT is the threshold for the system blocking probability. To solve the optimization problem, we need to determine the number of admitted users by using the proposed RAC scheme. Before doing so the next section first presents a detailed description of the proposed RAC scheme. III. PROPOSED RADIO ADMISSION CONTROL SCHEME A. Radio Admission Control Rules Based on the combined CS and VP resource allocation model described in Section II, for a new call request in service class k, we have the following admission rules: 1) For a new call in service class k that belongs to service group 1, if the remaining bandwidth in Group 1 is greater than or equal to rkg , the call is admitted into the system; otherwise the call is rejected. The allocated bandwidth rk is given by ⎧ ni l ⎪ ⎪ ⎪ rkg , rj + rkg ≤ C1 ⎪ ⎪ ⎨ i=1j=1 ni l rk = (2) ⎪ rkmax , rj + rkmax ≤ C1 ⎪ ⎪ ⎪ i=1j=1 ⎪ ⎩ 0, otherwise 2) For a new call in service class k that belongs to service group 2, if the remaining bandwidth in Group 2 and 3 is greater than or equal to rkg , the call is admitted into the system and the allocated bandwidth rk is given by ⎧ ni K g ⎪ ⎪ rj + rkg ≤ C2 ⎨ rk , i=l+1j=1 rk = (3) ni K ⎪ max max ⎪ ⎩ rk , rj + rk ≤ C2 i=l+1j=1
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
3) For a new call in service class k that belongs to service group 2, when the remaining bandwidth in Group 2 and 3 is less than rkg , according to preemption over Group 1, we assume that the bandwidth acquired by using the service degradation scheme described in the following subsection is Ra . If Ra is greater than or equal to the required bandwidth, the call is admitted into the system; otherwise the call is rejected. The allocated bandwidth is given by ⎧ ni K ⎪ ⎪ ⎪ rj + rkg > C2 and ⎪ ⎪ ⎪ ⎪ i=l+1 j=1 ⎨ g rk , ni K rk = g ⎪ ⎪ ≥ r − max(C − rj , 0) R ⎪ a 2 k ⎪ ⎪ ⎪ i=l+1 j=1 ⎪ ⎩ 0, otherwise (4) 4) For a new call in service class k that belongs to service group 3, if the remaining bandwidth in Group 2 and 3 is greater than or equal to rkg , the call is admitted into the system; otherwise the call is rejected. The allocated bandwidth rk is given by ⎧ ni K ⎨ g rj + rkg ≤ C2 rk , rk = (5) i=l+1j=1 ⎩ 0, otherwise B. Service Degradation Scheme According to our proposed RAC rules, when bandwidth C2 is fully utilized by service group 2 and 3, we can also admit some traffic in service group 2 by preemption of the bandwidth that was already assigned to traffic in service group 1. There are two basic preemption schemes: cutoff scheme[10] and service degradation scheme[11][12]. The main concept of the cutoff scheme proposed in [10] is that when executing preemption, the total bandwidth assigned to traffic in service group 1 will be preempted and as a result its QoS may no longer be satisfied. With the aim to maximize the system capacity while maintaining the basic QoS of all admitted users, we use a service degradation scheme when executing preemption and the proposed service degradation scheme must follow the following rules: 1) Service group 2 users can only preempt the bandwidth which has been assigned to the admitted users in service group 1, and not directly preempt the remaining bandwidth allocated to service group 1; 2) Preemption can only happen to the admitted users in service group 1 whose priority level is lower than the requested user in service group 2; 3) When preempting the bandwidth that has been assigned to users in service group 1, we just degrade the bandwidth assigned to each user with a degradation level ri − rig ; 4) When executing degradation, we should degrade the user with the lowest priority each time.
Furthermore, we assume that the required bandwidth is Rreq , the priority threshold is PT and the available bandwidth after service degradation is Ra . Then the details of the service degradation scheme over Group 1 are given as follows: Step 1 According to the priority level of each admitted user from service class 1 to l in Group 1, sort them in a priority decreasing order; Step 2 Select a user i with lowest priority from the set where the bandwidth has not been degraded and the priority level is lower than PT ; Step 3 Execute service degradation with a degradation level ri − rig , then Ra + = ri − rig ; Step 4 If Ra < Rreq , the procedure goes back to Step 2 until no in-progress user is available upon which the procedure ends. Otherwise the procedure ends. The proposed RAC scheme is composed of the radio admission rules and the service degradation scheme. In the next section, we give a detailed description of the performance analysis parameters by applying the proposed RAC scheme. IV. PERFORMANCE ANALYSIS MODEL As generally accepted, we assume that the arrivals of calls are Poisson distributed and call holding times follow the exponential distribution. For a service class k call, the arrival rate is λk and mean call-holding time is μ−1 k . Then, the performance analysis can be done based on a K-dimensional Markov Chain Model. Let n = (n1 , n2 , . . . , nK ) denote the state of the system with the number of users nk (0 ≤ k ≤ K) for class k traffic. Then, the state space Ω for the K-dimensional Markov Chain is obtained based on the following equations: ⎧ ni K ⎪ ⎪ rj ≤ C ⎪ ⎪ ⎪ i=1j=1 ⎪ ⎪ ⎨ ni l rj ≤ C1 ⎪ i=1j=1 ⎪ ⎪ ⎪ ni K ⎪ ⎪ ⎪ rj ≤ C2 ⎩
(6)
i=m+1j=1
where
ni
rj denotes the total bandwidth assigned to service
j=1
class i. We can specify the RAC scheme by mapping f(n) = (f1 (n), f2 (n), . . . , fK (n)) for each service class of calls. For a system state n in system state space Ω, fk (n) takes on the value 0 or 1 if a class k call is rejected or admitted, respectively. They are defined by 1, condition 1 (7) fk (n) = 0, otherwise where condition 1 refers to the situations that rk = 0 in equations (2) to (5). Let P (n) denote the equilibrium probability that the system is in state n. Then, the global-balance equation for the K-
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
where θk denote the probability that the next arrival call belongs to service class k and is given by
dimensional Markov Chain can be formulated as: K [λk fk (n) + nk μk ]P (n) k=1 K
=
P (n − ek )λk fk (n − ek )
θk = λ k /
(8)
+
P (n + ek )nk μk
k=1
where ek is a K-dimensional vector of all zeros except for a value of one in the kth place. nk + ek means that a call in service class k should be admitted while nk − ek means the termination of a call in service class k. With the normalized condition that the sum of all the state probabilities in state space Ω equals to 1, we can solve equation (8) and obtain the steady state probabilities of all the states in state space Ω. Based on these we can then determine the system performance parameters of interest. First, we define some parameters that may be used in system performance analysis. For a service class k belonging to group 2, we define the number of admitted users by sharing bandwidth with users in service group 3 as nsk and the number of admitted users by preempting the bandwidth assigned to users in service group 1 as ndk . Then the total number of admitted users nk is given by nk = nsk + ndk
n1 =0
n2 =0
max C2 /rl+1 +
ni l
rkg and
ni l i=1j=1
i=l+1j=1
(C1 −
ni ri
i=1
...
for service class k that belongs to service group 3, Bk is calcum n K si lated for all n satisfying (C2 − rj − ni rig ) < rkg . i=l+1j=1
i=m+1
Then the total call blocking probability can be calculated by B=
K
Bk
(12)
k=1
B. System Resource Utilization The system resource utilization for service class k is given
Uk =
n1 =0
max C2 /rl+1 +
ni
i=1
nl =0
g (rj −rjg )/rl+1
i=1j=1
)/rl
...
nl+1 =0
m−1
(C2 −
max nsi ri )/rm + (
ni l
m−1
(rj −rjg )−
i=1j=1
i=l+1
g (rj −rjg )/rl+1
ni rimax )/rlmax
...
n2 =0 l
l−1
(C1 −
C1 /r1max (C1 −n1 r1 )/r2max
nl =0
i=m+1
(rj − rjg ) < rkg ;
C1 /r1max (C1 −n1 r1 )/r2max
Bk =
i=1j=1
for service class k that belongs to service group 2, Bk is calm n K si culated for all n satisfying (C2 − rj − ni rig )