Efficient cooperative access class barring with load balancing and traffic adaptive radio resource management for M2M communications over LTE-A

Computer Networks 73 (2014) 268–281

Contents lists available at ScienceDirect

Computer Networks journal homepage: www.elsevier.com/locate/comnet

Efficient cooperative access class barring with load balancing and traffic adaptive radio resource management for M2M communications over LTE-A Yi-Huai Hsu, Kuochen Wang ⇑, Yu-Chee Tseng Department of Computer Science, National Chiao Tung University, Hsinchu 300, Taiwan

a r t i c l e

i n f o

Article history: Received 27 January 2014 Received in revised form 9 June 2014 Accepted 22 July 2014 Available online 3 September 2014 Keywords: Access class barring LTE-A M2M communications Random access Radio resource management

a b s t r a c t We propose two efficient cooperative access class barring with load balancing (CACB-LB) and traffic adaptive radio resource management (TARRM) schemes for M2M communications over LTE-A. The proposed CACB-LB uses the percentage of the number of MachineType Communication (MTC) devices that can only access one eNB between two adjacent eNBs as a criterion to allocate those MTC devices that are located in the overlapped coverage area to each eNB. Note that an eNB is a base station of LTE-A. In this way, the proposed CACB-LB can achieve better load balancing among eNBs than CACB, which is the best available related work. The proposed CACB-LB also uses the ratio of the channel quality indication that an MTC device received from an eNB over the number of MTC devices that attach to the eNB as a criterion to adjust the estimated number of MTC devices that may access the eNB. As a result, the proposed CACB-LB can have a better set of barring rates of access class barring than CACB and can reduce random access delay experienced by an MTC device, which is also applicable to user equipment (UE). After an MTC device successfully accesses to an eNB, the eNB needs to allocate radio resources for the MTC device. In addition, the proposed TARRM allocates radio resources for an MTC device based on the random access rate of the MTC device and the amount of data uploaded and downloaded by the MTC device in a homogeneous MTC device network, and the priority of an MTC device in a heterogeneous MTC device network. Furthermore, we use the concept from cognitive radio networks such that if there are unused physical resource blocks (PRBs) of UEs, an eNB can schedule MTC devices to use these PRBs to enhance the throughput performance. Simulation results show that either in a homogeneous MTC device network or in a heterogeneous MTC device network, the proposed CACB-LB’s average access delay of UEs/MTC devices and average throughput from UEs/MTC devices are better than CACB’s. The proposed CACB-LB with TARRM’s average throughput from UEs/MTC devices is also higher than CACB’s. Therefore, the proposed CACB-LB with TARRM is feasible for M2M communications over LTE-A. Ó 2014 Elsevier B.V. All rights reserved.

1. Introduction

⇑ Corresponding author. E-mail addresses: [email protected] (Y.-H. Hsu), [email protected] (K. Wang), [email protected] (Y.-C. Tseng). http://dx.doi.org/10.1016/j.comnet.2014.07.017 1389-1286/Ó 2014 Elsevier B.V. All rights reserved.

Machine-to-Machine (M2M) communications need the support of high coverage wireless networks, such as LTEA (Long Term Evolution-Advanced) cellular networks. The deployment of LTE-A is growing rapidly because of its

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281

better coverage, higher data rate, and lower network deployment cost. In this paper, we assume the M2M communications are through LTE-A. However, as the number of MTC devices which needs to upload/download data to/ from eNBs of LTE-A increases so greatly, it results in severe congestion in a physical random access channel (PRACH), which is used by UEs or MTC devices to send a preamble to an eNB for initial access, handover, and connection reestablishment. The 3GPP (3rd Generation Partnership Project) has raised the need to revisit the design of next generations of cellular networks in order to make them capable and efficient to provide M2M services [16]. One of the key challenges that has been identified is the need to enhance the operation of the random access channel of LTE-A [16]. The current mechanism to request access to the system is known to suffer from congestion and overloading in the presence of a huge number of MTC devices [16]. Therefore, the 3GPP proposed the access class barring (ACB) scheme [10] to resolve the congestion problem in the PRACH. In the ACB, classes 0–9 are randomly assigned to commercial users, and classes 10–15 are assigned for special services [10]. In this paper, we assume that UEs belong to class 0, and MTC devices belong to class 1 for a homogeneous MTC device network and are classified into four classes (1–4) for a heterogeneous MTC device network. In the ACB, when the random access procedure is initiated, an eNB broadcasts a mean durations of access control and barring rate p 2 [0, 1] for each class. An MTC device or UE draws a uniform random number q between 0 and 1 when initiating connection establishment and then compares q with the current barring rate p to determine whether it is barred or not. If q 6 p, the MTC device or UE is allowed to perform the random access procedure with the eNB. Otherwise, the MTC device or UE is blocked for a certain period. Therefore, the eNB can control the congestion by adjusting p. In addition, since M2M features a large number of devices, a mechanism to guarantee the performance of Human-to-Human (H2H) and M2M in the random access procedure of LTE-A should be considered [17]. That is, the MTC device should not affect the UE in terms of performance. Thus, how to efficiently handle the congestion in the PRACH and guarantee the performance of the UE has become a critical issue when designing M2M communications over LTE-A. The motivation of this paper is to conquer the congestion and overloading problem of MTC devices and UEs in the PRACH and guarantee the UE’s performance as well. To achieve the concept of the smart city, people focus on applying next generation information technologies in our daily life and forming Internet of Things (IoTs), which can be achieved by utilizing emerging technologies such as M2M communications and cloud computing [1]. M2M communications will be one of the main focuses in LTE-A, as the need for M2M communications is increasing so rapidly in recent years and might finally exceed that of H2H communications [2]. In M2M communications, MTC devices communicate with each other and exchange data without direct human intervention [3]. To enable full automation of smart cities, each MTC device can play multiple roles among a sensor, a decision maker, and an action executor [4].

269

In the first generation of M2M communications, a critical part of M2M leverages the short message service (SMS) as a communication means, and the MTC device only transmits a small amount of data each time. It may not be a problem when there are few MTC devices in the network [2]. In the mobile network, there is rapid growth of data traffic in the world. This is because increase of smart phones and video contents results in rapid growth of data traffic. In the near future, M2M communications are also going to increase the data traffic significantly [6]. Thus, managing a large number of MTC devices is one key future task [5]. Nowadays, the number of available M2M services is rapidly increasing in cellular communication systems. In order to guarantee a maximum system capacity, the impact of this special kind of traffic on H2H communication needs to be analyzed [7]. Supporting trillions of MTC devices is a critical challenge in M2M communications, which may result in severe congestion in random access channels of cellular systems, such as LTE-A, that have been recognized as promising scenarios enabling M2M communications [8]. In this paper, we propose two efficient load balancing cooperative access class barring (CACB-LB) and traffic adaptive radio resource management (TARRM) schemes for M2M communications, called CACB-LB with TARRM, over LTE-A. Our core objective is seeking a better set of barring rates of access class barring in order to achieve load balancing CACB so as to resolve the congestion of MTC devices and UEs in the PRACH and reduce the random access delay experienced by MTC devices or UEs while guaranteeing the performance of UEs, and improving the throughput performance as well. To the best of our knowledge, no existing work integrates access class barring and radio resource management for M2M communications over LTE-A. The rest of this paper is organized as follows. Section 2 briefly discusses related work. Section 3 describes the problem formulation. Section 4 presents the proposed CACB-LB with TARRM. In Section 5, an analytic model is described. Performance evaluation results, including analytic and simulation results, are discussed in Section 6. Section 7 gives some concluding remarks and future work.

2. Related work Lien et al. [4] proposed a group-based radio resource management, which allocates PRBs (physical resource blocks) according to the packet arrival rate and delay jitter of the MTC device. Although this approach may increase throughput performance, its improvement may be restricted because of the packet arrival rate and delay jitter of the MTC device cannot fully reflect the real random access rate and the amount of data uploaded and downloaded. In addition, it did not mention how to resolve the congestion problem in a PRACH. Lee et al. [9] used different combinations of H2H’s and M2M’s preambles based on access barring check to evaluate throughput performance. However, it did not propose any mechanism to resolve the congestion in a PRACH or any radio resource scheduling method to enhance throughput performance. Lien et al. [8] proposed a CACB scheme, which enables eNBs to cooperate with each other so that they can obtain a

270

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281

Table 1 Comparison of the proposed CACB-LB with TARRM with related work. Method

Grouping-based radio resource management [4] Access barring check [10] CACB [8]

Pang et al. [17]

Lin et al. [18]

Proposed CACBLB with TARRM

Feature Design focuses/improvements

Average throughput (for both UEs and MTCs)

Average access delay (for both UEs and MTCs)

UEs mixed with MTC devices

Multiple classes of MTC devices

MTC devices are grouped according to the packet arrival rate and delay jitter

Medium

High

Yes

No

Using different combinations of H2H’s and M2M’s preambles based on access barring check to evaluate network performance Enabling eNBs to cooperate with each other so that they can obtain a set of barring rates of access class barring to significantly reduce the congestion in a PRACH Using a game-theoretic framework to divide random access resources (preambles) into three groups: for H2H, for M2M, and for hybrid usage Using dedicated allocated PRBs of PRACH to each class of MTC device and dynamically adjusted the barring time of a MTC device according the loading of its serving eNB 1. Using the number of MTC devices that can only access one eNB as a criterion to distribute those MTC devices that may access multiple eNBs and the number of MTC devices that attach to an eNB and the channel quality indication that an MTC device received from an eNB as criteria to adjust the estimated number of MTC devices that may access an eNB to obtain a better set of barring rates of access class barring than CACB 2. MTC devices and UEs use different PRBs of a PRACH and separate preambles to perform random access with eNBs 3. MTC devices are clustered according to their random access rates and the amount of data uploaded and downloaded in a homogeneous MTC device network, and their priorities in a heterogeneous MTC device network

Low

High

Yes

No

Medium (MTCs only)

Medium (MTCs only)

No

No

NA

NA

Yes

No

NA

NA

Yes

Yes

High

Low

Yes

Yes

set of barring rates of access class barring to reduce the congestion in a PRACH. However, the scheme may allocate MTC devices that may access multiple eNBs to an eNB that is not possible for these MTC devices to access. Furthermore, the scheme did not use the number of MTC devices that attach to an eNB and the channel quality indication that an MTC device received from an eNB as criteria to modify the estimated number of MTC devices that may access an eNB. The scheme may restrict the improvement of access delay of each MTC device or UE. Moreover, the paper did not provide a radio resource management method to improve throughput performance, it did not consider the situation that UEs are mixed with MTC devices, and it did not address the situation when there are multiple classes of MTC devices. Pang et al. [17] used a game-theoretic framework to divide random access resources (preambles) into three groups: for H2H, for M2M, and for hybrid usage, to achieve RACH contention resolution. Lin et al. [18] dedicated allocated PRBs of PRACH to each class of MTC devices and dynamically adjusted the barring time of an MTC device according the loading of its serving eNB. Because our work focuses on obtaining a better set of barring rates (p) of access class barring (ACB) to resolve PRACH contention, which is different from the focuses of these two related work, we only compared our work with the best available related work, CACB [8].

The proposed CACB-LB with TARRM can reduce the random access delay experienced by the MTC device or UE and can improve the throughput performance. Table 1 summarizes the differences between the proposed CACB-LB with TARRM and related work. Note that the proposed CACBLB with TARRM also addresses the support of a heterogeneous MTC device network, which has multiple classes of MTC devices. The related work only addresses the support of a homogeneous MTC device network. 3. Problem formulation Consider the random access of M2M communications in LTE-A with M eNBs, indexed by m = 1, . . . , M and N active MTC devices, indexed by n = 1, . . . , N. In [8], Lien et al. assume that an MTC device is able to access an eNB unattached by the MTC device when the MTC device is located within the overlapped coverage area of multiple eNBs. This part is different from the access class barring in [10], which assumes each MTC device can only access an eNB attached by the MTC device [8]. To formulate such a random access problem, the following notations are used [8]. Note that an eNB attached by an MTC device means that the MTC device registers with the mobility management entity (MME), which is used to monitor the location and connection status of an MTC device, via that eNB. Conversely, an eNB unattached by an MTC device means that the MTC device

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281

An MTC device that can only access the macrocell

271

Macrocell

X2

eNB

Picocell

X2

eNB

eNB An MTC device that can access the macrocell or the picocell

Picocell An MTC device that can only access the picocell

Fig. 1. Architecture of an LTE-A cellular network [8].

could receive signals from the eNB, but the MTC device does not register with the MME via that eNB. (1) Denote Am as the set of MTC devices that attach to the mth eNB and ||Am|| as the norm of Am, which represents the number of MTC devices that attach to the mth eNB. With the ability to cooperate with other eNBs, Am for all m are known by all eNBs. Note that PM m¼1 kAm k ¼ N. (2) Denote Mn as the set of eNBs that the nth MTC device can probably access. The nth MTC device chooses one eNB from Mn to perform the random access. In LTE-A, an eNB can request an MTC device to perform the exploration of surrounding eNBs, and report the exploration result. Thus, Mn for all n are available for all eNBs. However, Mn is unknown by the other MTC devices, except for the nth MTC device. (3) Denote Nm as the set of MTC devices accessing the mth eNB. In the traditional access class barring, ||Nm|| for all m are known by each eNB because each MTC device can only access the eNB to which it attaches. However, if each MTC device can not only access the eNB to which it is attached but also the eNB it has not (yet) attached, ||Nm|| for all m are random variables unknown by eNBs, unless the eNB selection strategy for each MTC device is given. (4) Denote ln,m as an indicator function that, for the nth MTC device,

ln;m ¼



1; if the mth eNB is within Mn 0;

otherwise

ð1Þ

(5) Denote N 0m as the set of MTC devices that can only receive the signal from the mth eNB and these MTC devices can only access the mth eNB. Remind  that N 0m  is the norm of N 0m . In access class barring [10], it is possible to independently maximize the throughput of each cell by

individually setting pm = 1/||Nm|| (where pm is the barring rate of access class barring of the mth eNB) in each eNB, but the delay of each MTC device that attaches to the mth eNB may be extremely long when there is a large number of MTC devices that attach to the mth eNB [8]. 4. Proposed efficient cooperative access class barring with load balancing and traffic adaptive radio resource management (CACB-LB with TARRM) In the proposed CACB-LB, we use the percentage of the number of MTC devices that can only access one eNB between two adjacent eNBs as a criterion to allocate those MTC devices that are located in the overlapped coverage area to each eNB. In this way, we can have better load balancing among eNBs than CACB. Moreover, the proposed CACB-LB also uses the ratio of the channel quality indication that an MTC device received from an eNB over the number of MTC devices that attach to the eNB as a criterion to adjust the estimated number of MTC devices that may access the eNB. The proposed CACB-LB can have a better set of barring rates (p) of access class barring and can reduce random access delay experienced by an MTC device or a UE. Fig. 1 shows the architecture of an LTE-A cellular network. Some MTC devices may be located in the overlapped area of a macrocell and a picocell. Note that these MTC devices may also access an eNB that is not attached by them. In LTE-A, the eNBs communicate with each other by an X2 interface [8]. In the following, we described the proposed CACB-LB with TARRM in a homogeneous MTC device network and in a heterogeneous MTC device network. 4.1. Proposed CACB-LB with TARRM in a homogeneous MTC device network 4.1.1. Proposed CACB-LB in a homogeneous MTC device network In the proposed efficient load balancing cooperative access class barring with load balancing (CACB-LB), we

272

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281

want to acquire a better set of barring rates of access class barring p = [p1, p2, . . . , pM] that is jointly decided by M eNBs to minimize the access delay experienced by N active MTC devices. This objective can be achieved by balancing the number of accesses among M eNBs. Through cooperation among eNBs to make a joint decision of p = [p1, p2, . . . , pM], the problem can be formulated by

min

maxðkN1 k; kN2 k; . . . ; kNM kÞ

p1 ;p2 ;...;pM

ð2Þ

s:t: ðC1:1Þ 0 6 p ¼ ½p1 ; p2 . . . ; pM  6 1 kNM kpm 6 1; for m ¼ 1; . . . ; M

ðC1:2Þ

The eNB selection strategy for an MTC device is based on pm, for m = 1, 2, . . . , M. The eNB with higher pm has a higher probability to be accessed by MTC devices. Upon receiving {pi, pj} # p, the nth MTC device adopts the following strategy to select an eNB to access:

" bn ðpi ; pj Þ ¼ Q n;i

¼P

pi x2M n px

; Q n;j

¼P

pj

#

x2Mn px

ð3Þ

where Qn,x is the probability that the nth MTC device selects the xth eNB to access. Given that the above strategy in Eq. (3) on the selection of an eNB by each MTC device is known by all eNBs, the estimatednumber of MTC devices that may access the  mth eNB, N 00m , for all m can be obtained by M X  00  X N  ¼ ln;m Q n;m m x¼1 n2Ax

¼ 1; . . . ; M



 C n;m ; kAm k þ 1

Am P 0; for m

  parameters, ||Am|| and N 0m , required for the proposed work can be exchanged via the X2 interface among eNBs, the PHY and DLL do not involve in the proposed work. Note that the X2 interface is handled by the radio resource management (RRM) component in an eNB. In addition, the complexities of linking with macro and pico-cell structure are resulting from neighbor cell list   update and the parameters, ||Am|| and N 0m , exchanged via the X2 interface. Algorithm 1. The    load  balancing algorithm for obtaining N 1 ; kN 2 k; . . . ; N M .     1. Set N m  ¼ N 0m  for all m 2. Denote Lm,m0 as a set of MTC devices that can not only access the mth eNB in a macrocell but also the   m’th eNB in a picocell and Lm;m0  is the norm of Lm,m0 3. for each mth eNB in a macrocell for each m0 th eNB in a picocell that has an overlapped coverage area with the mth eNB in a macrocell   if Lm;m0 –0 then       kNm0 k N  ¼ N  þ Lm;m0   m m ðkNm kþkNm0 kÞ         kNm k N 0  ¼ N 0  þ Lm;m0   m m ðkNm kþkNm0 kÞ   Lm;m0  ¼ 0 end if end for end for       4. Output N 1 ; N 2 ; . . . ; N M 

ð4Þ

where Cn,m is the channel quality indication that the nth MTC device received from the mth eNB. Note that in contrast to [8], in Eq. (4), we add the number of MTC devices (Am) that attach to an eNB and the channel quality indication (Cn,m) that an MTC device received from an eNB to calculate the number of MTC devices that access an eNB. This is because the higher the number of MTC devices that attach to an eNB, the lower the chance for an MTC device to access that eNB, and the better the channel quality indication that an MTC device received from an eNB, the higher the chance for the MTC device to access that eNB. In this way, we can better reflect the number of MTC devices that access an eNB. Finally, a better set of barring rates p of access class barring can be found by the proposed Algorithm 1 and the proposed Algorithm 2. In Algorithm 1, we use the percentage of the number of MTC devices that can only access one eNB between two adjacent eNBs as a criterion to allocate those MTC devices that are located in the overlapped coverage area to each eNB. In this way, we can evenly distribute the load among M eNBs. In Algorithm 2, we use Eq. (4) to   obtain N 00m  so as to get a better set of p. Note that UEs use the same strategy as MTC devices to obtain a set of barring rates p of access class barring. Note that the overhead of the proposed scheme is that the X2 interface needs to be used for exchanging some parameters,   such as ||Am|| and N 0m , among eNBs. Since the

Algorithm 2. The procedure for obtaining the access class barring parameter p.       1. Input N 1 ; N 2 ; . . . ; N M  from Algorithm 1 2. Set pm ¼ kN1 k for all m m     P P Cn;m 3. Set N 00m  ¼ M x¼1 n2Ax ln;m Q n;m kAm kþ1 for m ¼ 1; . . . ; M 4. while  00         N   N  >  or . . . or N 00   N   > e do 1 M M    1  00  Set N  ¼ N  for all m m

m

Set pm ¼ N1 for all m k mk     P P Cn;m Set N 00m  ¼ M x¼1 n2Ax ln;m Q n;m kAm kþ1 for m ¼ 1; . . . ; M end while 5. Output p1, p2, . . . , pm

We assume that MTC devices and UEs use different PRBs of a PRACH and separate preambles to perform random access with eNBs so as to reduce the congestion between them. Fig. 2 shows the access class barring procedure for each MTC device or UE. At first, an eNB broadcasts a barring rate p 2 [0, 1]. An MTC device or UE also generates a random value q, which is a uniform random

273

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281

eNB

MTC device/ UE System Information (p) via broadcast

Access A Ac ccess Class Class ss Barring Bar Ba arr rring

Retry after a certain period

Draw a random number q-unif(0, 1)

q

Scheduled Transmission

Random Access

Random Access Response

Contention Resolution

Fig. 2. Access class barring procedure for an MTC device or UE [10].

number between 0 and 1 (q-unif(0, 1)). If q 6 p, the MTC device or UE is allowed to perform the random access procedure with the eNB. Otherwise, the MTC device or UE is blocked for a certain period [10]. Without loss of generality, we assume MTC devices (UEs) are blocked for 1 ms (0.5 ms). After an MTC device passes the access class barring, it needs to perform random access with an eNB before uploading data to the eNB or downloading data from the eNB. If there is any collision happened during the random access procedure, a colliding device needs to retransmit its random access preamble after a random backoff time,

l defined as unif- 0; 2 21 ms, where l is a backoff indicator, r which is set to 1 for UEs and unif-(3, 6) for MTC devices [11], and r is the retransmission count of the random access preamble. 4.1.2. Proposed TARRM in a homogeneous MTC device network After an MTC device or a UE successfully performs random access with an eNB, the eNB needs to allocate radio resource to the MTC device or the UE. Fig. 3 shows the proposed TARRM in a homogeneous MTC device network. In Fig. 3, we classify an MTC device into a cluster k according to its random access rate ak and the amount of data uploaded and downloaded bk. We use 1/(akbk) as the time interval to allocate PRBs to cluster k, where akbk can reflect cluster k’s real data arrival rate. Here we use only two clusters for illustration. Note that in [4], the packet arrival rate and delay jitter of an MTC device were used to classify MTC devices into different clusters. In contrast, ours can fully reflect the traffic load of MTC devices to an eNB so as to

efficiently allocate radio resources to increase throughput performance. This point will be verified via performance evaluation. We assume that a1b1 of cluster 1 is larger than a2b2 of cluster 2, so the PRB allocation time interval 1/(a1b1) for cluster 1 is lower than the interval 1/(a2b2) for cluster 2. We allocate PRBs to MTC devices by first come, first served (FCFS) within a cluster, and we also allocate PRBs to UEs by FCFS. Furthermore, we use the concept of cognitive radio networks such that if there are unused PRBs of UEs, the eNB can schedule MTC devices from clusters based on the allocation frequency of each cluster in the descending order to use these PRBs of UEs. Note that in LTE-A, there are 15 PRBs in a physical downlink shared channel (PDSCH) and physical uplink shared channel (PUSCH), which are used by UEs or MTCs to download and upload data traffic, and 6 PRBs in a physical random access channel (PRACH) for random access. By using a cluster’s real data arrival rate to allocate PRBs and utilizing unused PRBs of UEs for MTC devices, the proposed TARRM can improve throughput performance. 4.2. Proposed CACB-LB with TARRM in a heterogeneous MTC device network 4.2.1. Proposed CACB-LB in a heterogeneous MTC device network For the proposed CACB-LB in a heterogeneous MTC device network, we redefine the following parameters. Note that we assume there are four classes of MTC devices, as shown in Table 2. The lower the class id (k) an MTC device has, the higher the priority of the MTC device is.

274

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281

Table 2 Parameter setting for each class of MTC devices [11].

Class (k) Node percentage Random access rate (requests/s) Upload/Download data size (byte) Application type Random access backoff indicator (l)

Low priority

High priority

Scheduled

Emergency

3 2% 0.1 180 Fleet management 2–4

2 2% 0.2 180 Hospital e-care 1–4

4 95% 0.1 180 Smart meter 3–6

1 1% 1 180 Seismic alarm 1

(1) ||Ak,m|| is the number of class k MTC devices that attach to the mth eNB. (2) Mn,k is the set of eNBs that the nth class k MTC device can possibly access. (3) ||Nk,m|| is the number of class k MTC devices that access the mth eNB. (4) Denote ln,k,m as an indicator function that, for the nth class k MTC device,

ln;k;m ¼



1; if the mth eNB is within M n;k

ð5Þ

0; otherwise

    (5) N 0k;m  is the number of class k MTC devices that can only access the mth eNB. (6) pk,m is the random access barring rate for class k MTC devices and is used in the mth eNB. We want to acquire a set of barring rates of access class barring pk = [pk,1, pk,2, . . . , pk,M] for all k that are jointly decided by M eNBs to minimize the access delay experienced by N active MTC devices. This objective can be achieved by balancing the number of accesses for each class of MTC device among M eNBs. Through cooperation among eNBs to make a joint decision of pk = [pk,1, pk,2, . . . , pk,M], the problem can be formulated as

min

pk;1 ;pk;2 ;...;pk;M

maxðkNk;1 k; kNk;2 k; . . . ; kNk;M kÞ; for all k

kNk;M kpk;m 6 1; for m ¼ 1; . . . ; M ð6Þ

The eNB selection strategy for class k MTC device is based on pk,m, for m = 1, 2, . . . , M. The eNB with higher pk,m has a higher probability to be accessed by class k MTC devices. Upon receiving {pk,i, pk,j} # pk, the nth class k MTC device adopts the following strategy to select an eNB to access:

"

bn;k ðpk;i ; pk;j Þ ¼ Q n;k;i ¼ P

pk;i x2M n;k pk;x

; Q n;k;j ¼ P

pk;j

k;x

Ak;m P 0; for m ¼ 1; . . . ; M

ð8Þ

where Qn,k,m is the channel quality indication that the nth class k MTC device received from the mth eNB. Finally, a better set of barring rates pk of access class barring can be found by executing the proposed Algorithm 3 and the proposed Algorithm 4. If there is any device that does not pass the access class barring, we assume a class k MTC device (UE) is blocked for (0.5  k) ms (1 ms). The random access backoff time is l

defined as unif- 0; 2 21 ms, where l is the backoff indicar tor, which is set to 3 for UEs. The random access backoff indicators for different classes of MTC devices are shown in Table 2, and r is the retransmission count of the random access preamble. Note that the proposed Algorithm 3 and Algorithm 4 obtain a set of barring rates pk of access class barring that will be used by class k MTC devices, while Algorithm 1 and Algorithm 2 obtain a set of barring rates p of access class barring that will be used by all MTC devices. Algorithm 3. The load balancing algorithm for obtain            ing N k;1 ; N k;2 ; . . . ; N k;M .         1. Set N k;m  ¼ N 0k;m  for all m

s:t: ðC1:1Þ 0 6 pk ¼ ½pk;1 ; pk;2 . . . ; pk;M  6 1 ðC1:2Þ

    M C n;k;m  00  X X ; ln;k;m Q n;k;m Nk;m  ¼ kAk;m k þ 1 x¼1 n2A

#

x2Mn;k pk;x

ð7Þ where Qn,k,x is the probability that the nth class k MTC device selects the xth eNB to access. Given that the above strategy in Eq. (7) on the selection of an eNB by each class k MTC device is known by all eNBs, the estimated number of class k MTC devices that may     access the mth eNB, N 00k;m , for all m can be obtained by

2. Denote Lk,m,m0 as a set of class k MTC devices that can not only access the mth eNB but also the m’th eNB   and Lk;m;m0  is the norm of Lk,m,m’ 3. for each mth eNB in a macrocell for each m’th eNB in a picocell that has an overlapped coverage area with the mth eNB in a macrocell   if Lk;m;m0 –0 then       N  0          k;m      N ¼ N þ L 0   k;m   k;m  k;m;m kNk;m kþNk;m0        kN k         N k;m0  ¼ N k;m0  þ Lk;m;m0     k;m kNk;m kþNk;m0    Lk;m;m0  ¼ 0 end if end for end for         4. Output N k;1 ; N k;2 ; . . . ; kN k;M k

275

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281

Algorithm 4. The procedure for obtaining the access class barring parameter pk.

Cluster 1 with (α 1, β 1 )

            1. Input N k;1 ; N k;2 ; . . . ; N k;M  from Algorithm 3

Time

1

Frequency

2. Set pk;m ¼ N for all m k k   k;m    00  PM P C n;k;m 3. Set N k;m  ¼ x¼1 n2Ak;x ln;k;m Q n;k;m kAk;m kþ1 for m ¼ 1; . . . ; M                 4. while N 00k;1   N k;1  > or . . . orN 00k;M   N k;M  >  do         Set N k;m  ¼ N 00k;m  for all m

Cluster 2 with (α 2, β 2 )

PDSCH / PUSCH

...

...

Set pk;m ¼ N1 for all m k k   k;m    00  PM P Cn;k;m Set N k;m  ¼ x¼1 n2Ak;x ln;k;m Q n;k;m kAk;m kþ1 for m ¼ 1; . . . ; M end while 5. Output pk,1, pk,2, . . . , pk,m

4.2.2. Proposed TARRM in a heterogeneous MTC device network In the proposed TARRM in a heterogeneous MTC device network, we classify MTC devices into four clusters according to their priorities (k) from cluster 1 to cluster 4. The higher the priority a cluster has, the higher the PRB allocation frequency the cluster has. The PRB allocation time interval for cluster k is k ms. In Fig. 4, assuming two clusters for illustration, we allocate PRBs to cluster 1 every 1 ms and to cluster 2 every 2 ms. In this way, the proposed TARRM can fully reflect the traffic load of MTC devices to an eNB so as to efficiently allocate radio resources to increase throughput performance. We allocate PRBs to MTC devices by first come first served (FCFS) within a cluster. Moreover, we use the concept of cognitive radio networks such that if there are unused PRBs of UEs, the eNB can schedule MTC devices from clusters based on the priority of each cluster in descending order to use these PRBs of UEs. By utilizing a cluster’s priority to allocate PRBs and utilizing unused PRBs of UEs for MTC devices, the proposed TARRM can improve throughput performance. 5. Analytic model In this section, we analyze the performance of the proposed CACB-LB, which focuses on reducing the random access delay experienced by an MTC device or a UE. Two performance metrics, the random access failure probability and the expectation of the random access preamble retransmission count, for MTC devices or UEs are analyzed. In order to analyze the proposed CACB-LB, we model the random access procedure by the M/mixEk/1/1 queuing model [12] and assume the random access result of an MTC device or a UE is either in success state (SS) or failure state (FS). The SS state denotes that the random access preamble of an MTC device or a UE is granted by an eNB, and the FS state denotes that the random access preamble of an

Allocate 15 PRBs to cluster 2 every 1/(α 2β 2 )ms

PRACH

a PRB used by a UE a PRB in a PRACH

...

...

a PRB used by an MTC device in cluster 1 a PRB used by an MTC device in cluster 2

Fig. 3. The proposed TARRM in a homogeneous MTC device network.

MTC device or a UE is rejected by an eNB or is lost during transmission. Fig. 5 shows the state transition diagram of the Markov process for the random access procedure. We assume that the arrival of a failed random access is a Poisson process with arrival rate k, and the departure rate of the FS state is h. The inter-arrival time of two consecutive failed random accesses is exponentially distributed with the probability density function, as shown in Eq. (9) [13–15].

f ðtÞ ¼ kekt

tP0

ð9Þ

Let d denote a random variable that represents the duration of a failed random access. Assume that d has the mixed Erlang distribution with the probability density function

f d ðtÞ ¼

I X ð/ tÞri 1 ci i / e/i t ðr i  1Þ! i i¼1

where I X ci ¼ 1 i¼1

ð10Þ

276

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281

Type 1 cluster

Type 2 cluster

EðdÞ ¼

  I X ri ci / i i¼1

ð12Þ

Then, the departure rate, h, can be expressed as h = 1/E(d). Furthermore, let the state probabilities of SS and FS be p0 and p1, respectively. From the Birth–Death process of queuing theory, we get

Frequency

Time

p0 ¼ ð1 þ k=hÞ1

ð13Þ

and

p1 ¼ ðk=hÞð1 þ k=hÞ1 PDSCH / PUSCH

ð14Þ

The random access failure probability (RAFP) is expressed as

RAFP ¼ p0

Z

tp

t¼0

 tp  f ðtÞdt þ p1 ¼ p0 ekt t¼0 þ p1

¼ 1  p0 ektp

ð15Þ

where tp is the average random access delay experienced by an MTC device or a UE. Let r denote the random access preamble retransmission count for an MTC device or a UE successfully performing the random access procedure with an eNB. Therefore, the expectation of r is expressed as follows:

Allocate 15 PRBs to type 2 cluster every 2ms

EðrÞ ¼

1 X

n1

nð1  p0 ektp Þ

ðp0 ektp Þ  1

ð16Þ

n¼1

PRACH

Note P1

that in the first part of Eq. (16), n1  p0 ektp Þ ðp0 ektp Þ, it represents the expectation of 1 (the first random access preamble transmission) plus the random access preamble retransmission count. Thus, Eq. (16) represents the expectation of the random access preamble retransmission count for an MTC device or a UE successfully performing the random access procedure with an eNB. n¼1 nð1

a PRB used by a UE

a PRB used by a type 1 MTC device

a PRB in a PRACH

a PRB used by a type 2 MTC device

Fig. 4. The proposed TARRM in a heterogeneous MTC device network.

λ

1-λ

SS

6. Performance evaluation

FS

1-θ

θ Fig. 5. State transition diagram of the Markov process for the random access procedure.

Note that in Eq. (10), I, ci, ri and /i determine the shape and scale of the distribution [14]. We selected the mixed Erlang distribution because it has been proven that it can well approximate many other distributions [13–15]. Thus, the expectation of d is expressed as

EðdÞ ¼

Z

1

t¼0

f d ðtÞtdt ¼

Z 1 r I X ci /i i tri e/i t dt ðr i  1Þ! t¼0 i¼1

By Laplace transform, we get

ð11Þ

In this section, we first describe simulation and analytic setup, and evaluation metrics. Then, we compare the proposed CACB-LB with CACB [8] in homogeneous and heterogeneous MTC device networks in terms of the average access delay of UEs, average access delay of MTC devices, average throughput from UEs, and average throughput from MTC devices. We also compare the proposed CACB-LB with TARRM with CACB [8] in homogeneous and heterogeneous MTC device networks in terms of the average throughput from UEs and average throughput from MTC devices. Moreover, based on the analytic model derived in the last section, we also compare analytic results and simulation results of the proposed CACB-LB with CACB [8] in terms of random access failure probability and random access preamble retransmission count. A simulation model using NS-2 was developed to validate the analytic model. The simulation model is based on the analytic model described in Section 5. The same parameter settings shown in Table 3 were used for simulation as well as for analysis. Simulation results were obtained by the average of 10,000 runs, and the confidence interval for our simulations is 95%.

277

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281 Table 3 Parameter setting for performance evaluation [3,4,8,9,14,15]. Parameter

Values

Number of cells Inter-site distance of macrocells Uplink data rate Downlink data rate Number of PRBs in a PDSCH/PUSCH Number of PRBs in a PRACH Number of PRBs for UEs in a PRACH Number of PRBs for MTCs in a PRACH Number of preambles for UEs Number of preambles for MTCs Cells layout and MTC device deployment percentages

7 macrocells and 3 picocells 500 m 500 Mbps 1 Gbps 15 6 4 2 58 6 See Fig. 6 0.0001 0.2, 0.35, 0.5, 0.65, 1 request/s 36, 72, 108, 144, 180 bytes 5 bytes 0.01 request/s 500 bytes 600 s c1 = c2 = 0.5 /1 = 2, /2 = 3 (/ms)

 Random access rate of an MTC in a homogeneous MTC device network Amount of data uploaded and downloaded for an MTC device in a homogeneous MTC device network Packet size Random access rate of UEs Amount of data uploaded and downloaded for UEs Simulation time d is a mixed Eralng-2 random variable

Macrocell

Macrocell

6%

6%

20%

20%

6%

4% Picocell

6% Picocell

Macrocell

Macrocell 20% 6%

6% Picocell

Macrocell

Macrocell

Fig. 6. Cells layout and MTC device deployment percentages [8].

6.1. Simulation and analytic setup, and evaluation metrics Tables 2 and 3 show the simulation setup. First, we define the average throughput from MTC devices as the throughput averaged over all cells’ eNBs. Then we define the average access delay of MTC devices as the delay averaged over all active MTC devices. The definitions of throughput and access delay of UEs are similar to those of MTC devices. Moreover, for the analytic setup, without loss of generality, we assume d is a mixed Eralng-2 random variable and the coefficients for d are set to c1 = c2 = 0.5 [14,15]. Fig. 6 shows the cell layout and MTC device deployment percentages in each picocell and macrocell [8].

6.2. Comparison between proposed CACB-LB/CACB-LB with TARRM and CACB in a homogeneous MTC device network In this section, we show simulation results for a homogeneous MTC device network. Fig. 7 shows the average access delay of UEs. The access delay increases as the number of UEs increases. The reason is that when the number of UEs increases, collisions in a PRACH may increase. This results in the increase of access delay. The average access delay of the proposed CACB-LB is 45% lower than that of CACB. Fig. 8 shows the average access delay of MTC devices. The access delay increases as the number of MTC devices increases. The reason is that when the number of MTC devices increases, collisions in a PRACH may increase.

278

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281 3.5

10

І 95% confidence interval

Average delay of proposed CACB-LB

2 1.5

Average delay of CACB

1 0.5

Throughput (bps) X 10000

2.5

Delay (s)

І 95% confidence interval

9

3

8 7 6

Average throughput of proposed CACB-LB with TARRM Average throughput of proposed CACB-LB

5 4 3

Average throughput of CACB

2 1

0 15

20

25

30

35

0

10000

Number of UEs in each macrocell

20000

30000

40000

50000

Total number of acve MTC devices (N) Fig. 7. Access delay of UEs in a homogeneous MTC device network. Fig. 10. Throughput from MTC devices.

4.5 4

І 95% confidence interval

3.5

Delay (s)

3 Average delay of proposed CACB-LB

2.5

to uplink/downlink access may decrease. It results in degradation of the throughput. Fig. 10 also demonstrates that the proposed CACB-LB’s average throughput from MTC devices is 28% higher than CACB’s. The average throughput from MTC devices of the proposed CACB-LB with TARRM is 30% higher than that of CACB.

2 Average delay of CACB

1.5 1

6.3. Comparison between proposed CACB-LB/CACB-LB with TARRM and CACB in a heterogeneous MTC device network

0.5 0 10000

20000

30000

40000

50000

Total number of acve MTC devices (N) Fig. 8. Access delay of MTC devices.

This also results in the increase of access delay. The average access delay of the proposed CACB-LB is 30% lower than that of CACB. Fig. 9 shows the average throughput from UEs. We observed that the throughput degrades as the number of UEs increases. This is because collisions may occur as the number of UEs increases and the number of UEs granted to uplink/downlink access may decrease. It results in degradation of the throughput. Fig. 9 also demonstrates that the proposed CACB-LB’s average throughput from UEs is 26% higher than that of CACB, and the average throughput from UEs of the proposed CACB-LB with TARRM is 31% higher than that of CACB. Fig. 10 shows the average throughput from MTC devices. We found that the throughput degrades as the number of MTC devices increases. This is because collisions may increase as the number of MTC devices increases and the number of MTC devices granted

6.4. Analytic results between CACB-LB and CACB By Eq. (15), Fig. 17 shows that the average random access failure probability of UEs for the proposed CACB-LB

І 95% confidence interval

14

3.5

І 95% confidence interval

12

3

6

Average throughput of proposed CACB-LB with TARRM Average throughput of proposed CACB-LB

4

Average throughput of

10 8

2.5

Delay (s)

Throughput (bps) X 10000

16

In this section, we show simulation results for a heterogeneous MTC device network. Fig. 11 shows the proposed CACB-LB’s average access delay of UEs is 42% lower than that of CACB. The proposed CACB-LB’s average access delay experienced by class 1, class 2, class 3, and class 4 MTC devices are 34%, 28%, 30%, and 38% lower than CACB’s, respectively. Fig. 12 shows the proposed CACB-LB’s average throughput from UEs is 28% higher than that of CACB, and the average throughput from UEs of the proposed CACB-LB with TARRM is 33% higher than that of CACB. Figs. 13–16 show the proposed CACB-LB’s average throughput from class 1, class 2, class 3, and class 4 MTC devices are 39%, 42%, 46%, and 54% higher than CACB’s, respectively. Moreover, Figs. 13–16 also show the proposed CACB-LB with TARRM’s average throughput from class 1, class 2, class 3, and class 4 MTC devices are 44%, 48%, 53%, and 62% higher than CACB’s, respectively.

Average delay of proposed CACB-LB

2 1.5

Average delay of CACB

1

CACB

2

0.5

0 15

20

25

30

35

Number of UEs in each macrocell Fig. 9. Throughput from UEs in a homogeneous MTC device network.

0 15

20

25

30

35

Number of UEs in each macrocell Fig. 11. Access delay of UEs in a heterogeneous MTC device network.

279

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281 8

І 95% confidence interval

14 12

Throughput (bps) X 10000

Throughput (bps) X 10000

16

Average throughput of proposed CACB-LB with TARRM Average throughput of proposed CACB-LB

10 8 6

Average throughput of CACB

4 2 0

6

20

25

30

Average throughput of proposed CACB-LB with TARRM Average throughput of proposed CACB-LB

5 4 3 2

Average throughput of CACB

1 0

15

І 95% confidence interval

7

35

9500

19000

28500

38000

47500

Total number of class 4 MTC devices

Number of UEs in each macrocell

Fig. 16. Throughput from class 4 MTC devices.

Fig. 12. Throughput from UEs in a heterogeneous MTC device network.

1

І 95% confidence interval

І 95% confidence interval

8 7 6

Average throughput of proposed CACB-LB with TARRM Average throughput of proposed CACB-LB

5 4 3 2

Average throughput of CACB

1 0 100

200

300

400

500

Total number of class 1 MTC devices

Random access failure probabulity

Throughput (bps) X 10000

9

0.9 0.8

Average_proposed CACB-LB

0.7 0.6 0.5

Average_proposed CACB-LB-sim

0.4

Average_CACB

0.3

Average_CACB-sim

0.2 0.1

Fig. 13. Throughput from class 1 MTC devices.

0 0.01

0.1

1

Random access failure arrival rate of UEs, λ (/ms) Fig. 17. Average random access failure probability versus random access failure arrival rate of UEs.

І 95% confidence interval

8 7

Average throughput of proposed CACB-LB with TARRM Average throughput of proposed CACB-LB

6 5 4 3

Average throughput of CACB

2 1 0

200

400

600

800

1000

Total number of class 2 MTC devices Fig. 14. Throughput from class 2 MTC devices.

1

Random access failure probability

Throughput (bps) X 10000

9

0.9

І 95% confidence interval

0.8

Average_proposed CACB-LB

0.7 0.6

Average_proposed CACB-LB-sim

0.5

Average_CACB

0.4 0.3

Average_CACB-sim 0.2 0.1 0 0.01

Throughput (bps) X 10000

9 8 7 Average throughput of proposed CACB-LB with TARRM Average throughput of proposed CACB-LB

6 5 4 3 2

Average throughput of CACB

1 0

200

400

600

800

0.1

1

Random access failure arrival rate of MTC devices, λ (/ms)

І 95% confidence interval

1000

Total number of class 3 MTC devices Fig. 15. Throughput from class 3 MTC devices.

Fig. 18. Average random access failure probability versus random access failure arrival rate of MTC devices.

is 46% with k = 0.01, 43% with k = 0.1, and 14% with k = 1 lower than that of CACB, respectively. Fig. 18 shows that the average random access failure probability of MTC devices for the proposed CACB-LB is 48% with k = 0.01, 59% with k = 0.1, and 17% with k = 1 lower than that of CACB, respectively. In Figs. 17 and 18, analytic and simulation results are very close, which confirms the validity of our analytic model.

280

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281

Random access preamble retransmission count, E(σ)

12

І 95% confidence interval 10

Average_proposed CACB-LB

8

Average_proposed CACB-LB-sim

6

Average_CACB 4

Average_CACB-sim

2 0 0.01

0.1

1

Random access failure arrival rate of UEs, λ (/ms) Fig. 19. Average random access preamble retransmission count versus random access failure arrival rate of UEs.

Random access preamble retransmission count, E(σ)

18 16

І 95% confidence interval Average_proposed CACB-LB

14 12

Average_proposed CACB-LB-sim

10 8

Average_CACB

6

Average_CACB-sim

4 2 0 0.01

0.1

barring rates of access class barring than CACB so as to reduce the random access delay experienced by each MTC device or a UE. After an MTC device successfully accesses an eNB, the eNB schedules radio resources for the MTC device based on the random access rate and the amount of data uploaded and downloaded by the MTC device in a homogeneous MTC device network, and the priority of the MTC device in a heterogeneous MTC device network. In addition, we have applied the concept of cognitive radio networks such that if there are unused PRBs of UEs, an eNB can schedule MTC devices to use these PRBs of UEs so as to improve throughput performance. Simulation results show that, in a homogeneous MTC device network, the proposed CACB-LB’s average access delay of UEs is 45% lower than CACB’s. Its average access delay of MTC devices is 30% lower than CACB’s. Its average throughput from UEs is 26% higher than CACB’s. Its average throughput from MTC devices is 28% higher than CACB’s. The proposed CACB-LB with TARRM’s average throughput from UEs is 31% higher than CACB’s. Its average throughput from MTC devices is 34% higher than CACB’s. As to a heterogeneous MTC device network, similar simulation results have been obtained. Therefore, the proposed CACB-LB with TARRM is feasible for M2M communications over LTE-A under either in a homogeneous MTC device network or in a heterogeneous MTC device network.

1

Random access failure arrival rate of MTC devices, λ (/ms)

Acknowledgments Fig. 20. Average random access preamble retransmission count versus random access failure arrival rate of MTC devices.

By Eq. (16), Fig. 19 shows that the average random access preamble retransmission count of UEs of the proposed CACB-LB is 15% with k = 0.01, 42% with k = 0.1, and 49% with k = 1 lower than that of CACB-LB, respectively. Fig. 20 shows that the average random access preamble retransmission count of MTC devices of the proposed CACB-LB is 16% with k = 0.01, 49% with k = 0.1, and 32% with k = 1 lower than that of CACB-LB, respectively. In Figs. 19 and 20, analytic and simulation results are very close, which confirms the validity of our analytic model. 7. Conclusion In this paper, we have presented two efficient load balancing cooperative access class barring with load balancing (CACB-LB) and traffic adaptive radio resource management (TARRM) schemes for M2M communications over LTE-A. We use the percentage of the number of MTC devices that can only access one eNB between two adjacent eNBs as a criterion to allocate those MTC devices that are located in the overlapped coverage area to each eNB. In this way, the proposed CACB-LB can have better load balancing among eNBs than CACB, which is the best available related work. Moreover, the proposed CACB-LB also uses the ratio of the channel quality indication that an MTC device received from an eNB over the number of MTC devices that attach to the eNB as a criterion to adjust the estimated number of MTC devices that may access the eNB. As a result, the proposed CACB-LB can have a better set of

This work was supported in part by the Ministry of Science of Technology, Taiwan, under Grants NSC99-2221-E009-081-MY3 and NSC102-2221-E-009-090-MY3. The authors would like to thank the anonymous reviewers for their valuable comments that improved the quality of this paper and Dr. Jen-Shun Yang and Tiffany Y. Chin for their suggested improvements and editing of this paper.

References [1] M. Hasan, E. Hossain, D. Niyato, Random access for machine-tomachine communication in LTE-advanced networks: issues and approaches, IEEE Commun. Mag. 51 (6) (2013). [2] Y. Chen, W. Wang, Machine-to-machine communication in LTE-A, in: Proc. IEEE Vehicular Technology Conf., Sept. 2010, pp. 1–4. [3] K. Zheng, F. Hu, W. Wang, W. Xiang, M. Dohler, Radio resource allocation in LTE-advanced cellular networks with M2M communications, IEEE Commun. Mag. 50 (7) (2012) 184–192. [4] S.-Y. Lien, K.-C. Chen, Y. Lin, Toward ubiquitous massive accesses in 3GPP machine-to-machine communications, IEEE Commun. Mag. 49 (4) (2011) 66–74. [5] N. Gligoric, S. Krco, D. Drajic, S. Jokic, B. Jakovljevic, M2M device management in LTE networks, in: Proc. IEEE Telecommunications Forum, Nov. 2011, pp. 414–417. [6] H. Suda, Deployment strategies of LTE of DOCOMO and challenges for future mobile, in: Proc. IEEE Communication Technology and Application Conf., Oct. 2011, p. 1. [7] C. Ide, B. Dusza, M. Putzke, C. Muller, C. Wietfeld, Influence of M2M communication on the physical resource utilization of LTE, in: Proc. IEEE Wireless Telecommunications Symposium, April 2012, pp. 1–6. [8] S.-Y. Lien, T.-H. Liau, C.-Y. Kao, K.-C. Chen, Cooperative access class barring for machine-to-machine communications, IEEE Trans. Wireless Commun. 11 (1) (2012) 27–32. [9] K.-D. Lee, S. Kim, B. Yi, Throughput comparison of random access methods for M2M service over LTE networks, in: Proc. IEEE Global Telecommunications Conf., Dec. 2011, pp. 373–377.

Y.-H. Hsu et al. / Computer Networks 73 (2014) 268–281 [10] 3GPP TS 22.011 V9.4.0, 3rd Generation Partnership Project Technical Specification Group Services and System Aspects Service Accessibility (Release 9), June 2010. [11] J.-P. Cheng, C.-H. Lee, T-M. Lin, Prioritized random access with dynamic access barring for RAN overload in 3GPP LTE-A networks, in: Proc. IEEE Global Telecommunications Conf., Dec. 2011, pp. 368– 372. [12] M. Zobu, V. Saglam, Control of traffic intensity in hyperexponential and mixed Erlang queueing systems with a method based on SPRT, Math. Probl. Eng. 2013 (2013) 9. [13] S. Asmussen, Applied Probability and Queues, Wiley, New York, 1987. [14] A.-C. Pang, Y.-B. Lin, H.-M. Tsai, P. Agrawal, Serving radio network controller relocation for UMTS all-IP networks, IEEE J. Select. Areas Commun. 22 (4) (2004) 617–629. [15] Y. Fang, I. Chlamtac, Teletraffic analysis and mobility modeling of PCS networks, IEEE Trans. Commun. 47 (7) (1999) 1062–1072. [16] A. Laya, L. Alonso, J. Alonso-Zarate, Is the random access channel of LTE and LTE-A suitable for M2M communications? A survey of alternatives, IEEE Commun. Surveys Tutorials 16 (1) (2014) 4–16. [17] Y. Pang, S. Chao, G. Lin, H. Wei, Network access for M2M/H2H hybrid systems: a game theoretic approach, IEEE Commun. Lett. 99 (2014) 1–4. [18] T.-M. Lin, C.-H. Lee, J.-P. Cheng, W.-T. Chen, PRADA: prioritized random access with dynamic access barring for MTC in 3GPP LTE-A networks, IEEE Trans. Vehicular Technol. 99 (2014) 1–7.

Yi-Huai Hsu received the B.S. degree in Computer and Information Science from the National Taichung University, Taichung, Taiwan, in 2008 and M.S. degree in Computer Science from the National Chiao Tung University, Taiwan, in 2010. He is currently a Ph.D. student in the Department of Computer Science, National Chiao Tung University. His research interests include wireless (ad hoc/ sensor/VANET) networks, cloud computing, and LTE-A.

Kuochen Wang received the B.S. degree in control engineering from the National Chiao Tung University, Taiwan, in 1978, and the M.S. and Ph.D. degrees in electrical engineering from the University of Arizona in 1986 and 1991, respectively. He is currently a Professor and the Chair of the Department of Computer Science, National Chiao Tung University. He was a Director in the Institute of Computer Science and Engineering/Institute of Network Engineering, National Chiao Tung University from August 2009 to July 2011. He was an Acting/Deputy Director of the Computer and Network Center at this university from June 2007 to July 2009. He was a Visiting Scholar in the

281

Department of Electrical Engineering, University of Washington from July 2001 to February 2002. From 1980 to 1984, he was a Senior Engineer at the Directorate General of Telecommunications in Taiwan. He served in the army as a second lieutenant communication platoon leader from 1978 to 1980. His research interests include wireless (ad hoc/sensor/ VANET) networks, mobile/cloud computing, and power management for multimedia portable devices.

Yu-Chee Tseng received his B.S. and M.S. degrees in Computer Science from the National Taiwan University and the National Tsing Hua University in 1985 and 1987, respectively. He obtained his Ph.D. in Computer and Information Science from the Ohio State University in January of 1994. He was an Associate Professor at the Chung Hua University (1994–1996) and at the National Central University (1996–1999), and a Professor at the National Central University (1999–2000). He is Professor (2000–present), Chairman (2005–2009), Associate Dean (2007–2011), and Dean (2011– present) at the Department of Computer Science, National Chiao-Tung University, Taiwan. He is also Adjunct Chair Professor at the Chung Yuan Christian University (2006–present). Dr. Tseng received the Outstanding Research Award, by National Science Council, ROC, in both 2001–2002 and 2003–2005, the Best Paper Award, by Int’l Conf. on Parallel Processing, in 2003, the Elite I.T. Award in 2004, and the Distinguished Alumnus Award, by the Ohio State University, in 2005. His research interests include mobile computing, wireless communication, network security, and parallel and distributed computing. Dr. Tseng is a member of ACM and a Senior Member of IEEE. Dr. Tseng served as a Vice Program Chair in the Int’l Conf. on Distributed Computing Systems (ICDCS), 2004, as a Vice Program Chair in the IEEE Int’l Conf. on Mobile Ad-hoc and Sensor Systems (MASS), 2004, as an Associate Editor for The Computer Journal (2001–present), as a Guest Editor for ACM Wireless Networks special issue on ‘‘Advances in Mobile and Wireless Systems’’, as a Guest Editor for IEEE Transactions on Computers special on ‘‘Wireless Internet’’ (2003), as an Editor for Journal of Information Science and Engineering (2002–2008), as an Associate Editor for Telecommunication Systems (2005–present), as an Associate Editor for IEEE Trans. on Vehicular Technology (2005–2009), as an Associate Editor for IEEE Trans. on Mobile Computing (2006–present), and as an Associate Editor for IEEE Trans. on Parallel and Distributed Systems (2008–present).