Content uploaded by Chengchao Liang
Author content
All content in this area was uploaded by Chengchao Liang on Feb 27, 2016
Content may be subject to copyright.
Mobile Virtual Network Admission Control and
Resource Allocation for Wireless Network
Virtualization: A Robust Optimization Approach
Chengchao Liang and F. Richard Yu
Depart. of Systems and Computer Eng., Carleton University, Ottawa, ON, Canada
Email: chengchaoliang@sce.carleton.ca; richard.yu@carleton.ca
Abstract—Wireless network virtualization is a promising tech-
nology in next generation wireless networks. In this paper,
motivated by the experience of user equipment (UE) admission
control in traditional wireless networks, we propose a novel
concept of mobile virtual network (MVN) admission control
for wireless virtualization. By limiting the number of MVNs
embedded in the physical network, MVN admission control can
effectively guarantee quality of service (QoS) experienced by
users of MVNs and maximize the utilization of the physical
networks at the same time. Specifically, we propose a two-stage
MVN embedding mechanism that can decouple short-term phys-
ical resource allocation from long-term admission control and
resource leasing. With recent advances in robust optimization,
we formulate the MVN admission control problem as a robust
optimization problem. Both the long-term admission control
and short-term resource allocation problems are transformed
to convex problems, which can be solved efficiently. Simulation
results are presented to show the effectiveness of the proposed
scheme.
Index Terms—Wireless network virtualization, admission con-
trol, resource allocation, robust optimization
I. INTRODUCTION
Recently, virtualization has been considered as a promising
technology for next generation wireless networks [1]. With
virtualization technology, wireless network infrastructure can
be decoupled from the services that it provides, so that differ-
entiated services can share the same infrastructure, maximizing
their utilization.
Several research projects have been started around the
world in the area of wireless network virtualization, such as
Environment for Network Innovations (GENI) and Virtualized
dIstributed plaTfoRms of smart Objects (VITRO) [1], [2].
Virtualizing eNodeB in 3rd Generation Partnership Project
(3GPP) Long term evolution (LTE) is investigated in [3]
from the views of node virtualization and software defined
networks. In addition, several studies have tried to solve
the challenges in wireless network virtualization, such as
virtual resource allocation [4], full-duplex relay scenario [5],
scheduling [6] and spectrum sharing [7].
While these excellent works have improved the performance
of wireless network virtualization, the quality of service (QoS)
experienced by users in virtual networks can be significantly
affected by virtualization, because virtual networks share the
same underlaying physical network, and both the traffic and
wireless channels can change dynamically [1]. To guarantee
the QoS, user equipment (UE) admission control has been
studied extensively in traditional wireless networks (see [8],
[9] and the references therein). By limiting the number of UEs
entering the network, UE admission control can effectively
guarantee QoS and maximize network utilization simultane-
ously.
In this paper, motivated by the experience of UE admission
control in traditional wireless networks, we propose a novel
concept of mobile virtual network (MVN) admission control
for wireless virtualization. By limiting the number of MVNs
embedded in the physical network, MVN admission control
can effectively guarantee QoS experienced by users of MVNs
and maximize the utilization of the physical networks at the
same time. To the best of our knowledge, MVN admission
control has not been well studied in the literature. The
contributions of this paper are as follows. In addition to
the physical resource leasing and allocation issues studied
in previous works, we jointly study MVN admission control
with these issues for wireless virtualization. Specifically, we
propose a two-stage MVN embedding mechanism that can
decouple short-term physical resource allocation from long-
term admission control and resource leasing. With recent
advances in robust optimization (RO) [10], we formulate the
MVN admission control problem as a robust optimization
problem. RO is a novel technology in optimization theory
that is suitable for solving problem with uncertain data [11].
RO has been successfully used in signal processing [12]
and resource allocation [13], among others. In the MVN
admission control problem, uncertainty arises when estimating
the capacity of underlaying physical networks and the traffic
of MVNs. Both the long-term admission control and short-
term resource allocation problems are transformed to convex
problems, which can be solved efficiently.
The rest of this paper is organized as follows. Section
II introduces system model and describe the problem of
MVNs embedding. The proposed two-stage MVNs admission
mechanism is presented in Section III. Section IV discusses
simulation results. Finally, we conclude this study in Section
V.
II. SYSTEM MODEL AND PROBLEM STATEMENT
In this section, we first introduce the business model of
wireless virtualization. Then, we present the system model,
978-1-4799-5952-5/15/$31.00 ©2015 IEEE
followed by the problem statement considered in this paper.
A. Business Model
In this paper, we consider a virtualized mobile network
model introduced in [1] that includes four parts, Infrastruc-
ture Providers (InPs), Mobile Virtual Network Providers
(MVNPs), Mobile Virtual Network Operators (MVNOs) and
Service Providers (SPs). InPs operate several physical mobile
networks. Each InP has their own radio access networks
(RAN) including amount of macro basestations (BSs) and (or)
small BSs ) to cover a certain area. These physical mobile
networks can be shared with each other, which means a InP
can utilize other InPs’ physical resource by pay leasing fee.
MVNPs create and provide mobile virtual networks (MVNs)
to higher MVNOs. MVNs are embedded at physical mo-
bile networks by leasing physical resources. Without loss of
generality, to accommodate network sharing that has been
standardized, we assume MVNPs also can borrow physical
resources from any InPs through paying fees, because network
sharing has been considered in next generation mobile net-
works. MVNOs request MVNs from MVNPs and operate them
to satisfy the requirements of services. SPs directly provide
services (e.g., real time video and online game) to end users. In
this paper, we do not consider SPs due to the space limitation.
B. System Model
1) Physical wireless network model: Consider IInPs,
denoted by i∈I={1,2, ..., I }, in our model. All of the
InPs own their operating RANs that can fully cover the same
geographic area (a city or a district). This large area can be
separated to Nseveral sub-areas. For ease of presentation, we
assume that each sub-area is covered by one macro BS in
one InP. The covered sub-areas are based on the coverage of
macro BSs denoted by b∈B
i=1,2, ..., N . Specifically, total
Imacro BSs will cover area b, and the macro BS operated by
InP iat area bis denoted by ib. Assume full network sharing is
realized, the MVNP can borrow any physical network resource
from all InPs. We use αb
i∈[0,1] to denote the borrowing ratio
of the ith macro BS at area b. Without losing generality, we
assume the resources in BSs are not always full for borrowing,
which means the available ratio of resource is αb
i,max for all
BSs {ib}.ψc
i(crepresents cost) to denote the access cost
of fully leasing BS ib. Thus, the borrowing cost at area bis
finp,b =I
iαb
iψc
i.
2) Mobile virtual networks model: We assume KMVNOs
denoted by k∈K={1,2, ..., K}in our system model. For
simple notation, we use the same notation for MVNs operated
by MVNOs. For total KMVNs (one MVNO owns one MVN),
the MVNP creates KMVNs. These MVNs request to access
the network dynamically, which means embeding request. If
the physical network can embed these request MVNs at area b,
MVNP will admit the access of these MVNs. Let xb
k∈{0,1}
be the admission indicator of the kth MVN. Formally, if the
MVNO admits the kth MVN, xb
k=1; if block, xb
k=0.
μr
kis the weight parameter (rrepresents for admission
revenue) for admitting the kth MVN and μp
kis the weight
parameter (prepresent blocking penalty) for blocking the kth
MVN. Therefore, the utility that MVNP can get from MVNOs
at area bis fmvno,b =K
kxb
kψr
k−K
k(1 −xbk)ψp
k.
Let the arrival rate of the users in MVN kassociated with
area (macro BS) bbe Poisson and thus denoted by λb
k(unit in
users/second). The session holding time of MVN kis denoted
by 1/μb
k(unit in second). 1/μb
kis decided by data rate rkand
file length Fk, which is 1/μb
k=Fk/Rk. Since data rate Rk
and file length Fkare only related to service type, we assume
for all area rkand Fkare the same. However, unlike holding
time, the arrival rates are based on service type and varying
with time. Thus, it is very hard to evaluate the arrival rate
accurately. Following the research in [14], we define ρb
k=
Fkλb
kto be the mean traffic of MVN karriving at area b.
Therefore, the total expected traffic load (request throughput)
at area bis ρb=K
kρb
k.
The leased physical resources from InP iused for embed-
ding MVNs at area bis denoted by αb
i.Weusecb
kto denote
the capacity allocated to MVN kat area b. Obviously, cb
kis
decided by users distribution
db
k(the position of active users
served by MVN kat area b), leasing strategies {αb
i}and
physical resource allocation strategies denoted by {yb
ik}that
means the ratio of resource allocated to the active users served
by MVN kby using BS ib. Therefore, similar with [14], the
system is stable only if
K
kxb
kρb
k
cb
k{
db
k},{αb
i},{yb
k}≤1∀b, k, i (1)
This stable condition is the key constraint in our optimization
problem.
C. Problem statement
Let ub
k∈U
b
k=1.2, ..., |Ub
k|denote the active user at area
bserved by MVN k. As the short-term resource allocation
strategy is written as {yb
k}, the throughput gain by ub
kis thus
Rb
uk(yb
k). We use vector to represent the resource allocation
strategy here is because it includes two types of strategies
including BS associate and radio resource allocation. Detailed
description is in Subsection III-C. Thus, the objective function
denoted by F({xb
k},{αb
i},{yb
k})is trying to maximize the
utility of MVNP. Thus, we obtain the following optimization
problem:
max
{xb
k},{αb
i},{yb
k}F=
N
bfmvno,b −finp,b(2a)
s.t. xb
k∈{0,1},∀b, k (2b)
αb
i≤αb,max
i,∀b, i (2c)
Rb
uk(yb
k)≥rk,∀b, k, uk(2d)
K
kxb
kρb
k
cb
k{
db
k},{αb
i},{yb
k}≤1,∀b(2e)
III. THE PROPOSED MOBILE VIRTUAL NETWORK (MVN)
EMBEDDING MECHANISM WITH ADMISSION CONTROL
In this section, we present the proposed two-stage MVN
embedding mechanism. Then, we study robust admission con-
trol and resource leasing issues. Physical resource allocation
is studied next, followed by some implementation discussions.
A. Two-stage Mechanism of Embedding MVNs
In the above problem (2) described in the previous sec-
tion, it is difficult to solve the three types of variables
({xb
k},{αb
i},{yb
k})in the problem at the same time due to
different time scales and traffic realizations. {xb
k}used to
control the embedding decision of MVNs are in large time
scale (minutes to hours) and decided before traffic realization.
This is because a typical MVNO usually requests a MVN for
a period of time instead of a short moment. For instance, a
sports game broadcast virtual network will stay in the network
for hours, and a real time video conferencing network may
exist for minutes. If the MVNO is operating a VoIP MVN,
it may ask a MVN for days or months. This means {xb
k}
should be optimized before users arrival with uncertain arrival
rate. Similar to admission control, {αb
i}that can be considered
as coordinative parameters among InPs should be optimized
in large time scale and before traffic realization, because the
MVNP should reserve enough resource for embedding MVNs
and also reduce the overhead of coordinating InPs. However,
{yb
k}have to be optimized after user actually arrival and
waiting for scheduling (traffic load uncertainty realization),
which means {yb
k}is changing in ms to second. Thus, to
address these unmatched optimization appropriately, we divide
optimization problem (2) into two stages: 1) joint admission
control and leasing optimization and 2) physical resource al-
location optimization. This two-stage mechanism is described
in Fig. refFig:TwoStage. In the first stage, without knowing
the exact traffic load request from MVNs, the MVNP conducts
admission control and resource leasing based on the estimated
arrival traffic {λb
k}that are uncertain and provided by MVNs.
Obviously, in real networks, {λb
k}vary with different long
term time slots (e.g., rush hours and evening). Therefore, in the
first stage, the system has to be able to tolerate this variance.
In the second stage, the MVNP will perform a traditional
resource allocation. This stage can be considered as an ex-
tended stage of first stage after traffic arriving. As we consider
multiple InPs in our model, this resource allocation includes
radio resource allocation and BS association. It should note
that, in this paper, we only consider a macro BS to provide
radio access and leave the heterogeneous network scenarios
for future research.
B. Robust Admission Control and Resource Leasing
Based on the above analysis, admission indicators {xb
k}
and infrastructure leasing ratio {αb
i}are optimized before the
realization of uncertainty {λb
k}. Since the resource allocation
ratio {yb
k}is not included in the objective function, we only
need to reform our constraints related to uncertainty instead
of certain problem. Firstly, (2d) is eliminated as {yb
k}is
Area b
Traffi c
Traffi c
Traffi c
Uncert ain Arrival
Traffic
Admis sion Cont rol
Stage 1
Admission contro l and resourc e
borrowing without knowing the
traffi c
Stage 2
BS associat ion a nd resou rce
allocation with knowing the
traffi c and users di stri butio n
MVNs em bedding requests
(admi ssion requ es ts)
Fig. 1. The two-stage MVN embedding mechanism including admission
control, resource borrowing and resource allocation.
optimized after admission control and resource borrowing.
Then, (2b) and (2c) are left without any changes because there
is no uncertainty in them.
In constraint (2e), two parameters, users distribution and
resource allocation, are unknown when we conduct admission
control and resource borrowing. Therefore, decoupling them
from (2e) is necessary but very hard. In this paper, we use a
feedback method to evaluate the capacity of a specific area.
Assuming the maximum capacity provided by one InP iin the
area of bat time tis cb,t
i, our expected maximum capacity is
Et[cb
i]=ηcb,t
i+(1−η)Et−1[cb
i](3)
where t−1means the last evaluation time and η∈[0,1] is
a constant to adjust the ratio of current network status and
previous network status. Same estimation method for mobile
networks can be found in [4]. Then, we can rewrite the stability
constraints (2e) of system as
K
k
xb
kρb
k−
I
i
αb
iEt[cb
i]≤0,∀b, k (4)
As {xb
k}are binary variables, problem (2) is a mixed integer
linear problem (MILP), which is a NP-hard problem. For
tractable solution of it, we relax the binary variables {xb
k}to
real value {ˆxb
k}∈[0,1] so that problem (2) becomes a linear
problem. {ˆxb
k}can be interpreted as partial admission decision
of {xb
k}, which means the physical network cannot embed
the whole MVN’s request but part of it. In long term view,
{ˆxb
k}can be considered as the average time when MVN kis
embedded in the physical network. We obtain the first stage
(optimize {ˆxb
k}and {αb
i}) of problem (2) formally formed as:
max
{ˆxb
k∈[0,1]},{αb
i}F({ˆxb
k},{αb
i})(5a)
s.t. (2c),
K
k
ˆxb
kρb
k−
I
i
αb
iEt[cb
i]≤0,∀b, k (5b)
Note that in problem (5) formulated above, there is no uncer-
tainty in {λb
k}. However, in practical network virtualization,
the real arrival rates vary with time, the arrival rate of users
given by a specific MVN is an estimated mean value. Thus,
the estimated arrival rate is actually a nominal value and not
accurate so that we have to tolerate this uncertain inaccuracy.
We define the nominal values of arrival rate {λb
k}as long term
mean {¯
λb
k}. Thus, in this paper, we assume the relationship
between the nominal arrival rates and the actual arrival rate is
based on a bounded but random parameters θand as follows:
λb
k=(1+θ)¯
λb
kwhere >0is the largest magnitude affecting
the uncertainty of λb
k, and θis a random variable supported in
[−1,1] with zero mean, which defines the possible fluctuation
of the arrival rates. λb
kmeans that the individual arrival rate λb
k
cannot deviate by more than θ¯
λb
kof the estimated arrival rate.
Thus, the possible deviation level is controlled by parameter
[11]. The MVNP can change this parameter based on historical
statistics (network traffic data) and robust level. As a larger
leads a better robust and a less results in less resource
conservation, adjusting can be used to balance the robustness
and resource conservation. These random parameters λb
kresult
in ρb
krandom and uncertain. Based on the definition of robust
linear problem and the similar expression in [10], [11], [14],
the relaxed version problem of (5) is a robust problem. For ease
of presentation, if θ=0, that means there is no uncertainty in
(5), and we call it nominal (5). Otherwise, robust counterpart
problem (RCP) (5) is used to indicate the robust problem.
In [11], the authors propose a robust model where if
the uncertain coefficients have bounded, symmetric support,
the corresponding robust feasible solutions must satisfy the
constraints with high probability [10]. As λb
khave bounded
and symmetric support in [¯
λb
k−¯
λb
k,¯
λb
k+¯
λb
k]with mean ¯
λb
k,
problem (5) follows the model proposed in [10]. Thus, it is
able to find a feasible solution ({ˆxb
k},{αb
i})∗that can satisfy
the constraints (4) with high probability only if ({ˆxb
k},{αb
i})∗
is also feasible to nominal (5).
Based on the conclusion in [11] and [10], it is effortless
to deal with RCP (5). Following the approach in them, if we
define the reliability level as κthat represents the violation
probability of (4) is at most κ, finding a feasible solution of
RCP (5) is equivalent to solving the following problem (6):
max
{ˆxb
k∈[0,1]},{αb
i}F({ˆxb
k},{αb
i})
s.t. (2c)
K
k
ˆxb
k¯ρb
k−
I
i
αb
iEt[cb
i]+Ω
K
k
ˆxb2
k¯ρb2
k≤0,∀b, k (6a)
where κ=exp{−Ω2/2}[11] and ¯ρb
k=(
¯
λb
krk)/μb
k.κcan be
interpreted as the overload probability of embedding a MVN.
In practice, since network capacity and traffic status vary with
time, it is reasonable to allow this low probability that some
users of MVN kat area bare not satisfied. Therefore, we also
can use probabilistic constraints to represent (4):
Pr{xb
kρb
k−
I
i
αb
iE+[cb
i]≥0}≤κ, ∀b, k (7)
Moreover, recall that {Et[cb
i]}are the evaluated capacity. To
guarantee a safer resource reservation, which means more
robust feasible solution to problem (6), we use E+[cb
i]=
(1+ξ)E[cb
i]in (6a) and (7) instead of only E[cb
i].ξ∈(0,1) is a
given and called feasibility tolerance [11]. Because both {ˆxb
k}
and {αb
i}are nonnegative, problem (6) is a convex problem
based on robust parameters (, κ, ξ). The objective function,
constraints (2c) are linear function while constraints (6a) are
convex (second-order norm plus some linear part) function.
When (, κ, ξ)are given, (6) is called (, κ, ξ)-(6). Solving it
can be done by many efficient methods, because the compu-
tational complexity of (, κ, ξ)-(6) is only O(|K|·|B|·|I|).
It should note that {ˆxb
k}are real values bounded on [0,1]
that is represented the partially admitting. However, if partial
admission is not allowed in the network, we have to recover
{xb
k}from {ˆxb
k}. There are many methods to round up real
value to integer, such as arbitrary selection [15] and marginal
benefit [16]. Due to the uncertainty of our optimization param-
eters that leads inaccurate solution, in this paper, we deploy
marginal benefit rounding method described as follows:
xb
k=0,if ∂F/∂xb
k|xb
k=0 >∂F/∂x
b
k|xb
k=1
1,otherwise (8)
where ∂F/∂x is the first partial derivation of x. After rounding
{ˆxb
k}to {xb
k}, we need to re-optimize the borrowing ratio
{αb
i}. Indeed, we can solve the problem (6) to get new {αb
i}
by fixed {ˆxb
k}. Here, we use {xb∗
k}and {αb∗
i}to indicate the
first-stage decision made by the MVNP for admission control
and resource borrowing, respectively.
C. Physical Resource Allocation for Users
In this subsection, we adopt a distributed algorithm [17] to
optimize the resource allocation for individual users in order
to embed the admitted MVNs.
Firstly, we need to present the users who are admitted to the
network and waiting for scheduling. If we use Ka
bto indicate
the set of MVNs that have been admitted to be embedded in
the physical networks area b, we can get
k∈Ka
b,if xb∗
k>0
Kb−K
a
b,otherwise (9)
As we mentioned above, multiple InPs are assumed to be exist-
ing at any area b. Thus, the resource allocation strategy actually
includes two variables, BS associate indicator and resource
ratio. {yb
k}used to represent the resource allocation strategy
have to extend to BS associate indicator ab
iukand resource ratio
βb
iuk. Assuming at each scheduling slot the users of MVN k
in the area of bare indicated by ub
k∈U
b
k=1.2, ..., |U b
k|,
the BS associate indicator represented user ub
k’s association
relationship with BS iat area bis denoted by ab
iukand
the resource allocation ratio is βb
iuk. Normally, in a certain
area, a user can only associate with one BS, which leads
ab
iuk∈{0,1}. However, coordinated transmission technology
enables multiple association, which results in ab
iuk∈[0,1].In
this paper we firstly assume multiple association is available.
Then we will adapt a rounding method to get binary ab
iukif
the system does not allow multiple association.
So far, the second stage of problem (2) becomes a traditional
joint BS association and resource allocation problem, because
the traffic load from each MVN is realized and available
physical resources are known. Since the utility and cost is
related to the {ˆxb
k}and {αb
i}, optimizing
βb
iukis used to
provide a feasible strategy that guarantee the QoS requirement
of MVNs. Thus, based on above discussion and resource
allocation policy that is fairness maximum allocation in this
paper, we can get the following optimization problem:
max
ab
iuk,βb
iuk
N
b
I
i
|Ka
b|
k
|Ub
k|
uk
ab
iuklog(βb
iukWb
irb
iuk)(10a)
s.t.
I
i
ab
iuk=1,∀uk,b (10b)
|Ub
k|
uk
ab
iukβb
iuk≤αb
ik,∀b, i, k (10c)
where Wb
iis the available bandwidth of BS iband rb
iukis
the spectrum efficiency of ukserved by BS ib. Following
the researches and similar expression in [18], problem (10)
can be transformed to an equivalent convex problem if we
define ˜
βb
iuk=ab
iukβb
iukand subtract them to problem (10).
Detailed proof can been found in [18] and the references
therein. Since it is a convex problem, it is not difficult to
solve it by using many methods (e.g., interior method [19] or
ADMM [20]). Due to the space limitation, we cannot give the
detailed procedure of solving this problem by ADMM. More
information can be found in our earlier paper [17]. Solving
problem (10) provides the solution to the second stage of
embedding MVNs and also estimated capacity cbiused in (3)
of the physical networks.
IV. SIMULATION RESULTS AND DISCUSSIONS
In the simulations, we consider three LTE-based RAN InPs
with 20MHz licensed spectrum. The locations of the macro
BSs are fixed in the center of each cell with a coverage 250
meters representing an urban environment. Transmit power
of 49dBm is set for macro BSs who have 20MHz spectrum
respectively [21]. Referred to [22], we user a path loss
L(d)=34+40log(d)to model the macro cell propagation.
The lognormal shadowing with standard deviation 8dB for
macro cell is assumed in our paper. The power density of
thermal noise power is -174dBm/Hz [21]. The unit leasing
price price set by these three RAN InPs are all 1unit/MHz.
One MVNP is assumed to provided MVNs to maximum 10
MVNOs. The unit access price of one MVN is set to be 2
units and blocking penalty is set to be 1 unit/MVN. The traffic
model assumed in our paper is similar to [14].
To compare our proposed algorithm, two benchmarks with-
out using RO are also considered. The first scheme, named
fixed arrival rate AC (fixed-AC), operates admission control
without considering the uncertainty of arrival rate. It only
controls the access of MVNs based on average users arrival
rate. Unlike first scheme, the second scheme, named max
arrival rate AC (max-AC), considers the maximum uncertainty
when performing admission control, which means they always
use at least (1+λ)as the arrival rate. For simple notation, our
proposed two-stage MVN embedding mechanism is denoted
by robust optimization admission control (ROAC).
In simulations, we select two metrics to verify our proposed
scheme. They are blocking probability of MVNs and number
of satisfied users. In the first simulation, we evaluate the effect
of the MVN arrival rate and users arrival rate on the blocking
probability of MVNs. In Fig. 2, the dotted lines represent user
arrival rate λu=6users/s and the solid lines represent user
arrival rate λu=4users/s. The x-axis is the arrival rate of
MVNs. As shown in Fig. 2, increases of both MVNs arrival
rate (2 to 3) and users arrival rate (4 and 6) higher the blocking
probability of MVNs. Firstly, higher users arrival rate leads
that more resources are needed for each MVNs. This results
in less position of potential accessed MVNs which affects
the blocking probability directly. Secondly, according to the
property of Poisson distribution, block probability increases
with the increase of arrival rate of MVNs as the network load
ρn=λn/μn>1. From Fig. 2, we can see that the max-
AC scheme blocks more MVNs. This is because the max-
AC scheme has the least accessible MVNs, and the max-AC
scheme has to always consider the maximum users arrival
rate. Meanwhile, if the admission control scheme (shown by
fixed-AC) does not consider the uncertainty in arrival rate, it
will allow the most MVNs to access. Obviously, our proposed
scheme shows a balance between max-AC and fixed-AC. Since
ROAC takes uncertainty into consideration, it shows more
cautious about admitting MVNs leading appropriate number
of positions of MVNs.
In addition, we evaluate the QoS performance of our pro-
posed ROAC. Satisfied ratio (a ratio between the number of
users served by required data rate to the total number of users)
is used to verify QoS performance. Fig. 3 shows that both max-
AC and ROAC guarantee the QoS requirement (satisfied ratio)
that is 95%. For max-AC, as it considers the maximum traffic
when performs AC, it has less users in the network. Thus,
almost every user can be served with full requirement. Our
proposed ROAC scheme cannot satisfy all the users because
we allow a probability (QoS requirement) that the system is
not stable in RO model. The worst case is given by the fixed
arrival rate AC, because it allows too much MVNs without
considering the uncertainty of arrival rate that directly affects
thetrafficinthenetwork.
Average arrival rate of MVNs
2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3
Blocking probability of MVNs
0
0.05
0.1
0.15
0.2
0.25
max-AC λu=6
ROAC λu=6
fixed-AC λu=6
max-AC λu=4
ROAC λu=4
fixed-AC λu=4
Fig. 2. The effect of user arrival rate in MVNs on the blocking probability.
Total arrival rate of users in MVNs
4 4.5 5 5.5 6 6.5 7 7.5
Satisfied ratio
0.8
0.82
0.84
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
max-AC
ROAC
fixed-AC
Fig. 3. Performance of different admission control schemes.
V. C ONCLUSIONS AND FUTURE WORK
In this paper, we jointly studied admission control and
resource allocation for wireless network virtualization. Firstly,
we decoupled the admission control and physical resource
leasing from physical resource allocation. Then, we formulated
the virtual network admission control and physical resource
leasing as an robust optimization problem. In addition, we
adopted an efficient virtual resource allocation scheme to
perform radio resource allocation and cell selection. Simu-
lation results were presented to show that the performance of
physical resource reservation can be substantially improved
and QoS requirements can be guaranteed. Future work is in
progress to consider full-duplex relaying [23] and information-
centric networking [24] in the proposed scheme.
ACKNOWLEDGMENT
This work was supported in part by the Natural Sciences
and Engineering Research Council of Canada and in part by
Huawei Technologies Canada CO., LTD.
REFERENCES
[1] C. Liang and F. R. Yu, “Wireless network virtualization: A survey, some
research issues and challenges,” IEEE Comm. Surveys Tutorials, vol. 17,
pp. 358–380, Firstquarter 2015.
[2] C. Liang and F. R. Yu, “Wireless virtualization for next generation
mobile cellular networks,” IEEE Wireless Comm., vol. 22, pp. 61–69,
Feb. 2015.
[3] K. Pentikousis, W. Yan, and H. Weihua, “Mobileflow: Toward software-
defined mobile networks,” IEEE Comm. Mag., vol. 51, pp. 44 – 53, Jul.
2013.
[4] Y. Zaki, LTE Optimization and Mobile Network Virtualization. Springer,
2012.
[5] L. Chen, F. R. Yu, H. Ji, G. Liu, and V. Leung, “Distributed virtual
resource allocation in small cell networks with full duplex self-backhauls
and virtualization,” IEEE Trans. Veh. Tech., 2015. accepted.
[6] M. I. Kamel, L. B. Le, and A. Girard, “Lte wireless network virtualiza-
tion: Dynamic slicing via flexible scheduling,” in Proc. IEEE Veh. Tech.
Conf. (VTC) - Spring, pp. 1–5, 2014.
[7] M. Yang, Y. Li, J. Liu, D. Jin, J. Yuan, and L. Zeng, “Opportunistic
spectrum sharing for wireless virtualization,” in Proc. IEEE Wireless
Commun. & Networking Conf. (WCNC), pp. 1803–1808, 2014.
[8] F. Yu, V. Krishnamurthy, and V. C. M. Leung, “Cross-layer optimal
connection admission control for variable bit rate multimedia traffic in
packet wireless CDMA networks,” IEEE Trans. Signal Proc., vol. 54,
pp. 542–555, Feb. 2006.
[9] S. Zhang, F. R. Yu, and V. Leung, “Joint connection admission control
and routing in IEEE 802.16-based mesh networks,” IEEE Trans. Wireless
Commun., vol. 9, pp. 1370 –1379, Apr. 2010.
[10] D. Bertsimas, D. B. Brown, and C. Caramanis, “Theory and applications
of robust optimization,” SIAM review, vol. 53, no. 3, pp. 464–501, 2011.
[11] A. Ben-Tal and A. Nemirovski, “Robust solutions of linear programming
problems contaminated with uncertain data,” Mathematical program-
ming, vol. 88, no. 3, pp. 411–424, 2000.
[12] B. K. Chalise and L. Vandendorpe, “Optimization of mimo relays
for multipoint-to-multipoint communications: Nonrobust and robust
designs,” Signal Processing, IEEE Transactions on, vol. 58, no. 12,
pp. 6355–6368, 2010.
[13] M. Hasan, E. Hossain, and D. I. Kim, “Resource allocation under
channel uncertainties for relay-aided device-to-device communication
underlaying lte-a cellular networks,” IEEE Trans. Wireless Commun.,
vol. 13, no. 4, 2014.
[14] S. Yun and C. Caramanis, “System-level optimization in wireless net-
works: managing interference and uncertainty via robust optimization,”
ACM/IEEE Trans. Networking, vol. 20, pp. 339–352, Apr. 2012.
[15] S. Gortzen and A. Schmeink, “Optimality of dual methods for discrete
multiuser multicarrier resource allocation problems,” IEEE Trans. Wire-
less Commun., vol. 11, no. 10, pp. 3810–3817, 2012.
[16] G. Liu, F. R. Yu, H. Ji, and V. Leung, “Energy-efficient resource
allocation in cellular networks with shared full-duplex relaying,” IEEE
Trans Veh. Tech., 2014, on-line. doi: 10.1109/TVT.2014.2359892.
[17] C. Liang and F. R. Yu, “Distributed resource allocation in virtualized
wireless cellular networks based on ADMM,” in Proc. IEEE INFO-
COM’15 Workshops, Apr. 2015.
[18] G. Liu, F. R. Yu, H. Ji, and V. Leung, “Distributed resource allocation
in full-duplex relaying networks with wireless virtualization,” in Proc.
IEEE Globecom, pp. 4959–4964, Dec. 2014.
[19] S. Boyd and L. Vandenberghe, Convex optimization. Cambridge univer-
sity press, 2009.
[20] S. Boyd, N. Parikh, E. Chu, B. Peleato, and J. Eckstein, “Distributed
optimization and statistical learning via the alternating direction method
of multipliers,” Foundations and Trends in Machine Learning,vol.3,
no. 1, pp. 1–122, 2011.
[21] 3GPP, “Evolved Universal Terrestrial Radio Access (E-UTRA); Further
advancements for E-UTRA physical layer aspects,” TR 36.814 V9.0.0,
3rd Generation Partnership Project (3GPP), Mar. 2010.
[22] Q. Ye et al., “User association for load balancing in heterogeneous
cellular networks,” IEEE Trans. Wireless Commun., vol. 12, pp. 2706–
2716, Jun. 2013.
[23] G. Liu, F. R. Yu, H. Ji, V. Leung, and X. Li, “In-band full-duplex
relaying: A survey, research issues and challenges,” IEEE Comm.
Surveys Tutorials, vol. 17, pp. 500–524, Secondquarter 2015.
[24] C. Fang, F. R. Yu, T. Huang, J. Liu, and Y. Liu, “A survey of green
information-centric networking: Research issues and challenges,” IEEE
Comm. Surveys Tutorials, vol. 17, pp. 1455–1472, Thirdquarter 2015.
