Content uploaded by Sunday Oladayo Oladejo
Author content
All content in this area was uploaded by Sunday Oladayo Oladejo on Dec 30, 2019
Content may be subject to copyright.
Latency-Aware Dynamic Resource Allocation
Scheme for 5G Heterogeneous Network: A
Network Slicing-Multitenancy Scenario
Sunday O. Oladejo
Department of Electrical Engineering
University of Cape Town
Cape Town, South Africa
oldsun002@myuct.ac.za
Olabisi E. Falowo
Department of Electrical Engineering
University of Cape Town
Cape Town, South Africa
olabisi.falowo@uct.ac.za
Abstract—In 5G Slice Networks, the multi-tenant multi-tier
heterogeneous network will be critical in meeting the Quality
of Service (QoS) requirement of the different slice use cases;
Capital Expenditure (CAPEX) and Operational Expenditure
(OPEX) reduction for mobile network operators. Hence, the 5G
slice networks should be as flexible as possible to accommodate
the different network dynamics such as of user location and
distribution, different slice use case QoS requirement, cell load,
intra-cluster interference and Mobile Virtual Network Opera-
tors (MVNOs) Service Level Agreement (SLA). Motivated by
this, this paper addresses a latency-aware dynamic resource
allocation problem for 5G Slice Networks in a multi-tenant
multi-tier heterogeneous environment, for efficient radio resource
allocation. The latency-aware dynamic resource allocation prob-
lem is formulated as a maximum utility optimisation problem.
The optimisation problem is transformed and the hierarchical
decomposition technique is adopted to reduce the complexities
in solving the optimisation problem. Furthermore, we proposed
a Genetic Algorithm (GA) Intelligent Latency-Aware Resource
Allocation scheme (GI-LARE). Our findings have been obtained
by extensive Monte Carlo numerical simulations and comparisons
were carried out with the static allocation scheme and a branch
and bound method. The genetic algorithm schemes out perform
the static scheme.
Index Terms—Multi-tenancy, Network Slicing, Heterogeneous
Network, Genetic Algorithm, Resource Allocation
I. INT ROD UC TI ON
The IMT-2020 widely regarded as the Fifth Generation (5G)
mobile networks will play a critical role in the realisation
of the Industry 4.0 revolution [1], which will reshape the
way humans and machines co-habit and live. Although this
fact is indisputable, the deployment of the 5G Networks
requires substantial financial investment and commitment by
the different mobile network players such as the Infrastructure
Providers (InP), Mobile Virtual Network Operators (MVNO),
Over-the-top players (OTP), backhaul Operators (BO). In a
bid to reduce the Capital Expenditure (CAPEX) and Operating
Expenditure (OPEX), different multi-tenancy business models
for the 5G Networks have evolved and emerged. In this work,
we modify the models in [2] [3] to suite multi-tenant network
slicing, as shown in Fig 1. In this business model, the InPs
have the ownership of the physical resources of the mobile
Fig. 1. InP-MVNO Model
network resources such as the power, cooling systems, towers
and mast, antennas, spectrum, virtualisation equipment such as
hypervisors. These resources are virtualised and allocated to
the MVNOs, who in turn provide services to their respective
slice users. In this multi-tenancy scenario, the major challenge
is how the InP dynamically, efficiently and effectively allocate
resources to the MVNOs without breaching the Service Level
Agreement (SLA) with their clientele and also meeting the
QoS requirement of different slice use cases. In this work,
without loss of generality, we consider two broad types of slice
use cases: the enhanced Mobile broadband (eMBB) and the
massive Machine-Type Communications (mMTC) slice use
cases. The eMBB slice use case is a bandwidth-hungry use
case, while the mMTCs is quite a latency or delay-dependent
use case. Unlike most works in the literature, we investigate
the concept of Network Slicing not only from the received data
rate QoS requirement; but also from the latency or delay QoS
requirement of the different slice use case. Besides, unlike the
widely investigated 2-Tier heterogeneous network environment
in the study of 5G Network Slicing, we examine the concept of
Network Slicing deployed or implemented in a hierarchical 3-
tier clustered heterogeneous network. While being modest, this
comes with its complexities. Necessarily, these are conditions
that sufficiently motivate the purpose of this paper to consider
latency-aware radio resource allocation in 5G Slice Network
in a multi-tenancy environment.
2019 International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)
978-1-7281-3316-4/19/$31.00 ©2019 IEEE
A. Contributions
The contributions of this paper are as follows:
•The radio resource allocation problem between the InP
and the MVNOs; the different Slice use cases; the dif-
ferent tiers of the heterogeneous networks; and between
the slice users is jointly formulated as an optimization
problem with feasible constraints in order to examine the
impact of network parameters on the network utility.
•A matching game is played between the slice users and
3-tier clustered heterogeneous network in a multi-tenancy
scenario in order to associate slice users to particular
access points optimally.
•Other than the data rate QoS constraints, we examine the
latency constraints are examined in the Network Slicing
concept.
•A Genetic Algorithm (GA) Intelligent Latency-Aware
Resource Allocation scheme (GI-LARE) is proposed to
enhance the overall network utility. The Genetic Algo-
rithm (GA) Intelligent Latency-Aware Resource Allo-
cation scheme (GI-LARE) is compared with an exact
solution from the Spatial Branch and Bound method and
a static resource allocation scheme.
II. RE LATE D WOR KS
The authors in [4], [5] presented maximum capacity and
profit-aware resource allocation schemes for network slicing
in a multi-tenancy scenario. Slice priorities and bandwidth-
Power cost were considered, however, a static resource scheme
between the InP and MVNOs was adapted. Besides, the
latency constraints requirement of the respective slices was
not considered. Also, it is pertinent to state that a single-
tier homogeneous network was considered, which does not
entirely depict a 5G network with its complexities. In [6],
a maximum-revenue resource allocation optimisation problem
is formulated for a virtual network in a 2-tier heterogeneous
network. In solving this problem, the authors pre-allocated
radio resources to the respective base stations or access point.
In [7], a dynamic resource sharing scheme for a single-tier
homogeneous C-RANs multi-tenancy is proposed. A network
utility maximisation problem was formulated while consid-
ering the tenants’ priorities. Although the two-step proposed
sub-optimal solution improved the network utility, however,
users were not categorised based on their slice requirements.
Moreover, the authors in [8], presented a dynamic radio
resource slicing scheme for a 2-tier heterogeneous wireless
network. An alternating concave search algorithm is designed
to solve the maximum network utility optimization problem,
the 2-tier heterogeneous network due to it simplistic model
may not fully represent a 5G network environment with its
many tiers of access networks in order to meet the ever-rising
user demands. Besides the authors do not address the concept
of multi-tenancy, which is critical for CAPEX and OPEX
reduction in 5G networks. Motivated by the shortcomings
in the literature, we investigate the latency-aware dynamic
resource allocation problem in a 3-tier clustered heterogeneous
network for multi-tenancy network slicing.
Fig. 2. A 3-Tier Clustered Heterogeneous Network
III. SYS TE M MOD EL
A. General Model
In this work, the downlink of a 3-tier clustered heteroge-
neous 5G Slice Network, as illustrated in Fig. 2 is considered.
The physical resources of this network are owned, managed
and controlled by the Infrastructure Provider I, and these re-
sources (Bandwidth in this case) are allocated to the MVNOs.
Let the set of MVNOs be H. Each respective MVNO, h∈ H is
totally independent and unaware of the other MVNOs h0∈ H,
its set of Slices is denoted by S={E ∪ M}. Where Eand
Mdenotes the eMBB slice and mMTC slice sets which are
explained in details in subsection III B. The hierarchical 3 -
tier network comprises of the femtocells, picocells and macro-
base stations respectively. The set of femtocells is denoted by
SC ={F ∪ PF }.F={1,2, ...f}represents the set of femto
access points located only in the coverage area of the Macro
base station, while PF ={1,2, ..., pf }denotes the set of
clustered femto in the coverage area of a picocell, p∈ P, and
macro base station, m.Pis the set of picocells in the pico
tier. As stated earlier, the coverage area of a picocell, p, helps
to create a cluster area for a set of femtocells. The entire
heterogeneous network environment is overlaid by a macro
base station, m. For a particular MVNO, h, slice user, iis
categorized according to its requested slice use case (Eor M)
and besides, its geographical position in the heterogeneous
network.
B. Slice User Categorization
Without loss of generality, slice users are categorised into
four types.
•Cat. I: Em, is the set of eMBB users belonging to MVNO,
h, in the coverage region of monly. Besides, the mMTC
slice user belonging to MVNO, h, in the coverage area
of monly is denoted as Mm.
•Cat. II: The set of eMBB users in the coverage of a femto,
f∈ F, is denoted as Efand similarly, Mffor the set of
mMTC slice users in the coverage of a femtocell f∈ F.
•Cat. III: For the pico-tier, the set of eMBB slice users in
the coverage area of a picocell p∈ P but does not belong
2019 International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)
to PF is represented as Ep. Similarly, Mpdenotes the
set of mMTC slice users which are under the coverage
area of p∈ P and not in the clustered femto set PF.
•Cat. IV: Likewise, Epf denotes the set of eMBB slice
users in one of the clustered femtocells, pf ∈ PF. For
mMTC slice users in the clustered femto, pf ∈ PF, its
set is denoted by Mpf . This categorisation is illustrated
in Fig. 2. The total number of cluster, c∈ C is the same
as |P|.
C. Channel Model
With the downlink of a link layer model given in [9] [10],
we consider a user iwith a path loss given as:
ρi,j,h =
30 + 35 log(di,j,h),if i∈ {Em,Mm}, j =m
35 + 35 log(di,j,h),if i∈ {Ep,Mp}, j =p
40 + 35 log(di,j,h),if i∈ {Ef,Epf ,Mf,Mpf }
(1)
Where di,j,h is the distance of the slice user, i, belonging
to MVNO, h, from base station (or access points), j∈
{m, f, p, pf }. For a user iwhich request for slice use case
{Em,Mm}and belongs to MVNO, h, and is in the coverage
of monly, its spectrum efficiency is given as:
γi,m,h = log21 + ψm,hΓi,m,h
σ2(2)
Similarly for a user i∈ {Ef,Mf}, its spectrum efficiency
is given as:
γi,j,h = log21 + ψj,hΓi,j,h
X
k∈{F}
j6=k
ψk,hΓi,k ,h +σ2(3)
Likewise for a user i∈ {Ep,Mp}, its spectral efficiency is
given as:
γi,j,h = log21 + ψj,hΓi,j,h
X
k∈{P}
j6=k
ψk,hΓi,k ,h +σ2(4)
For a user i∈ {Epf ,Mpf }who subscribes to the services of
MVNO, h, its spectrum efficiency is given as:
γi,j,h = log21 + ψj∈c,h Γi,j∈c,h
X
p0∈c0
c6=c0X
k∈{PF 0}
j6=k
PF 6=P F0
ψk,hΓi,k ,h +σ2(5)
ψj,h is the transmit power of the access point or the base
station, j∈ {m, F,P,PF}.
X
p0∈c0X
k6=j, k∈{PF 0}
ψk,hΓi,k ,h in (5) is the inter-
cluster interference of for a user iin the clustered
femto cells . The channel gain Γi,j,h, between a user
i,which belongs to a MVNO, h, and associated with base
station, j∈ {m, F,P,PF} is given as:
Γi,j,h = 10−ρi,j,h
10 (6)
D. Latency and Delay Model
For a mMTC slice user, its minimum achievable rate ui,m,h
, must be greater or equal to the effective bandwidth, uthres
h,
[9] [10] of the network source. The effective bandwidth is
a function of the delay bound violation probability, (q), the
average data packet arrival rate, λh,Mj, of mMTC slice users,
the maximum delay bound threshold, D, and the data packet
size, Lh,Mj, of the mMTC user. Therefore from [9] [10],
uthres
h=−Lh,Mjlog(q)
Dln(1 −log(q)
Dλh,Mj
)(7)
In this work, Dand qare set at 100ms and 0.001, although
these values could be varied. The network parameters for
λh,Mjand λh,Ejare 5 and 20 packets per seconds. The data
packet size for the eMBB and mMTC (Lh,Ej&Lh,Mj) are
9000 bits and 2000 bits respectively.
E. Dynamic Network Slicing
In this work, the bandwidth is virtualised and pooled to a
centralised cloud server and reallocated to MVNOs. In a multi-
tenancy scenario, this is carried out in a 2-step hierarchical
method and shown in Fig. 1. The bandwidth allocated to each
MVNO in each tier t, is given as:
X
t∈{MB S,SC ,P}
βt,hB(8)
However, to fully maximise the bandwidth of the InP, the
total network slicing ratio must be equal to 1, as seen in (8):
X
h∈H X
t∈{MB S,SC ,P}
βt,h = 1 (9)
Where βt,h is the network slice ratio of MVNO, h, in tier
t, while Bis the network bandwidth. As stated earlier, βt,h
is a function of slice user distribution and location, cell load
characteristics and user QoS.
Furthermore, we consider a logarithmic utility for each cate-
gory of user specified in Section III (B).
log ui,j,h= log Bβt,h αi,j,h γi,j,h(10)
Where αi,j,h is the fraction of the bandwidth allocated to
a slice user i, belonging to a tier j∈ {m, F,P,PF} with
bandwidth Bβt,h in MVNO h.
IV. PROBLEM FORMULATIO N
In this section, we consider the latency-aware dynamic
resource allocation problem in a 3-tier clustered heterogeneous
network in a network slicing multi-tenancy scenario. In
order to fully maximise the capacity of the network, we
formulate a joint user-association InP-MVNO resource
allocation problem in (11). The utility of each MVNO,
is the summation of the utility (or rate) of the four
categories of slice users explained in Section III (B) and
2019 International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)
max X
h∈H "X
i∈(Em∪Mm)
log ui,m,h+X
f∈F X
i∈(Ef∪Mf)X
j∈{m,f}
yi,j,h log(ui,j,h)
+X
p∈P X
i∈(Ep∪Mp)X
j∈{m, p}
y0
i,j,h log ui,m,h+X
p∈P X
pf∈PF X
i∈(Epf ∪Mpf )X
j∈{m, p, pf}
y00
i,j,h log ui,m,h#
(11)
s.t.
C1: ui,m,h ≥λh,EmLh,Em∀i∈ Em,∀h∈ H
C2: ui,m,h ≥uthres
h∀i∈ Mm,∀h∈ H
C3: yi,j,hui,m,h −uthres
h≥0∀i∈ Mf, j ∈ {m, f },∀h∈ H
C4: yi,j,hui,m,h −λh,EmLh,Em≥0∀i∈ Ef, j ∈ {m, f },∀h∈ H
C5: y0
i,j,hui,m,h −uthres
h≥0∀i∈ Mp, j ∈ {m, p},∀h∈ H
C6: y0
i,j,hui,m,h −λh,EpLh,Ep≥0∀i∈ Ep, j ∈ {m, p},∀h∈ H
C7: y00
i,j,hui,m,h −uthres
h≥0∀i∈ Mpf , j ∈ {m, p, pf},∀h∈ H
C8: y00
i,j,hui,m,h −λh,Epf Lh,Epf ≥0∀i∈ Epf , j ∈ {m, p, pf },∀h∈ H
C9: X
j∈{m,f}
yi,j,h = 1 ∀i∈(Ef∪ Mf)
C10: X
j∈{m, p}
y0
i,j,h = 1 ∀i∈(Ep∪ Mp)
C11: X
j∈{m, p, pf}
y00
i,j,h = 1 ∀i∈(Epf ∪ Mpf )
C12: yi,j,h ∈ {0,1} ∀ i∈(Ef∪ Mf), j ∈ {m, f },∀h∈ H
C13: y0
i,j,h ∈ {0,1} ∀ i∈(Ep∪ Mp), j ∈ {m, p},∀h∈ H
C14: y00
i,j,h ∈ {0,1} ∀ i∈(Epf ∪ Mpf ), j ∈ {m, p, pf},∀h∈ H
C15: X
i∈(Em∪Mm)
αi,m,h +X
f∈F X
i∈(Ef∪Mf)
yi,j,hαi,m,h +X
p∈P X
i∈(Ep∪Mp)
y0
i,j,hαi,m,h +X
p∈P X
pf∈PF X
i∈(Epf ∪Mpf )
y00
i,j,hαi,m,h = 1
C16: X
f∈F X
i∈(Ef∪Mf)
yi,f,h αi,f,h = 1 ∀h∈ H
C17: X
p∈P X
i∈(Ep∪Mp)
y0
i,p,h αi,p,h +X
p∈P X
pf∈PF X
i∈(Epf ∪Mpf )
y00
i,p,hαi,p,h = 1 ∀h∈ H
C18: X
p∈P X
pf∈PF X
i∈(Epf ∪Mpf )
y00
i,pf,hαi,pf ,h = 1 ∀h∈ H
C19: αi,m,h ∈(0,1) i∈(Em∪ Mm∪ Ef∪ Mf∪ Ep∪ Mp∪ Epf ∪ Mpf )
C20: αi,f,h ∈(0,1) i∈(Ef∪ Mf)
C21: αi,p,h ∈(0,1) i∈(Ep∪ Mp∪ Epf ∪ Mpf )
C22: αi,pf,h ∈(0,1) i∈(Epf ∪ Mpf )
Constraints C1, C4, C6and C8in (11) ensures that the
minimum data rate for the eMBB users is guaranteed for all
tiers ı.e for all the four categories of users in the respective
tiers. Also, C2, C3, C5and C7guarantees that the received
rate, ui,m,h, of an mMTC slice user in the different categories
is by no means less than the effective bandwidth threshold
uthres
hof the mMTC slice. Besides, yi,j,h,y0
i,j,h and y00
i,j,h are
BS-user association indicator; in this work, we consider C9,
C10 and C11 to ensure that a Cat. II slice user is either associ-
ated with a femtocell, f, or the MBS, m; Cat. III slice user is
either associated with a picocell, por MBS, m; and a Cat. IV
slice user is either associated with a clustered femtocell, pf,
or picocell p, or the MBS, mrespectively. While constraints
C12, C13 and C14 is a relaxation of constraint C9, C10 and
C11. Constraints C15, C16, C17, C18 in general highlights the
bandwidth allocation requirement of the individual slice users
in each of the tiers. C19, C20, C21 and C22 is a relaxation
of the fractional allocation indicator which must be between
0and 1ı.e a positive fractional value.
V. PROPOSED SOLUTIONS
The proposed solution to the problem is discussed in this
Section IV is a combination of different algorithms. To make
2019 International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)
(11) tractable and solvable, we first simplify (10):
log Bβt,h αi,j,h γi,j,h= log Bβt,hγi,j,h+ log αi,j,h
(12)
The expression in (12) is then applied to the transformed
objective function in (11). The transformation of the objective
function (11) is shown the appendix.
A. Base Station-User Association
Using the hierarchical decomposition method in [11], firstly,
we solve the Base station-User association problem by adapt-
ing the maximum SINR association to determine the tier or
base station a user associates. A seen in Fig.2, slice users who
are in the coverage of the Macro base stations only, cannot
associate with any other tier because they are not under the
coverage of other tiers. Algorithm 1 solves the Base station-
user association in a 3-tier clustered heterogeneous network
through matching game.
B. Genetic Algorithm
We solve the transformed dynamic resource allocation prob-
lem in a clustered heterogeneous network in a multi-tenancy
scenario via Continuous Genetic Algorithm (CGA) [12], meta-
heuristic bio-inspired approach because of its ability to find
optimal solutions in a brief time. In solving (11), we restrict
the CGA search space using the constraints C1 to C22 for
optimality and fast computational convergence. We adapt
Elitism [13] to the CGA in the process of gene selection. The
probabilities of crossover, Pc, mutation, Pmand elitism Pe
play critical roles in the performance of the CGA. In this work,
Pc= 0.9,Pm= 0.9and Pe= 0.2. We choose number of
chromosomes, NChrom, to be 400; the number of variables,
NoV , is the product of |H| and the number of tiers (4). The
maximum iteration (MaxIter)is set at 50. The complexity of
the GI-LARE is given as: ONC hrom ∗NoV ∗MaxIter
A Spatial branch and Bound (BnB) method is adapted to
solve the maximisation problem in (11) in order to verify the
optimality of the CGA-based results. The BnB algorithm gives
a global optimal solution, and it is shown in Algorithm 3.
VI. RE SU LTS AN D DIS CU SS IO N
We consider 3-tier clustered heterogeneous network
whereby slice users subscribed to 3MVNOs over the area
of interest of 800 m radius. The femtocells, picocells and
macrocells all have coverage radius of 50 m,250 m and 800 m
respectively and transmit powers of 30 dBm,36.9 dBm,
40 dBm. The different categories of slices users are randomly
distributed across the different base stations in all the tiers,
with data packet arrival rate of 5 packets/s; 20 packets/s for
mMTC and eMBB users in the different categories. The per-
formance of the proposed algorithm is evaluated over a large
number of different simulations. A Monte-Carlo simulation
in Matlab environment is created, with a 10,000 iterations
for each simulation to validate the results. Fig. 3 show the
effect of the maximum delay bound, D, on the effective
bandwidth, uthres
h. The role of the uthres
hcan be seen in
constraints C2, C3, C5 and C6 which greatly affects the
Algorithm 1: Base Station-User Association
Data: Set Parameters
Set: ψj,h,di,j,h
if i is under the coverage of a femtocell then
if femtocell is clustered then
calculate: γi,m,h,γi,p,h ,γi,pf,h (2)(4)(5)
if γi,pf,h ≥(γi,m,h &γi,p,h)then
y00
i,pf,h = 1
else
if γi,p,h ≥(γi,m,h &γi,pf,h )then
y00
i,p,h = 1
else
y00
i,m,h = 1
end
end
break ;
else
calculate γi,m,h,γi,f ,h (2)(3);
if γi,f,h ≥γi,m,h then
yi,f,h = 1
else
yi,m,h = 1
end
break ;
end
else
Calculate γi,p,h,γi,m,h (4)(5);
if γi,p,h ≥γi,m,h then
y0
i,p,h = 1
else
y0
i,m,h = 1
end
end
Algorithm 2: Continuous Genetic Algorithm
Set Pm;Pe;Pc,MaxIter,NChrom,NoV
Step 1: Initialize the random population
Step 2: while iteration ≤MaxIter
iteration = iteration + 1
a: using (11) det. the fitness of each chromosome
b: sort the chromosomes fitness in descending order
c: select the chromosomes wrt. to fitness
d: carry out crossover using the Pc
e: mutate the genes of chromosomes with Pm
f: perform elitism on the initial population wrt. Pe
g: select population based on (d),(e),(f)
Step 3: end while
Algorithm 3: Spatial Branch and Bound
Step 1: Initialize the upper bound, ωub, of (11)
Set the list of region Gto a single domain
Step 2: Use the least lower bound rule to choose a
subregion A ∈ G
if G=∅then Go to Step 6
if the lower bound in region A,ωA,lb, is infeasible
or ωA,lb ≥ωub −πthen Go to Step 5
Step 3: Obtain the upper bound, ωA,ub
if ωA,ub > ωub then Go to Step 4
else ωub := ωA,ub delete all subregions in G
if ωA,ub −ωA,lb ≤πthen Go to step 5
Step 4: Partition A into subregions Aleft andArig ht
Step 5: Delete Afrom Gand Go to Step 2
Step 6: End Search
if ωub =∞then problem is infeasible
else ωub is the global optimal of the solution
2019 International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)
received rate of the mMTC slice users. We observe that as
the maximum delay bound increases, the threshold decreases
which is in tandem with (7). Fig.4. presents the relationship
between the average arrival rate of slice users and the slicing
ratio of the network. Fig. 4 clearly shows that, for the different
MVNOs, the slicing ratio of the dynamic GI-LARE algorithm
varies with the varying network parameters rather than the
Static slicing algorithm for the different average arrival rate
examined. Fig 4. shows that the dynamic GI-LARE algorithm
ensures fairness to the different tiers while at the same time
maximising the network utility.
Fig. 3. The effect of the maximum delay bound on the effective bandwidth
threshold
Fig. 4. Impact of slice user average arrival rate λon the Slicing Ratio, βt,h
for respective MVNOs
Fig.5 shows the impact of the average arrival rate of the
slice users on the total network utility. In Fig. 5 we assume
the slice users in the different categories have the same average
arrival rate. It can be observed that as the average arrival
increases, the total network utility improves. Besides, the
Dynamic GI-LARE algorithm performs quite better than the
static-allocation algorithm. The Dynamic GI-LARE scheme
performance is quite the same as the BnB approach, which
gives a globally optimal solution.
Fig. 5. Effect of the slice user average arrival rate on the total utility of the
network
Fig. 6 presents the impact of the eMBB data packet size on
the network utility. It can be observed that as the packet size
increases the net utility increases until about 4000 bits and
then a dip occurs in the net utility. This can be ascribed to the
bandwidth limitation of the network. However, the dynamic
GI-LARE approach outperforms the static resource allocation.
Fig. 6. Effect of the eMBB data packet size on the total network utility
VII. CO NC LU SI ON
In this paper, a Genetic Algorithm (GA) Intelligent Latency-
Aware Resource Allocation scheme (GI-LARE) that explicitly
takes into consideration the latency and data rate constraints
slices in a multi-tenant heterogeneous is proposed. The opti-
mization problem is transformed and solved using the hierar-
chical decomposition method. The slice users are associated
with base stations in different tiers and, the bandwidth resource
allocation problem is solved using GA, BnB and a static
approach. With the aid of Monte-Carlo simulations, extensive
numerical analysis was performed showing different scenarios,
2019 International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)
and the dynamic GI-LARE scheme far outperforms the static
approach.
VIII. ACK NOWLEDGEMENT
The authors gratefully acknowledge the support of the
National Research Foundation (NRF) and Telkom Centre of
Excellence (COE) in Broadband Networks, South Africa.
REF ER EN CE S
[1] N. Dao, Y. Lee, S. Cho, E. Kim, K. Chung, and C. Keum, “Multi-
tier multi-access edge computing: The role for the fourth industrial
revolution,” in 2017 International Conference on Information and Com-
munication Technology Convergence (ICTC), Oct 2017, pp. 1280–1282.
[2] A. Belbekkouche, M. M. Hasan, and A. Karmouch, “Resource discovery
and allocation in network virtualization,” IEEE Communications Surveys
Tutorials, vol. 14, no. 4, pp. 1114–1128, Fourth 2012.
[3] S. Kukli´
nski, L. Tomaszewski, K. Kozłowski, and S. Pietrzyk, “Business
models of network slicing,” in 2018 9th International Conference on the
Network of the Future (NOF), Nov 2018, pp. 39–43.
[4] S. O. Oladejo and O. E. Falowo, “5g network slicing: A multi-tenancy
scenario,” in 2017 Global Wireless Summit (GWS), Oct 2017, pp. 88–92.
[5] ——, “Profit-aware resource allocation for 5g sliced networks,” in 2018
European Conference on Networks and Communications (EuCNC), June
2018, pp. 43–9.
[6] C. Liang, F. R. Yu, H. Yao, and Z. Han, “Virtual resource allocation
in information-centric wireless networks with virtualization,” IEEE
Transactions on Vehicular Technology, vol. 65, no. 12, pp. 9902–9914,
Dec 2016.
[7] Z. Jian, W. Muqing, M. Ruiqiang, and W. Xiusheng, “Dynamic resource
sharing scheme across network slicing for multi-tenant c-rans,” in
2018 IEEE/CIC International Conference on Communications in China
(ICCC Workshops), Aug 2018, pp. 172–177.
[8] Q. Ye, W. Zhuang, S. Zhang, A. Jin, X. Shen, and X. Li, “Dynamic radio
resource slicing for a two-tier heterogeneous wireless network,” IEEE
Transactions on Vehicular Technology, vol. 67, no. 10, pp. 9896–9910,
Oct 2018.
[9] Dapeng Wu and R. Negi, “Effective capacity: a wireless link model
for support of quality of service,” IEEE Transactions on Wireless
Communications, vol. 2, no. 4, pp. 630–643, July 2003.
[10] ——, “Effective capacity-based quality of service measures for wireless
networks,” in First International Conference on Broadband Networks,
Oct 2004, pp. 527–536.
[11] D. P. Palomar and M. Chiang, “A Tutorial on Decomposition Methods
for Network Utility Maximazation,” IEEE Journal on Sel. Areas in
Comms. (JSAC), vol. 24, no. 8, pp. 1439–1451, 2006.
[12] S. Mirjalili, Evolutionary Algorithms and Neural Networks. Reading,
Massachusetts: Springer International Publishing, 2018.
[13] H. Du, Z. Wang, W. Zhan, and J. Guo, “Elitism and distance strategy
for selection of evolutionary algorithms,” IEEE Access, vol. 6, August
2018.
2019 International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)