Professional Documents
Culture Documents
https://doi.org/10.1007/s11276-020-02292-5 (0123456789().,-volV)(0123456789().
,- volV)
Abstract
This paper provides a V2X slicing logical architecture based on Software Defined Networking. We devote a special
concern to inter-slice mobility management. In this context, we propose a complete mobility solution that handles
resources management through the use of bargaining games enhanced by a heuristic and particular swarm optimization
algorithm. More specifically, this work proposes a resource borrowing scheme to readjust resources of an overloaded slice
in case of a inter-slice handover. The borrowing procedure is modeled via two bargaining games. Performance analysis
shows that our proposed solution guarantees fairness between V2X slices, reduces the probability of call dropping and
blocking probabilities and enhances resource utilization. Moreover, a comparative study between the proposed games is
provided and analyzed.
Keywords Network slicing Vehicle to everything Mobility management Inter-slice handover Bargaining game
Resource borrowing
123
3884 Wireless Networks (2020) 26:3883–3903
Admission control consists of checking the availability In [6], authors formulate a new network model for
of resources in the Radio Access Network (RAN). More providing 5G network slices computational and storage
specifically, in case there are no resources to accept a resources in order to guarantee QoS requirements. The
handover request in a slice, an efficient resource manage- slicing mechanism is based on an auction model designed
ment should be achieved between slices before the han- to maximize network resources.
dover occurrence. Admission control becomes very critical Authors of [7] introduce a resource management
in a V2X environment. In fact, dropping handover requests framework that aims at provisioning and auto-scaling slices
should be avoided since exchanged data carry urgent road in real time. This scheme uses a utility function subject to
safety messages. Thus, an efficient admission control with network bandwidth and cloud processing capabilities.
a resource management scheme should be carefully In [8], authors implement static and dynamic resource
studied. management schemes for different slices supporting vari-
The main concern of this work is to study the inter-slice ous services. The proposed algorithm considers the fluc-
handover management by deriving an admission control tuations in network traffic load conditions in different
with an efficient resource management scheme. More network slices. It dynamically shares available radio
precisely, we focus on the case of inter-slice handover resources based on a pre-defined sharing agreement.
while the target slice is overloaded. A resource borrowing A solution named Resources nEgotiation for NEtwork
from other available slices is performed in this case, in Virtualization (RENEV) is proposed in [9]. The algorithm
order to be able to accept the handover request. The aims at achieving an efficient mapping of radio virtual
problem is modeled using two different bargaining games. network elements onto the radio resources of the existing
Moreover, a simulation is conducted in order to validate physical network.
the resource borrowing scheme. A comparative analysis is In [10], authors introduce two schemes to allocate LTE
provided in order to understand each proposed game resources to Machine-to-Machine emergency deployments
characteristics and draw pertinent conclusions on the in an adaptive manner. These schemes ensure that the
advantages of each proposed game. resource allocation responds to the changing needs of the
The paper is structured as follows. In Sect. 2, we underlying emergency application.
overview literature works related to our problem and shed In [11] authors propose a dynamic bandwidth slicing
the light on our contributions. Section 3 presents 5G ref- and resource allocation problems to support a mixture of
erence slices and introduces V2X slices. We present the IoT (Internet of Things) and video streaming services. This
adopted 5G slicing architecture in Sect. 4 and dedicate is achieved by employing Lyapunov optimization method
Sects. 5 and 6 to give a detailed explanation about our with time-scale separation approach. Authors formulate
solution. Section 7 is conceived to evaluate our proposed algorithms for long time-scale bandwidth slicing, and short
algorithms. Finally, we conclude our paper in Sect. 8. time-scale IoT device scheduling, power allocation (for IoT
service) and quality decision (for video streaming service).
In [12], a Network Slicing Resource Management
2 Related works (NSRM) mechanism is proposed. NSRM aims to ensure the
isolation of allocated resources, fair resource sharing and
As mentioned earlier, resource management can help customized slice configuration. Authors propose a novel
improving the call admission procedure. Network slicing is architecture framework for virtualizing the LTE network in
envisioned by two different models concerning the order to maximize network resources utilization.
resource management: dedicated resources model and The work in [13] provides an analysis of a well-known
shared resources model. In the former, each slice is model for resource sharing, the ’share-constrained pro-
assigned a fixed amount of resources isolated from other portional allocation’ mechanism, to realize network slicing.
slices. This can ensure isolation among slices and guar- This mechanism enables tenants to reap the performance
antee a committed amount of resources to each slice. benefits of sharing, while retaining the ability to customize
However, it reduces the slice elasticity and can limit their own users allocation. This results in a network slicing
multiplexing gain. The second approach (shared resources game in which each tenant reacts to the user allocations of
model) allows the slice to share its physical resources. This the other tenants so as to maximize its own utility.
approach exploits statistical multiplexing. Despite that, Authors of [14] propose Hap-SliceR: a novel radio
sharing resources will raise outstanding challenges in terms resource slicing framework for 5G networks with haptic
of QoS guarantee, fairness among slices and mobility communications. An optimal slicing strategy, based on a
support. In this context, many researchers focus on finding reinforcement learning approach is proposed. It consists of
a wise resource management scheme for network slices. allocating radio resources to different slices while taking
into account the dynamics and utility requirements of
123
Wireless Networks (2020) 26:3883–3903 3885
different slices. Second, Hap-SliceR provides customiza- introduces the concept of channel borrowing strategy. The
tion of radio resources for haptic communications over 5G proposed scheme allocates the guard channels for handoff
networks. requests dynamically, based on the traffic load for certain
In [15], a resource reservation and allocation (RRA) time period. Moreover, a new originating call can use these
with uncertain demands of mobile users is formulated as a guard channels if they are unused. More precisely, when a
robust optimization model. Logarithmic utility functions new originating call arises and if all the available channels
are defined to capture the mobile users satisfaction, which are occupied it will check for the guard channels. If the
show how to match the allocations between radio resources latter are unused, the new call will occupy a guard channel.
and virtual machine resources according to the resource In this approach, the number of allocated guard channels
demands of the mobile applications. is determined through optimizing certain performance
In classical wireless networks, several research works goals with service quality constraints. When a base station
have tackled resource management and channel reservation experiences high handover call dropping rate, the number
strategies while taking into account mobility constraints. of guard channels will be increased until this rate drops to
The latter are designed by either reserving a fixed number below a predefined threshold. When a base station does not
of channels or managing the reservations dynamically. The get to use a significant portion of the guard channels over a
fixed reservation schemes ([16, 17]) are very simple with period of time, the number of guard channels is gradually
no computation overhead. Nevertheless, these schemes are decreased until most of the guard channels are used fre-
not flexible to cope with load fluctuation and induce an quently. We note that in this scheme, requests are buffered
inefficient use of network resources. over a certain duration in order to check if available
Dynamic reservation schemes are proposed to overcome resources are released.
the disadvantages of the fixed reservation schemes. In [18], Several resource management works proposed in liter-
the number of reserved channels is related to the requested ature are based on bargaining games. In [26], a novel
bandwidth of the ongoing connection. Each base station bargaining approach for dynamic spectrum management
monitors the handover call dropping probability and (DSM) based on alternating-offer bargaining games is
channel utilization to adapt the channel reservation. proposed. The scheme introduces a trading agent (TA), a
In [19], channel reservation is adjusted according to a changing bargaining ability and a revenue-sharing mecha-
prediction method that aims at provisioning handover nism to facilitate the spectrum trading.
probability based on handover occurrences history. The authors in [27] introduce a dynamic bargaining
Authors in [20] present an adaptive QoS (AQoS) algo- game for radio resource sharing among the primary users
rithm that consists of reducing QoS levels of traffic calls, in (the licensed mobile operators) and the secondary users
order to accept handover calls requests. (Mobile virtual operators). The study also suggests that the
In [21], a flexible resource-allocation (FRA) strategy is secondary users relay the primary users’ traffic. This model
designed. It consists of prioritizing QoS of particular ser- allows both licensed and virtual mobile operators to offer
vice types over the others. FRA aims at releasing band- higher data rates to their end-users.
width from the low priority calls based on a prioritized call Authors of [28] model datacenter bandwidth allocation
degradation policy to accept higher priority call requests. as a cooperative bargaining game, with two main objec-
Authors of [22] design an algorithm to reserve resources tives: 1) guarantee bandwidth for virtual machines based
with neighbor cells by exchanging information related to on their bandwidth requirements, and 2) share residual
movement and direction of users. bandwidth in proportion to the weights of virtual machines.
In [23], channel borrowing is proposed by dividing The before mentioned works can be discussed as
traffic into two classes: real time traffic and non-real time follows:
traffic. The scheme attempts to allocate desired bandwidth
• For the resource management works in slicing envi-
to every connection, and considers that in case of insuffi-
ronment, the dynamic aspect of these schemes relies on
cient bandwidth, resources will be borrowed from existing
load fluctuation and bandwidth availability only. In
connection.
these research papers, a slight effort is devoted to
In [24], authors investigate the offloading policy in
mobility which is a challenging concern for V2X slicing
heterogeneous wireless networks. In this paper, a queue
environment. In fact, in a vehicular architecture, users
model is built to formulate the mobile users workload
mobility requires a reduced handover call dropping
offloading problem and Lyapunov optimization framework
probability. This is achieved by an efficient resource
is proposed to make trade-off between system offloading
management scheme.
utility and queue backlog.
• The resource management schemes in wireless net-
In [25], authors propose a scheme that dynamically
works cannot be applied in a slicing environment. In
allocates guard channels for handover requests and
123
3886 Wireless Networks (2020) 26:3883–3903
fact, network slicing consists of isolating the resources • We provide a V2X slicing logical architecture that
of a physical base station, aiming at guaranteeing high details the different components involved in 5G vehic-
degree of QoS for each slice. Thus, the concept of ular networks.
reserving bandwidth from ongoing connections or • We propose a complete mobility solution in a V2X
reducing QoS of lower priority connections is contra- network slicing environment.
dictory to the slicing concept. • We shed the light on inter-slice handover management
Moreover, in a vehicular environment, bandwidth solution. More specifically, we present a resource
reduction may not be always efficient. In fact, ongoing borrowing scheme based on bargaining game.
sessions may be related to safety and remote controlling • We distinguish between two different formulation of
vehicles. Thus, non of the latter services can handle a bargaining games in order to resolve the problem of
QoS reduction in order to accept more requests. overloaded slices. Moreover, in an attempt to reduce the
• Bargaining games based approach are used to achieve complexity of games solutions, we propose a heuristic
fairness among users. However, all the mentioned and apply particle swarm optimization.
approaches, based on bargaining games for resource • We validate our platform by studying the following
management, do not consider mobility constraints. performance parameters: handover dropping probabil-
Table 1 compares the various resource management ity, new call blocking probability and resource utiliza-
schemes proposed in literature. tion; the studied scenario tackles a tele-operated driving
Motivated by the before mentioned issues, we orient our use case where vehicles are communicating to a
efforts to derive an admission control with a resource monitoring center.
management scheme dedicated to a V2X slicing architec-
ture. Unlike the aforementioned papers, our work aims at
considering mobility of users in V2X slicing environment. 3 Network slices types
The novelty of our proposed scheme is based on a resource
management scheme that can guarantee the best QoS of 3.1 5G reference slices
V2X messages (specially tele-operated driving messages),
while reducing the handover call dropping probability. The Fifth Generation Public Private Partnership (5G-PPP)
More precisely, if a handover request cannot be accepted in [29] defines three reference slices: enhanced Mobile
a slice due to the lack of resources, a resource borrowing BroadBand slice, massive Machine-type Communications
from other available slices is performed in order to guar- slice, and Ultra-Reliable and Ultra-Low Latency
antee the acceptance of the handover request. The bor-
rowing mechanism should respect several constraints such
as QoS of lender slices, and should guarantee fairness
among these slices. Our contributions are listed as follows:
[6–15] Slicing 7 Dynamic Resource allocation based on load factor and user
requirements
[16, 17] No Slicing 4 Static Simple, with no computation complexity
Classical wireless Cannot cope with load fluctuation
networks
[18–25] No Slicing 4 Dynamic Mobility consideration, with a priority to handover calls
Classical wireless QoS reduction of ongoing sessions.
networks
[26–28] Based on bargaining 7 Dynamic Fairness among users
games No mobility consideration
123
Wireless Networks (2020) 26:3883–3903 3887
Communications slice. These slices are described as fol- driving slice (exchanges safety messages) and infotainment
lows (Fig. 1): slice (provides video stream). Authors in [30] present four
sets of V2X slices according to the variety of V2X use
• Enhanced mobile broadBand (eMBB) slice requires
cases. These slices, listed as follows (Fig. 2), are used in
very high data rates to fulfill requirements of multime-
our work:
dia content, like ultra-high definition video streaming.
• Massive machine-type communications (mMTC) slice • Autonomous driving slice This slice is concerned with
This slice should sustain the massive traffic load of two types of communication. First it supports ultra low-
connected devices, transmitting non-delay sensitive latency V2V communication. Second, low-latency and
information, e.g., sensor networks deployed in smart reliable video/data exchange needs to be supported with
cities. a V2X Application Servers (AS) deployed at the
• Ultra-low latency communications (URLLC) slice this network edge.
slice should provide services that are extremely sensi- • Tele-operated driving slice This slice should ensure
tive to latency, such as autonomous driving, tactile ultra-low latency and highly reliable end-to-end con-
internet and augmented reality. It requires reliability, nectivity between the controlled vehicle and a tele-
low latency, and security. operation station which is hosted outside the Core
Network (CN).
• Vehicular infotainment slice This slice should guarantee
3.2 V2X slices
the high throughput infotainment data and services
exchanged with an infotainment V2X AS.
The variety of V2X services do not allow a straightforward
• Vehicle remote diagnostics and management slice This
mapping into the aforementioned reference slices. In fact,
slice is configured to support the exchange of low-
safety applications, autonomous driving, and vehicular
frequency data between many vehicles and remote V2X
remote controlling should be simultaneously supported.
AS outside the CN.
Each type of these services present different requirements
in terms of QoS and mobility support. This raises the need
to design network slices dedicated for each V2X use case.
V2X network slices are proposed and elaborated in 4 V2X slicing architecture
papers [30–33]. Authors in [32] shed the light on the
benefits of using network slicing in V2X to increase the Inspired by the slicing concept proposed in 5G NORMA
possibility of creating smart and secure traffic. In [33], a [4] project and ETSI logical architecture [34], we propose
network slicing based communication model for vehicular the following slicing architecture depicted in Fig. 3 . The
networks is proposed. Authors have modeled a multi-lane latter consists of three planes: the infrastructure plane, the
highway scenario with vehicles having heterogeneous control plane and the service plane. In addition, we add
traffic requirements. Messages are sent through two logical orchestration capabilities for NFV MANagement and
slices created on a common infrastructure: autonomous Orchestration (MANO) on the top of this architecture.
123
3888 Wireless Networks (2020) 26:3883–3903
123
Wireless Networks (2020) 26:3883–3903 3889
4.3 The service plane the SDO via Interface 2. At the same time, SDO can
configure these resources via this interface.
The service plane includes services and use cases of each – Interface 3 This interface ensures the exchange between
vertical market for which slices are designed. S-SDNC and D-SDNC. This exchange is related to
attachment, or mobility issues.
4.4 The management and orchestration (MANO) – Interface 4 It provides interaction between S-SDNC/D-
plane SDNC and forwarding elements in order to achieve
path configuration and radio resources management.
The MANO plane is responsible of the slice description,
instantiation and life-cycle management. The MANO plane
4.6 Physical architecture
consists mainly of an SDN controller named Software
Defined Orchestrator (SDO). The latter enables the bro-
This paper considers geographical areas owned by different
kering of resources among multiple slices of the same
operators (Fig. 4). Each operator deploys two technologies,
operator. Moreover, SDO exchanges information with peer
namely IEEE 802.11p [35], LTE-V [36]. Moreover, oper-
entities of other mobile network operators to enable
ator PoAs are connected to the available slices; each slice
seamless inter-slice handover among operators.
is controlled by a D-SDNC. It is noteworthy that all slices
in one operator domain are under the control of a S-SDNC.
4.5 Interfaces
123
3890 Wireless Networks (2020) 26:3883–3903
download the route from its current location to its desti- will be connected to the previous and target slice
nation. When a signal degradation is detected, the vehicle simultaneously (the vehicle is considered with two
sends a Signal Going Down Message to the D-SDNC. LS interfaces). A make before break handover is achieved
periodically forwards the calculated route information to in this case, by adjusting network settings before the
the D-SDNC that can predict if intra/inter slice is disconnection from the previous PoA. Packets dupli-
imminent. cation is used to guarantee a seamless handover. This
According to the prediction results, the mobility man- procedure is elaborated in our previous work in [37].
agement proceeds as follows (Fig. 5): 2. Whenever inter-slice handover is about to happen,
D-SDNC sends a request to the S-SDNC via Interface 3
1. When an intra-slice handover is imminent, a network
which transfers the request to the MANO plane SDO
selection procedure is performed within the same slice.
via Interface 2. A communication between operators
The selection of the target PoA is based on QoS
SDO is performed via Interface 1. In this case several
metrics and the estimated load of available networks.
scenarios should be considered:
The handover to the selected PoA is achieved as
follows: after the choice of the target PoA, the vehicle
123
Wireless Networks (2020) 26:3883–3903 3891
• Whenever the target operator possesses the same guarantee all the amount of resources required for the
slice type than the current slice, the D-SDNC of this handover request in the overloaded slice. This game is
slice performs admission control in order to check mainly concerned in the first and second constraints related
the resources availability: to QoS of the handover request and slices load. The second
game (Game II) proposes a compromise between the
– In case of the estimated load is sufficient, the
required resources for the imminent handover request and
target slice D-SDNC prepares and config-
the available resources of the lender slices. Thus, it slightly
ures path to the handover request in the new
reduces QoS of handover sessions. This game combines the
slice. Path configuration and handover execu-
first, second and fourth constraints. Concerning the third
tion is achieved according to our work in [37].
constraint related to fairness, we will show that Nash
– In case of congestion in the target slice, the
Bargaining Solution (NBS), used as game solution, can
S-SDNC of the target operator executes a
achieve fairness among slices. Our main purpose is to study
resource borrowing procedure. This is achieved
the performance of each formulation in order to draw
with the help of target operator SDO that is
important conclusions concerning the best resource bor-
responsible of resources brokering among
rowing strategy to be deployed.
slices. The resource borrowing algorithm will
After a performance study and a comparative analysis of
be detailed in the next section.
the proposed games, we will provide a discussion in order
• If the target operator does not have the same slice to draw conclusion about the advantages and disadvantages
type, a slice selection should be achieved in order to of this game.
choose the most adequate slice that can guarantee
QoS requirements of the imminent handover
request. This idea is studied in [38].
123
3892 Wireless Networks (2020) 26:3883–3903
6.2 Bargaining games for resource borrowing among bargaining players, i.e, in our case each involved
slices will offer a fair amount of resources. One proposition
Our proposed scheme is based on resource borrowing to achieve this concept is max-min method [44], where the
between slices in order to guarantee the acceptance of a performance of the worst case player is maximized.
handover call while taking into account QoS constraints. In However, this can penalize players with better conditions
this context, due to the selfishness [40] of lender slices and and thus can generate a reduced overall system perfor-
the scarcity of spectrum resources, the problem of realizing mance. In this paper, we use the Nash Bargaining Solution
fair and effective resources borrowing has become of a (NBS) that can provide a unique and fair Pareto optimality
primordial importance. In fact, the rational behavior of solution if it satisfies six axioms listed in [45]. The coop-
slices will make them offer a small amount of resources, erative bargaining problem solution maximizes the Nash
making other slices offer a higher amount. This may product as follows:
damage the overall benefit of the system. Therefore, rele- Y
K
vant constraints need to be established among the slices to max ðRk Rmin
k Þ ð1Þ
enable them to cooperate with each other and to improve k¼1
the overall efficiency of the system as much as possible. Since the ln is a continuous strictly increasing function,
In daily life basis, people can negotiate about goods or solving the problem in Eq. (1) is equivalent to the fol-
transactions in order to achieve their satisfaction. Similarly, lowing solution:
in a resource borrowing problem, network slices can !
negotiate about their offered resources in a way to achieve YK XK
min
maxln ðRk Rk Þ ¼ max lnðRk Rmin
k Þ ð2Þ
a common agreement and maximize their gains. This idea
k¼1 k¼1
motivates us to model our problem using bargaining games
([41, 42]) described as follows. Indeed, maximizing the sum in Eq. (2) is more simpler to
implement than maximizing the product in Eq. (1).
6.3 Bargaining game definition
6.5 Bargaining game for resource borrowing
Bargaining game is a cooperative game in which players
have the opportunity to reach a win-win situation. In this As mentioned earlier, whenever an inter-slice handover is
game, there is a conflict of interest among the participants. imminent, and the estimated load of the target slice is
If one player cannot accept the bargaining solution, the beyond a predefined threshold, the resource borrowing
overall agreement will not be reached. algorithm should be triggered. We model the resource
Let K ¼ f1; 2; . . .K g be the set of players. Let S be a borrowing between slices using a bargaining problem.
closed and convex subset of RK that represents the set of The slices that can participate in the bargaining game,
feasible payoff that the players can get if they cooperate. i.e, slices that can offer resources to the overloaded slice
Let Rmin be the minimal payoff that the kth player would should respect the following conditions:
k
expect in order to participate in the game and cooperate. • Let Bktotal is the amount of resources allocated for slice
Suppose Rk 2 SjRk Rmin k 8k 2 K is a nonempty boun- k 2 K.
ded set. Then, with Rmin ¼ ðRmin min min
1 ; R2 ; . . .RK Þ, the pair • A slice can be considered as a lender slice if
min
ðS; R Þ , is called a K-person bargaining problem [43]. Bk Bktotal :b, where Bk is the actual amount of resources
of the slice k and b is a load factor.
6.4 Nash bargaining definition • Bkfree the amount of free resources of slice k where
Bkfree ¼ Bktotal Bk .
Within the feasible set S , the notion of Pareto optimal is
• Each slice k presents a priority ak , where 0\ak 1.
defined as a selection criterion for the bargaining solutions
The higher the priority of the slice is, the lower ak is.
[43].
• The amount of offered resources should be limited in
A utility vector ðR1 ; R2 ; . . .RK Þ is considered as Pareto
order to avoid an overload occurrence in the slice itself.
optimal, if and only if there is no other R0k such that
To this end, we define a borrowable window bk for each
R0k Rk , i.e., there exists no other distribution that results
slice k calculated as follows: bk ¼ Bkfree :ak :x, where x
with a higher utility for some players (slices) without
causing a utility degradation for other slices. is the borrowing factor. It is to be noted that slices with
There may be several Pareto optimal solutions. Thus, in higher priority will present a small borrowable window.
order to find a unique result, a criterion could be added to
the problem: the fairness. The latter refers to a fair result
123
Wireless Networks (2020) 26:3883–3903 3893
• Let L ¼ fk 2 Kjbk 6¼ 0g be the set of slices that can slice, by respecting the borrowable window and required
offer resources to the overloaded slice. We consider resources constraints.
thus L lender slices. Complexity of NBS The NBS is conventionally found
• xk is the amount of offered resources by slice k. based on the exhaustive search, i.e., all feasible utility
• xreq is the amount of resources required by the values in the entire feasible utility set are examined.
overloaded slice. However, the exhaustive search becomes significantly
inefficient when the feasible utility set becomes large, in
In our case, we consider that Rmin as a null vector, thus
terms of a higher number of slices and larger borrowable
ðL; 0Þ defines the bargaining problem.
windows. Hence, in order to reduce the search complexity,
we present in the following, a low complexity heuristic in
6.6 Game I formulation
order to solve NBS for bargaining game I.
Low complexity heuristic In the following, we give a low
We consider that the players are the lender slices that aim
complexity heuristic in order to solve NBS for bargaining
at maximizing their utility while offering an amount of
game I.
resources in order to guarantee xreq .
In order to render the model tractable, we consider that
6.6.2 Low complexity heuristic algorithm
the utility function depends only on the utility of each slice
k 2 L offering an amount of resources xk . The utility of
We propose a heuristic (Algorithm 1) that aims at solving
each slice k is formulated as follows:
problem in Eq. (6) and reducing its complexity.
Rk ¼ Bkfree xk ð3Þ For the sake of simplicity, we consider the following
assumption: all lender slices have the same borrowable
Additionally, in order to guarantee the handover request
window b ¼ minðbk Þ8k 2 L. This is done in order to
requirements, all players should obey to the following
reduce the amount of offered resources from each slice.
constraint:
The algorithm proceeds as follows.
X
L
xk ¼ xreq ð4Þ 1. Let Sb be the set of all possible values of xk , that is
k¼1 Sb ¼ f1; 2; 3; . . .; bg.
2. Let S0b the subset of Sb where
Moreover, each slice should respect the borrowable win- n X o
dow constraint that is formulated as follows: S0b ¼ xk jxk 2 Sb ; xk ¼ xreq and CardðS0b Þ ¼ L :
0 xk bk 8 k 2 L ð5Þ
6.6.1 Nash bargaining solution for game I
123
3894 Wireless Networks (2020) 26:3883–3903
Heuristic complexity In the proposed heuristic, we aim maxxk ln u xk þ lnðRk ðxk ÞÞ ð10Þ
k¼1
at reducing the borrowable window. Moreover, unlike the X
combinatorial problem imposed by NBS, we propose to s.t. xm xk xreq ð11Þ
decompose the feasible set into several feasible subsets.
The complexity of this heuristic is OðL2 CardðSb ÞÞ. The 0 xk bk 8k 2 L ð12Þ
algorithm has linear complexity in the number of subsets Where Rk and bk are defined in Eqs. (3) and (5)
and quadratic complexity in the number of slices, and thus respectively.
could be easily implemented in real-time. The NBS obtained by solving problem (10) aims at
achieving a compromise between the lender slices and the
6.7 Game II problem formulation overloaded slice. The amount of resources offered to the
overloaded slice is obtained according the sigmoid utility
In this game formulation, we consider that the players are function. Moreover, a fair amount of resources is offered
the lender slices, in addition to the overloaded slice. The from each lender slice. In this case, lender slices will offer
utility of each slice k is given by Eq. (3). While the utility a reduced amount of resources compared to game I solu-
of the borrowing (overloaded) slice is given by the fol- tion, and the overloaded slice will obtain a fair amount of
lowing sigmoid utility function: resources that can guarantee the QoS of the ongoing ses-
8
>
> xreq x c sion in request.
>
> As mentioned earlier, the NBS is very complex since it
>
<1 xreq xm
if xm \x xreq involves an exhaustive search. Thus, in order to reduce the
uðxÞ ¼ xreq x c
>
> 1þ search complexity of Game II solution, we propose to use
>
> xreq xm
>
: the Particle Swarm Optimization algorithm defined in the
1 if x [ xreq
following paragraph.
Where c and n values determine the steepness of the utility
curve and make it possible to model the slice sensitivity to 6.7.2 Particle swarm optimization
nðxreq xm Þ m xa Þ
amount of resources. c ¼ xm xa and n maxð2ðx
xreq xm ; 2Þ.
Particle Swarm Optimization (PSO) [47] is a heuristic
Moreover, xm is the minimal acceptable required resources
optimization algorithm, characterized by few parameters.
that can satisfy the need of the incoming handover request,
PSO is based on social behavior and presents a solution set
while xa is the threshold value, that is for a number of
called a swarm, and each solution in the set is referred as
resources less than xa , we obtain a null utility. Further
particle. Particles move in the problem search space
explanation about this utility function can be found in [46].
striving to reach the optimal solution. This is similar to the
This utility function is a non-decreasing function of x:
P collective behavior of birds and fish, that exchange their
the amount of borrowed resources where x ¼ xk .
knowledge of the search space to find the best solution.
Additional resources result in a higher utility value.
At each iteration, PSO ensures that each particle’s
Moreover, the improvement of the utility fades when the
velocity Vk and the amount of borrowable resources xk are
offered resources reach xreq . This implies the concavity of
updated as follows:
u(x) for x greater than a given value. Moreover, it is to
indicate that this utility function is chosen in order to Vktþ1 ¼ k Vkt þ c1 ðPbestk xtk Þ þ c2 ðGbest xtk Þ
conserve a good QoS for the handover request in the xtþ1 ¼ xtk þ Vktþ1
k
123
Wireless Networks (2020) 26:3883–3903 3895
Where t is the iteration number and c1 ; c2 are learning rates speed ranging between 50 km/h and 80 km/h, in order to
coefficients representing the weight of memory of a parti- specify the amount of time that will be spent in each cell.
cle’s best position: Pbestk , toward the memory of the swarm This paper focuses on a specific V2X use case: the tele-
best position: Gbest . k is the weighting value that indicates operated driving where a vehicle is supposed to commu-
the effect of the previous velocity on the velocity update. nicate with a tele-operation station. The communication
A particle keeps track of its coordinates in the search requires extreme real-time and exhibits tight requirements
space and aims to reach Gbest . The best solution is deter- for the network to ensure fast vehicle control and feedback.
mined by the value of the fitness function, which is the It should guarantee ultra-reliable and low latency com-
utility function to be maximized (Eq. 10) The PSO fitness munication. In order to guarantee these requirements,
function, F, for the resource borrowing problem of game II, mobility management should be carefully studied.
is described in Eq. 13: The load and arrival rate in this slice are adequately
(P P PL adjusted, in order to obtain an admission control denial and
k lnðRk ðxk ÞÞ þ lnðuð xk ÞÞ if k¼1 xk xm thus trigger the resource borrowing algorithm.
F¼ P P PL
k lnðRk ðxk ÞÞ þ lnðuð xk ÞÞ þ g xm k¼1 xk otherwise Table 2 presents the parameters values used in the
ð13Þ simulation.
We proceed the evaluation as follows. First, the fairness
where the penalty value g [ 0. The penalty value accom-
obtained by using NBS is evaluated. Second, a comparison
modates the practical constraint that the sum of offered
of Game I and Game II is conducted. Third, some network
resources should exceed the minimum threshold value xm .
performance parameters such as the handover call dropping
probability, the new call blocking probability and resource
utilization are evaluated. And finally, the effectiveness of
using PSO as a method to solve Game II NBS is validated.
123
3896 Wireless Networks (2020) 26:3883–3903
mathematical formulation of the NBS that aims at In the following, we compare the utility of borrowing and
achieving fairness. lender slices obtained with each NBS1 and NBS2.
The arrival rate is fixed to 15 vehicle/min.
7.1.2 Amount of offered resources
123
Wireless Networks (2020) 26:3883–3903 3897
123
3898 Wireless Networks (2020) 26:3883–3903
7.3.2 New call blocking probability in lender slices Resource utilization is a primordial parameter that can
evaluate a resource management scheme. In fact, an ade-
Another metric to evaluate our solution is the new call quate scheme should maximize the resource utilization in
blocking probability in lender slices. In fact, the lender order to avoid their inefficient use.
slices are supposed to offer an amount of resources to the Figure 12 confirms that our proposed method results in
overloaded slice (tele-operated driving slice). This may high values of resource utilization. This is due to the fact
cause a reduction in the available bandwidth of these slices, that static and static guard schemes are based on a static
and thus, may cause a high call blocking probability. To
this end, we evaluate the call blocking probability in these
slices.
Figure 11 shows that static guard scheme results with a
high new call blocking probability. This is due to the static
reserved amount of resources offered to the handover
requests and that cannot be used by new requests. For the
dynamic guard scheme, with a low arrival rate, the new call
blocking probability is low, since the number of guard
channels reserved for handover calls is low in this case.
With the arrival rate increase, the dynamic guard
scheme results in higher call blocking probability than our
approach, since this scheme will reserve more guard
schemes for handover calls. However, our proposed
schemes can manage resources dynamically between slices
while taking into account the load constraint of lender Fig. 12 Resources utilization
123
Wireless Networks (2020) 26:3883–3903 3899
allocation and thus for a low arrival rate, resources are not As we can notice, NBS scheme induces additional delays
used efficiently. due to the computation and signaling overhead. In order to
Moreover, with the dynamic guard scheme, resource measure the handover latency and the overhead brought by
utilization is near to the static resource allocation NBS scheme, we present simulation results illustrated in
scheme results. This is due to the fact, that despite the Fig. 13.
guard channels reservation used in dynamic guard As we can see in this figure, NBS presents the highest
scheme method, the latter allows the use of these reserved handover latency among other schemes. This is due to the
resources by new calls. Our proposed method outperforms additional computations to obtain NBS and signaling
the dynamic guard scheme. This is due to the fact that our exchanged between D-SDNC and S-SDNC, in order to
proposed resource borrowing method allows a resource determine the number of offered resources. The dynamic
exchange between slices. While with the dynamic guard guard scheme presents lower handover latency since it
scheme method, even though this method adjusts the guard includes distributed decisions, without signaling overhead
channel dynamically and allows resource borrowing for between S-SDNC and D-SDNC. Moreover, the static and
new calls, some reserved resources may not be used effi- static guard schemes present the lowest handover latency
ciently due to the isolation of resources between slices. since no additional computations should be done for the
More precisely, if a request arrives to a slice and all the handover execution.
resources are used in this slice, the request will be dropped/ Despite the fact that our approach causes an increase in
blocked. Meanwhile, other slices may have available the handover latency, we can see that this latency reaches
resources that are not used efficiently. It is to be noted that, 12 ms for a high arrival rate. This shows that the our
the resource utilization is obtained as the average value of approach respects V2X latency constraints [30], and does
resource utilization of all slices in the network. not violate ultra reliable and low latency communication
Regarding NBS1 and NBS2, we can see that NBS1 requirements for a V2X tele-operated driving use case.
results with a higher resource utilization since it aims at
offering more resources than NBS2. 7.4 Studying the impact of x
7.3.4 Handover latency We evaluate the impact of the borrowing factor x on the
new call blocking probability and handover call dropping
In order to evaluate the handover duration, we present first probability. Results are illustrated in Figs. 14 and 15. It is
an analysis about the handover duration. To this end, we remarkable that varying x reduces the handover call
define the following parameters in Table 3. dropping probability in the tele-operated driving slice.
We can obtain the handover duration of each scheme as However increasing x results in increasing the new call
follows: blocking probability in lender slices, since the latter need to
Tstatic ¼ Tcheck þ Tconfig þ Tatt release more resources; this reduces their available
Tstaticguard ¼ Tcheck þ Tconfig þ Tatt bandwidth.
Tdynamic ¼ Tcheck þ Tconfig þ Tatt þ Tbuffer
TNBS ¼ Tcheck þ Tconfig þ Tatt þ Tsignaling þ TNBSC
123
3900 Wireless Networks (2020) 26:3883–3903
7.6 Discussion
123
Wireless Networks (2020) 26:3883–3903 3901
123
3902 Wireless Networks (2020) 26:3883–3903
multimedia networks using the shadow cluster concept. IEEE/ double auction. IEEE Transactions on Wireless Communications,
ACM Transactions on Networking, 5(1), 1–12. 14(8), 4093–4104.
23. Ei-Kadi, M., Olariu, S., & Abdel-Wahab, H. (2002). Rate-based 41. Yaı̈che, H., Mazumdar, R. R., & Rosenberg, C. (2000). A game
borrowing scheme for QOS provisioning in multimedia wireless theoretic framework for bandwidth allocation and pricing in
networks. IEEE Transactions on Parallel and Distributed Sys- broadband networks. IEEE/ACM Transactions on Networking,
tems, 13(2), 156–166. 8(5), 667–678.
24. Li, Y., Xia, S., Cao, B., Liu, Q., et al. (2019). Lyapunov opti- 42. Grosu, D., Chronopoulos, A. T., & Leung, M. Y. (2002). Load
mization based trade-off policy for mobile cloud offloading in balancing in distributed systems: An approach using cooperative
heterogeneous wireless networks. IEEE Transactions on Cloud games. In Ipdps (p. 52).
Computing. 43. Han, Z., Ji, Z., & Liu, K. R. (2005). Fair multiuser channel
25. Alagu, S., & Meyyappan, T. (2012). Efficient utilization of allocation for OFDMA networks using nash bargaining solutions
channels using dynamic guard channel allocation with channel and coalitions. IEEE Transactions on Communications, 53(8),
borrowing strategy in handoffs. arXiv preprint arXiv:1206.3375. 1366–1376.
26. Pan, M., Liang, S., Xiong, H., Chen, J., & Li, G. (2006). A novel 44. Rhee, W., & Cioffi, J. M. (2000). Increase in capacity of mul-
bargaining based dynamic spectrum management scheme in tiuser ofdm system using dynamic subchannel allocation. In
reconfigurable systems. In Proceedings of the international Proceedings of the 51st IEEE Vehicular Technology Conference,
conference on systems and networks communications (ICSNC’06) VTC-Spring (pp. 1085–1089).
(pp. 54–54). 45. Stefanescu, A., & Stefanescu, M. V. (1984). The arbitrated
27. Yan, Y., Huang, J., & Wang, J. (2012). Dynamic bargaining for solution for multi-objective convex-programming. Revue Rou-
relay-based cooperative spectrum sharing. IEEE Journal on maine de Mathematiques Pures et Appliquées, 29(7), 593–598.
Selected Areas in Communications, 31(8), 1480–1493. 46. Nazim Nguyen-Vuong, Q. T., Ghamri-Doudane, Y., & Agoul-
28. Guo, J., Liu, F., Lui, J. C., & Jin, H. (2015). Fair network mine, N. (2008). On utility models for access network selection in
bandwidth allocation in IaaS datacenters via a cooperative game wireless heterogeneous networks. In Network Operations and
approach. IEEE/ACM Transactions on Networking, 24(2), Management Symposium NOMS (pp 144–151). IEEE.
873–886. 47. Eberhart, R., & Kennedy, J. (1995). Particle swarm optimization.
29. Elayoubi, S., & Maternia, M. (2016). 5G-PPP use cases and Proceedings of the IEEE International Conference on Neural
performance evaluation modeling. 5G PPP white paper. Networks, Citeseer, 4, 1942–1948.
30. Campolo, C., Molinaro, A., Iera, A., & Menichella, F. (2017). 5G 48. Fontes, R. R., Afzal, S., Brito, S. H., Santos, M. A., & Rothen-
network slicing for vehicle-to-everything services. IEEE Wireless berg, C. E. (2015). Mininet-wifi: Emulating software-defined
Communications, 24(6), 38–45. wireless networks. In Proceeding of the 11th international con-
31. Campolo, C., Molinaro, A., Iera, A., Fontes, R. R., & Rothenberg, ference on network and service management (CNSM) (pp.
C. E. (2018). Towards 5G network slicing for the v2x ecosystem. 384–389).
In Proceedings of the 4th IEEE conference on network soft- 49. Tomonori, F. (2013). Introduction to RYU SDN framework.
warization and workshops (NetSoft) (pp. 400–405). Open Networking Summit.
32. Šeremet, I., & Čaušević, S. (2019). Benefits of using 5G network 50. Behrisch, M., Bieker, L., Erdmann, J., & Krajzewicz, D. (2011).
slicing to implement vehicle-to-everything (V2X) technology. In Sumo–simulation of urban mobility: An overview. In: Proceed-
Proceedings of the 18th International Symposium INFOTEH- ings of the 3rd international conference on advances in system
JAHORINA (INFOTEH) (pp. 1–6). simulation.
33. Khan, H., Luoto, P., Bennis, M., & Latva-aho, M. (2018). On the 51. Han, Gh, Chen Xr, Yu Y, & Yq, Li. (2012). A study of micro-
application of network slicing for 5G-V2X. In European Wireless scopic traffic simulation based on sumo platform. Computer
2018; 24th European Wireless Conference, VDE (pp. 1–6). Engineering and Science, 34(7), 195–198.
34. Ersue, M. (2013). ETSI NFV management and orchestration-an 52. Jain, R. K., Chiu, D. M. W., & Hawe, W. R. (1984). A quanti-
overview. In Proceedings of 88th IETF meeting. tative measure of fairness and discrimination. Hudson, MA:
35. Group, I. W., et al. (2010). IEEE standard for information tech- Eastern Research Laboratory, Digital Equipment Corporation.
nology–telecommunications and information exchange between 53. Koza, J. R. (1997). Genetic programming.
systems–local and metropolitan area networks–specific require-
ments–part 11: Wireless LAN medium access control (MAC) and Publisher’s Note Springer Nature remains neutral with regard to
physical layer (PHY) specifications amendment 6: Wireless jurisdictional claims in published maps and institutional affiliations.
access in vehicular environments. IEEE Std 802(11).
36. Molina-Masegosa, R., & Gozalvez, J. (2017). LTE-V for sidelink
5G V2X vehicular communications: A new 5G technology for
short-range vehicle-to-everything communications. IEEE Vehic-
ular Technology Magazine, 12(4), 30–39.
37. Nadia, M., Rola, N., & Samir, T. (2019). SDN based fast and soft
handover (FSH) for a shuttle bus monitoring system. In: Pro-
ceedings of the international symposium on computers and
communications (ISCC).
38. Mouawad, N., Naja, R., & Tohme, S. (2019). Inter-slice mobility
management solution in v2x environment. In: Proceedings of the
15th international conference on wireless and mobile, networking
and communications (to appear).
39. Gibbons, R. (1992). A primer in game theory. London: Harvester
Wheatsheaf.
40. Li, Y., Liao, C., Wang, Y., & Wang, C. (2015). Energy-efficient
optimal relay selection in cooperative cellular networks based on
123
Wireless Networks (2020) 26:3883–3903 3903
Nadia Mouawad earned a Ph.D. Springer. Actually, she is associate professor at Lebanese University.
in Telecommunications at She is as well researcher at Li-parad Laboratory at University of
University of Versailles and Versailles.
Lebanese University. Her
research studies lie in the area Samir Tohme was born in
of mobility management and Damascus, Syria, in 1949. He
Quality of Service provisioning graduated from the
for 5G vehicular networks. She EcoleSupérieured’Electricité
received her bachelor in (1975), Paris, in Electrical
Telecommunication and Com- Engineering, and received the
puter Engineering at the Faculty Doctorat Degree from the
of Engineering II , Lebanese EcoleNationaleSupérieure des
University, Lebanon and a Télécommunications ENST,
Masters in Technology of Paris, in 1979. After having
Information and Communica- been with HP and SITA for five
tion Systems at the Doctoral School of Science and Technology, years where he worked in the
Lebanese University, Lebanon. distributed system field, he
joined the EcoleNa-
Rola Naja earned a Ph.D. in tionaleSupérieure des Télécom-
Computer Networking at Tele- munications in 1984 where he was appointed Head of the Networks
comParitech-Paris and a HDR at Department between 1988 and 1998. He joined the University of
University of Versailles. Her Versailles in 2003 where he was Director of PRISM Laboratory
research studies lie in the area between 2003 and 2013. His is presently in leave within VEDECOM
of resource allocation and Institute since January 2015 where he is in charge of the C2X
mobility handling in wireless Telecommunication Activities. Author of one hundred and fifty
multiservice networks and published research papers, his main research activities concern mobile
intelligent transportation sys- networks, satellite networks, C2X communications, high speed net-
tems (ITS). She has published works and traffic control.Professor Tohmé is an IEEE senior member.
several papers in International
journals and conferences deal-
ing with network modeling and
performance, network adapt-
ability and quality of service
provisioning. She has edited and co-authored a book in ITS with
123