VIMENO: A Virtual Wireless Mesh Network Architecture for Operators
Katarzyna Kosek-Szott, Janusz Gozdecki, Krzysztof Loziak, Marek Natkaniec,
Szymon Szott and Michal Wagrowski
AGH University of Science and Technology, Faculty of Computer Science, Electronics and Telecommunications,
al. Mickiewicza 30, 30-059 Krakow, Poland
Keywords:
Wireless Mesh Networks, Virtualization, Virtual Network Operators, Resource Separation.
Abstract:
Network virtualization is one of the key concepts of the Future Internet. However, the use of virtualization
techniques in wireless mesh networks has thus far not been adequately studied. In this paper we describe
VIMENO: a Virtual wireless Mesh Network architecture for Operators, i.e., the rst virtualization-enabled
wireless mesh network architecture designed for operators. In the proposed architecture, virtualization is used
to divide the mesh resources governed by a single mesh network operator among multiple virtual network
operators. The advantage of this outsourcing approach is that each virtual network operator can focus on
providing a service to its end-users, leaving the network administration to the mesh network operator. This
leads to the lowering of the operating expenditure of virtual network operators and allows tailoring of each
virtual network to individual operator and service requirements.
1 INTRODUCTION
Network virtualization is one of the key concepts
of the Future Internet. It enables decoupling
the provided services from the underlying infras-
tructure (Chowdhury and Boutaba, 2010). There
have been multiple research efforts devoted to this
topic, mostly with respect to fixed core networks
(4WARD, http://www.4ward-project.eu/; AKARI,
http://akari-project.nict.go.jp/;ANA, http://www.ana-
project.org/; GENI, http://www.geni.net/). In the do-
main of wireless local area networks, virtualization
has been implemented at the interface level, allow-
ing one WiFi access point (AP) to control multiple
networks (Banchs et al., 2012). However, the use of
virtualization techniques in wireless mesh networks
(WMNs) has thus far not been adequately studied
even though, as we will show, it can introduce new
business scenarios for network operators.
WMNs are known for their ability to extend the
reach of existing networks (Azcorra et al., 2009). In
traditional concepts such a mesh cloud is governed by
a single mesh network operator (MNO). We propose
a business scenario in which network virtualization is
used to divide the mesh resources among multiple vir-
tual network operators (VNOs) (Figure 1). Each VNO
has dedicated resources in the form of a Virtual Net-
work (VN). The VN extends from traffic entry points
(either WiFi APs or, in general, Access Routers, ARs)
to gateways (GWs) leading to the Internet core net-
work. Each VN has statistically guaranteed QoS pa-
rameters. The advantage of this outsourcing approach
is that each VNO can focus on providing a service to
its end-users, leaving the network administration to
the MNO. This leads to the lowering of the operating
expenditure of VNOs. Furthermore, each VN can be
tailored to both operator and service requirements.
In this paper we describe VIMENO: a Virtual
wireless Mesh Network architecture for Operators.
This architecture provides the aforementioned virtu-
alization features. Our approach is based on WiFi de-
vices (IEEE 802.11, 2012), which are known for their
large popularity, low cost, license-free operation, high
extendibility, and constant improvement through new
IEEE 802.11 amendments. Adopting an approach
based on statistical packet multiplexing is in contrast
with existing solutions, which are based on circuit
switching (most commonly TDMA). However, our
approach ensures greater flexibility which is impor-
tant when dealing with both bursty traffic as well as
varying radio channel conditions. Furthermore, our
approach is the first virtualization-enabled WMN ar-
chitecture designed for operators: previous efforts fo-
cused either on networks for research (Shrestha et al.,
2008) or on providing community network access
(Matos et al., 2011) or were proposed for ad hoc net-
works (Dedecker et al., 2011).
The architecture described in this paper is an ini-
207
Kosek-Szott K., Gozdecki J., Loziak K., Natkaniec M., Szott S. and Wagrowski M..
VIMENO: A Virtual Wireless Mesh Network Architecture for Operators.
DOI: 10.5220/0004535402070214
In Proceedings of the 10th International Conference on Signal Processing and Multimedia Applications and 10th International Conference on Wireless
Information Networks and Systems (WINSYS-2013), pages 207-214
ISBN: 978-989-8565-74-7
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
tial study of the concept of virtual WMNs for oper-
ators. We would like to share this concept with the
reserach community to stimulate discussion on the
numerous research challenges which we have identi-
fied. We plan to implement VIMENO to further study
these issues and provide suitable solutions.
The remainder of this paper is organized as fol-
lows. In Section 2 we discuss the business aspects of
our approach. In Section 3, we provide an overview
of the node types forming the network architecture
as well as the relevant networking modules. In Sec-
tions 4 to 7 we discuss the most important features of
VIMENO: network monitoring, resource separation,
traffic shaping, and routing, respectively. The conclu-
sions and future work are summarized in Section 8.
Set of APs
Mesh Cloud
Internet
1
n
Figure 1: Concept of the proposal.
2 CONTRACT
The network architecture and its functionality is de-
termined by the contract agreed upon between the
MNO and its customers, the VNOs. The main task
of the MNO is to ensure the contract execution for all
VNOs and in the case of temporary resource scarcity
(due to, e.g., congestions, degradation of propaga-
tion conditions, or failures) perform appropriate re-
configurations to provide fair, or in accordance with
an agreed policy, resource distribution. Moreover, we
define a method for flexible and dynamicresource dis-
tribution according to generated traffic as a part of the
VIMENO architecture’sself-organisingfunctionality.
Multiple detailed schemes for the contract can be
considered, however, for the purpose of our inves-
tigations we have decided to define a flexible one
to provide smart resource distribution within VNs
and create a solid base for further developments.
Thus, the VNO buys uplink and downlink through-
put (R
UL
, R
DL
) and specifies a set of Mesh Access
Nodes {MAN
n
}, where n (0, N
MAN
] and N
MAN
is
the total number of MANs in the network, on which
this VN will be available. The throughput guaran-
tee is statistical with a given rate probability function,
which is in line with current mobile network opera-
tor business models for services based on packet data
transmission. It is up to the network planning process
and in particular its dimensioning phase to ensure that
there are, statistically, enough resources to meet the
contract demands. We can determine the relation be-
tween contract commitments and the network capac-
ity as follows:
N
VNO
i=1
R
UL
i
C
UL
(1+ ε), (1)
N
VNO
i=1
R
DL
i
C
DL
(1+ ε), (2)
where R
UL
i
and R
DL
i
is the contracted throughputin the
uplink and downlink direction, respectively, for the i-
th operator; N
VNO
is the number of VNOs, C
UL
and
C
DL
are the uplink and downlink network capacity,
respectively; and ε is the capacity under-provisioning
factor. These equations reflect our assumption that
typically the total load of the WMN will be smaller
than its capacity, and, therefore, we can assume sta-
tistical resource sharing.
Based on the signed contracts, the MNO is re-
sponsible for setting up VNs and providing connec-
tivity within the mesh network. The access part of the
network is separated between VNOs by having each
AP located at a MAN broadcast individual SSIDs for
each VN. VNs are also treated independently by the
remaining physical mesh infrastructure. Thus, mul-
tiple routing protocol instances must be involved to
define transmission paths. The physical network ar-
chitecture is a mesh, however all paths are established
between MANs and Mesh Gateways (MGW) to pro-
vide Internet access, which results in the functional
architecture of multiple trees. Periodically, or in re-
sponse to radio link quality degradation or failures in
the network, the paths can be modified by the rout-
ing protocol. Hence, some paths can also be redi-
rected to other MGWs. Such an operation may result
in load variation at the MGW and other intermediate
mesh nodes. Moreover,the load can be distributedun-
equally among VNs, since VNOs define independent
sets of MANs and independent throughput contracts.
”If there is a risk of exceeding the maximum capac-
ity of any link of an MGW (or any other mesh node),
the throughput values currently configured for partic-
ular VNOs must be fairly decreased at that node. This
means that each mesh node must know the proportion
of particular VN load requirements that are to be met.
This can be reported by the MANs or MGWs cur-
rently attached to the mesh node or by a central ded-
icated management entity that is to serve as a central
resource distribution coordination point. These issues
are further discussed in Sections 5 and 6.
In order to simplify the description, we assume
that all the connections in the mesh network havebeen
WINSYS2013-InternationalConferenceonWirelessInformationNetworksandSystems
208
configured. We therefore bypass the initial bootstrap-
ping stage. Additionally, we exclude mobility issues
(i.e., the movement of VNO clients between MANs)
from our field of study. Finally, we assume that all
traffic flows initiate or terminate, from the mesh net-
work perspective, in the MANs and MGWs.
3 NETWORK
AND NODE ARCHITECTURE
Figure 2 presents a detailed view of the network ar-
chitecture in VIMENO, highlighting the data paths in
a given example topology. In the following, we give
a brief overview of the elements presented in this fig-
ure. There are three mesh node types in the network:
(i) Mesh Gateway (MGW) interconnects the mesh
and the core networks using either wireless or fixed
links, (ii) Mesh Access Node (MAN) offers net-
work access to end users, typically equipped with at
least two wireless interfaces: one for providing net-
work access and the other for connecting to the mesh
network, (iii) Mesh Relay Node (MRN) — intercon-
nects MANs with MGWs and relays user traffic, typ-
ically equipped with at least two wireless interfaces:
one in the uplink and one in the downlink data transfer
direction, respectively. In practice, to achieve better
mesh network performance, it is recommended that
each MRN have at least three wireless interfaces.
Each mesh node (Figure 3) contains the follow-
ing building blocks: (i) wireless interfaces — typical
IEEE 802.11 interfaces, (ii) virtual wireless inter-
faces to support the various VNs transported in the
mesh network, (ii) schedulers — to properly plan the
transmission from each virtual interface, (iv) traffic
shapers to limit the amount of traffic transported
in the network, they are present only in the ingress
nodes: MANs and MGWs for the uplink and down-
link directions, respectively, (v) monitoring module
to measure and estimate current PHY and MAC
parameters, (vi) routing module to orchestrate
traffic forwarding within the network, a separate in-
stance is executed for each VN.
In principle, the network operates as follows.
Traffic from VNs is separated using virtual interfaces
and appropriately configured schedulers (Section 5).
The traffic shapers are configured in such a way as to
fulfil the contract and minimise the traffic transported
in the mesh network (Section 6). Each VN employs a
separate routing protocol (Section 7). All these mech-
anisms are supported by passive network monitoring
(Section 4). In the following sections, we describe
each mechanism in detail.
4 MONITORING
Since the parameters of wireless links cannot be de-
termined in advance, they need to be estimated af-
ter the setup of wireless links. Additionally, during
network operation, the parameters can change due to
fluctuations of the radio propagation conditions or in-
terference from other networks. Typical operating
system tools do not provide reliable estimates on the
available and used data rates on the wireless links.
The only possibility to obtain such information is to
use a dedicated monitoring mechanism. In the VI-
MENO architecture we define the Monitoring Mod-
ule (MM), which provides a passive monitoring ser-
vice able to measure several parameters related to ra-
dio channel conditions, capabilities of neighbouring
nodes and IEEE 802.11 MAC parameters estimation.
MM provides precise network measurements without
disrupting network operation.
The results of MM measurements are utilized
mainly by the routing module, shapers, and sched-
ulers. During normal network operation the most im-
portant parameters that have to be delivered by the
MM for each link are: the available data rate, the data
rate used by each VN, delay, jitter, and packet loss ra-
tio. The MM module performs measurements at the
PHY and MAC layers within the time-scale of mi-
croseconds, based on all types of 802.11 frames (data,
management, and control). To ensure a comprehen-
sive view of the current wireless channel conditions,
including traffic from networks of other operators, the
promiscuous mode of operation of wireless network
cards is used. The use of passive mode means that all
measurements are performed along with the normal
activity of the wireless card.
The MM works on a frame level, i.e., all frames
sent and received by each network interface must be
examined by the MM functions. This imposes high
requirements on the implementation effectiveness of
the frame analysis (i.e., the limited computational
power available at the nodes should be taken into
account). In the Linux operating system the above-
mentioned requirements force an implementation of
MM as a kernel module closely interworking with the
mac80211 wireless framework (mac80211, 2013) .
5 RESOURCE SEPARATION
In VIMENO, network virtualization is realized dif-
ferently in the access and mesh parts of the network.
This is illustrated in Figure 4. The differences be-
tween the two virtualization methods (link and inter-
face) are subtle. We discuss them below.
VIMENO:AVirtualWirelessMeshNetworkArchitectureforOperators
209
C
H
4
C
H
3
C
H
2
C
H
1
Internet
V
i
rt
u
a
l
L
i
n
k
V
P
1
V
i
r
t
u
a
l
L
i
n
k
V
P
n
V
i
r
t
u
a
l
L
i
n
k
V
P
1
V
i
r
t
u
a
l
L
i
n
k
V
P
n
V
i
r
t
u
a
l
L
i
n
k
V
P
1
V
i
r
t
u
a
l
L
i
n
k
V
P
n
V
i
r
t
u
a
l
L
i
n
k
V
P
1
V
i
r
t
u
a
l
L
i
n
k
V
P
n
Access Network Mesh Network Core Network
VI
1
VI
1
VI
n
Mesh Relay Node
VI
n
VI
1
VI
1
VI
n
Mesh Relay Node
VI
n
SSID
1
SSID
n
STA
1
STA
n
VI
1
VI
n
TS
1,1
TS
n,n
VI
1,1
VI
1,n
VI
n,1
Vi
n,n
Access
Point
Mesh Access
Node
TS
n,1
TS
1,n
VI
n
VI
1
TS
n,n
TS
1,1
VI
n,n
VI
n,1
VI
1,n
VI
1,1
Mesh Gateway
TS
1,n
TS
n,1
Figure 2: Network architecture with basic elements.
VP
2
VP
1
VP
2
VP
1
Mesh Node
Monitoring Module
Traffic
Shapers
Virtual
Interfaces
Schedulers
Physical
Interfaces
TS
TS TS
TS
VI
VI VI
VI
Routing
Module
Figure 3: Architecture of the mesh node.
Access Point
Access Point
Link Virtualization
Interface Virtualization
SSID
1
SSID
2
STA1
STA2
VI
1
VI
2
Access Point
Virtual
Network 1
Virtual
Network 2
Mesh Cloud
Access RouterAccess Router
Internet Edge Router
CH
2
CH
1
Virtual Link 1
Virtual Link 2
VI
1
VI
2
Mesh Node
VI
1
VI
2
Virtual Link 1
Virtual Link 2
VI
1
VI
1
VI
2
Mesh Node
VI
2
VI
1
VI
1
VI
2
Mesh Node
VI
2
Figure 4: Separation between virtual networks.
In the access part of VIMENO, the VNs are sep-
arated using standard interface virtualization. There-
fore, the wireless interface of each AP (at the MAN)
is virtualized and a separate virtual interface (VI) is
provided for each VNO. This allows separation be-
tween VNOs by setting a different SSID for each of
them. This type of interface virtualization has been
described in numerous papers and, therefore, we re-
fer the reader to (Chandra and Bahl, 2004; Smith
et al., 2007; Rivera and Zucci, 2010; Al-Hazmi and
de Meer, 2011; Braham and Pujolle, 2011) for fur-
ther details. Currently, we assume standard 802.11
fairness in scheduling between VIs. However, in the
future we foresee remote configuration of the client
node MAC parameters by the AP.
In the mesh part of VIMENO, the VNs are sep-
arated using link virtualization. Therefore, a sepa-
rate virtual link is provided for each VNO on a sin-
gle wireless channel
1
. The separation of traffic (i.e.,
packet flows) between VNs is based on statistical
packet scheduling (described below). Such a solution
is cheap, scalable, flexible, does not require synchro-
nization, and finally, it is in line with the packet nature
of the Internet.
Mesh Node #1
VI
1
VI
2
1,1,0
T
1,2,0
T
Mesh Node #3
B
1,1
Mesh Node #2
3,1,2
T
3,2,2
T
2,1,0
T
B
1,2
B
2,1
VI
1
VI
2
1,1,1
T
1,2,1
T
2,1,1
T
2,2,1
T
Stage 1 Stage 2
Traffic flow (upstream)
2,2,0
T
TS
1,1
TS
2,1
TS
1,2
TS
2,2
VI
1
VI
2
Incoming
Traffic
Figure 5: Generalized intra-mesh scheduling assuming two
VNOs.
The generalized intra-mesh scheduling method is
illustrated in Figure 5. In principle, the goal of the
scheduler’s operation is, in the absence of sufficient
resources, to fairly divide the available link capac-
ity among the VNs. This is performed in proportion
to current traffic shaper settings. At the first stage
(i = 1), i.e., at the MAN or MGW, the scheduler is
configured so that the outgoing traffic rate (in b/s) of
each VI j of the k-th mesh node is set as follows
T
1, j,k
= min
TS
j,k
N
VNO
h=1
TS
h,k
B
1, j
, T
0, j,k
!
, (3)
where TS
j,k
is the setting of the j-th traffic shaper of
the k-th mesh node, N
VNO
is the number of VNOs,
and B
1, j
is the capacity of the outgoing link of the k-
th mesh node. This means, that under non-saturation
T
1, j,k
= T
0, j,k
(i.e., the current incoming traffic rates
correspondto the outgoing trafficrates). Additionally,
under saturation T
1, j,k
=
TS
j,k
N
VNO
h=1
TS
h,k
B
1, j
, which means
1
A channel allocation scheme is out of the scope of this
paper. We refer the reader to (Lv et al., 2012) for an exem-
plary OFDM-based scheme.
WINSYS2013-InternationalConferenceonWirelessInformationNetworksandSystems
210
that the outgoing traffic rate is scaled according to the
capacity of the outgoing link of the k-th mesh node
and the current settings of appropriate traffic shapers.
At stages i > 1 the outgoing traffic rate from each
VI j of the k-th mesh node is dependent on the rate
configurations of the traffic shapers located at the up-
stream ingress nodes:
T
i+1, j,k
=
min
lK
k
TS
j,l
N
VNO
h=1
lK
k
TS
h,l
B
i+1, j
,
lK
k
T
i, j,l
!
,(4)
where K
k
is the set of upstream one-hop neighbors
of k. Under non-saturation all traffic arriving to
the k-th mesh node is transmitted to the next stage
T
i+1, j,k
=
lK
k
T
i, j,l
. Under saturation, the outgo-
ing traffic rate from the k-th mesh node is scaled ac-
cording to the value of the capacity of the outgoing
link
T
i+1, j,k
=
lK
k
TS
i, j,l
N
VNO
h=1
lK
k
TS
i,h,l
B
i+1, j
. Such an ap-
proach requires the dissemination of current TS rate
configurations, but assures fairness in relation to both
the contracts as well as the current VNO traffic re-
quirements.
In order to discriminate between VNs VLAN tags
are employed (IEEE 802.1Q, 2011). They are placed
after the IEEE 802.11 header in each frame. Such a
solution is flexible, backward compatible, and allows
straightforward implementation in wireless devices.
6 TRAFFIC SHAPING
In the VIMENO architecture it is necessary to as-
sure that the throughput contracts are not breached,
i.e., that the traffic admitted into the mesh network
for each VN does not exceed the VNOs contracted
throughput values. This is achieved through the ap-
propriate configuration of traffic shapers, which are
placed at the ingress nodes. At the MAN (MGW),
they guarantee that the uplink(downlink) contracted
throughput is within limits.
To ensure that contracts are fulfilled and to pro-
vide flexible resource distribution, dedicated sig-
nalling is required to periodically inform traffic
shapers about current data rates for VNs. This can
be performed using a centralized or distributed ap-
proach. The final solution for the signalling archi-
tecture should be selected according to the required
functionality, efficiency, and scalability of the net-
work. For the purpose of VIMENO deployment we
assume a centralized management approach. A ded-
icated server is located on the outside of the consid-
ered mesh network (physically, it may be at one of
the MGWs or in the core network). As future work,
we will consider a distributed management approach
which is required for large mesh networks.
We assume that the contracted throughput values
will be fairly divided among the MANs and MGWs
for each VNO. This requires an appropriate configu-
ration of shapers located at these nodes. This config-
uration is performed using the following algorithm.
First, the central server sends updates to the traf-
fic shapers located at the MANs and MGWs with
their initial configuration. These updates contain the
throughput values for the uplink and downlink direc-
tions, as contracted by each VNO, divided by the se-
lected number of MANs (for uplink) and MGWs (for
downlink). Next, the obtained throughput values are
set as the shaping rates for each VN at both MANs
and MGWs. This means that for a given VN, the
shaping rates are equal for each selected MAN as well
as for each selected MGW. In order to effectively use
the available resources in the mesh network, for each
VN the shapers are independently controlled using the
AIMD (Additive Increase, Multiplicative Decrease)
algorithm (Crisostomo et al., 2005). This algorithm
uses information about the resources available at the
MANs and MGWs as input to provide a configura-
tion which assures proportional sharing of resources
between VNs as feedback.
After the initial configuration of shapers, if there
is both on-going traffic as well as available resources,
the shaping rates can be reconfigured in the follow-
ing way. Every time interval t, the rate of each shaper
is independently increased for each VN by an incre-
ment of c kb/s until a given threshold for each MAN
or MGW is exceeded (this threshold is set to 95% of
the available resources to ensure that the network op-
erates in non-saturation). The sum of the configured
shaping rates at MANs and MGWs for each VN can-
not be larger than the contracted throughput values
(either for the uplink or downlink direction). It should
also be noted, that when the shaping rate substantially
exceeds the actual rate, there is the risk of transmitting
data bursts without due control, which may affect link
utilization. In order to avoid this problem, the rate
controller located at the same entity monitors the cur-
rent transmission, and regulates the shaping rate in
order to not exceed the actual rate in more than a gap
percent of the actual rate. The AIMD shaping algo-
rithm is given in Algorithm 1 in the form of pseudo
code.
VIMENO:AVirtualWirelessMeshNetworkArchitectureforOperators
211
Algorithm 1:
1: receive updated parameters r
i
, c
i
, a
i
, g
i
, w
i
, z
i
2: if (u > 0.95)
s
i
s
i
× (1 r
i
)
3: else
4: if (w
i
< z
i
)
s
i
s
i
+ c
i
5: if ((s
i
a
i
) > a
i
× g
i
)
s
i
a
i
× (1+ g
i
)
6: return;
where: u is the link utilization; s
i
is the shaping rate of
shaper i for the i-th VN; w
i
is the sum of actually con-
figured upstream shaping rates for the i-th VN; z
i
is
the sum of the contracted upstream throughput values
for the i-th VN; r
i
is the multiplicative decrease factor
for the i-th VN; c
i
is the additive increase increment
for the i-th VN; a
i
is the actual outgoing traffic rate of
the shaper for the i-th VN; g
i
is the maximum gap of
r
i
concerning a
i
for the i-th VN.
Moreover, if there is no traffic in the VN at spe-
cific MAN or MGW, the shaping rate at MAN and/or
MGW for this VN can be decreased down to the mini-
mum defined throughput value m
i
using the following
rule: every time interval t, the rate of each shaper is
decreased by a multiplicative factor r
i
.
It could also happen that a new traffic flow appears
in a VN while the shaping rate for this VN was de-
creased at specific MAN and/or MGW to the level of
m
i
. If this is the case, a reconfiguration procedure
should be executed. First, the server checks the actual
value of w
i
. If w
i
< z
i
and there are enough resources
to increase the shaping rate at specific MAN and/or
MGW, it is increased using Algorithm 1. In any other
case, to prevent saturation conditions and assure pro-
portional and fair distribution of resources for each
VN, the server sends to all MANs and MGWs a re-
quest to proportionally decrease the shaping rate for
all VNs by a multiplicative factor r
i
. This procedure
is repeated until there are enough resources to start
the process of the shaper reconfiguration. Second,
the shaping rate at newly used MAN and/or MGW
for this VN is increased proportionally to the values
of throughput configured at all other MANs and/or
MGWs of the same VN (proportionally to the con-
tracted throughput values of each VN) using AIMD.
To summarize, the server plays an important role
in the process of guaranteeing VN throughput. It
collects information about the contracted upstream
throughput values, existence of VN traffic at each
MAN and MGW, available resources at MANs and
MGWs, and the actual rate of shapers for each VN.
This allows for efficient and fair distribution of the
available mesh network resources between the VNs.
7 ROUTING
Routing plays a leading role in the data exchange
within WMNs, especially in terms of service avail-
ability and reliability. For a large number of MRNs
in a WMN topology, the role of the routing protocol
is even more important, because it has to deal with
the complexity of a multi-path topology. Additionaly,
taking into account the dynamics caused by the time
varying wireless links, it can be concluded that only
a dynamic, link-state routing protocol, supported by a
constant network monitoring, is capable of providing
a reliable service for VNOs.
The most important objectives of a dynamic rout-
ing protocol in WMNs are: (i) remote network
discovery and information sharing among network
nodes, (ii) determination of the best path towards the
destination (between MANs and MGWs), (iii) main-
tenance of routing information and routing table up-
dated upon topology or link state change, and (iv)
ability of traffic rerouting along a new path in case
of transmission condition changes.
The most essential component of each rout-
ing protocol is the routing protocol metric, since
it drives the best path selection procedures. Cur-
rently, a variety of different metrics for WMNs ex-
ist. They are mainly adapted from mobile ad-hoc
networking. The most typical metrics proposed
for WMNs are the following (Bin Ngadi et al.,
2012): ETX (EXpected Transmission Count) with
its modifications, ETT(Expected Transmission Time),
WCETT (Weighted Cumulative ETT), RARE (Re-
source Aware Routing for MESH), EDR (Expected
Data Rate), CATT (Contention Aware Transmission
Time), and ACRM (Airtime Cost Routing Metric, the
default metric for the IEEE 802.11s standard).
Unfortunately, none of these typical metrics could
be adopted without change to provide a complete
radio link state description, taking into account the
virtualization concept, as this requires a more de-
tailed image of a variety of cross-layer parameters
(L1/L2/L3). Such a cross-layer approach allows to
cover not only the radio channel state (number of
transmitting nodes or estimated free bandwidth), but
extends it with a specific link parameters (delay, aver-
age throughput, etc.).
Since a simple mapping of the existing wireless
routing protocol metrics into virtualized WMNs is not
sufficient, a question arises: what kind of metric could
satisfy the requirements of WMN reliability with re-
spect to the demands of VNOs? It seems to be evident
that only a cross-layer approach together with a com-
posite metric design addresses such a problem.
The commonly known example of a composite
WINSYS2013-InternationalConferenceonWirelessInformationNetworksandSystems
212
metric is the one used by the EIGRP routing proto-
col (EIGRP, 2005). It is formulated as follows:
M =
K
1
× B+
K
2
×B
256L
+ K
3
× D
K
5
R+ K
4
, (5)
where M metric, B bandwidth (in bits per sec-
ond), L load, D delay, R reliability. Addition-
ally, K
1
to K
5
are the cumulative weights. By default
K
1
and K
3
are set to 1, while the others are set to 0.
The EIGRP protocol uses the minimum bandwidth on
the path towards the destination network and the total
delay to calculate its routing metric. The load compo-
nent reflects link saturation, while the reliability com-
ponent represents the measure of the likelihood that
a link will fail. A detailed description of the EIGRP
metric can be found in (EIGRP, 2005).
In case of time varying wireless links, the for-
mula of a composite metric is not so evident as for
EIGRP, since we cannot rely on the stability of the
bandwidth and delay values.
2
Therefore, the pro-
posed VIMENO composite metric should be based on
components selected from the following four generic
groups: (i) Basic: hop count, estimated maximum
available bandwidth; (ii) Quality of Service: link
capacity, average transmission rate, aggregated per-
hop and overall delay, jitter, BER; (iii) Interference:
number of nodes operating on the same channel, SNR
ratio, RxPwr ratio; (iv) Route Stability: considered
as the radio link stability with respect to a rate uti-
lization histogram analysis and link availability in a
long timescale. The following parameters have been
selected as the most representative elements of the
VIMENO composite metric: (i) Bandwidth estima-
tions: ratio of the required, contractual, bandwidth or
average available throughput (per VNO) compared to
the estimated free channel bandwidth; (ii) Interfer-
ence: number of nodes transmitting on a single radio
channel; (iii) Route stability: multi-rate utilization
and link availability.
These parameters will be provided by the Moni-
toring Module operating on all the physical interfaces
of the mesh node. However, the exact definition of the
final VIMENO composite metric formula requires a
detailed mathematical and simulation analysis, which
is beyond the scope of this paper.
Importantly, the VIMENO architecture is able to
support traffic separation between VNOs. This re-
quires that at each physical radio link a dedicated vir-
tual link is assigned for a single VNO. The virtual link
is then established between a set of virtual interfaces
on the physical ones at neighbouring WMN nodes.
2
Originally, both parameters, bandwidth and delay, in
the EIGRP metric are assumed to be static.
The set of virtual links composes a VN topology for a
VNO. To assure end-to-end connectivity, a dedicated
link-state routing protocol is required to operate sep-
arately for each VN topology.
The main goal of the routing protocol is to provide
optimal path selection between MANs and MGWs
and activate a rerouting procedure in case of degra-
dation of radio conditions. The rerouting process as-
sumes that multiple paths exist between MANs and
MGWs. To satisfy this assumption, during the VNO
network setup phase, the virtual interfaces dedicated
to a specific VNO should be configured. Additionally,
virtual links should be set up over all physical ones,
so as not to limit the possibilities of alternative path
selection.
Due to the limitation of the radio resources, we
cannot assume resource over-provisioning at a simi-
lar level as it is possible in fixed networks, but some
over-provisioning may need to be considered to avoid
link congestion. In the case of virtualized WMNs
link congestion may be caused by synchronization of
rerouting decisions within the same physical node for
different VNOs. To avoid this situation, the value of
the composite metric should be calculated separately
for each VN and the metric computation algorithm
should avoid synchronization of the rerouting pro-
cess. As a result, a slight time shift in the signalling of
radio link parameter degradation to each VNO routing
protocol is considered. This helps to avoid the prob-
lem of path flipping between interfaces at the same
physical node for all existing VNOs at the same time.
The selection of the VN which should be triggered to
activate the rerouting procedure may be based on the
VNO contract with respect to the required QoS pa-
rameters (delay, jitter, etc.), business-level parameters
(VNO priority), currently used bandwidth, etc.
At the first stage of the VIMENO architecture de-
velopment, for simplicity and taking into account that
multiple VNO networks exist in a single WMN node,
load balancing takes place at the VNO level. There-
fore, we do not consider multi-path routing inside
VNO networks. The analysis of multi-path routing
is planned as our future work.
8 CONCLUSIONS
AND FUTURE WORK
This article has presented an initial study of VI-
MENO, the first virtual wireless mesh network ar-
chitecture for operators. The novelties proposed in
the article include: definition of a new business sce-
nario in which network virtualization is used to divide
the mesh resources among multiple VNOs; distinc-
VIMENO:AVirtualWirelessMeshNetworkArchitectureforOperators
213
tion between interface and link virtualization for the
access and mesh network parts, respectively; archi-
tecture of a virtualized mesh node; virtualization of
the mesh network composed of intra-mesh scheduling
combined with the assignment of VLAN tags; shap-
ing technique based on the throughput metric, provid-
ing upstream and downstream guarantees for VNOs;
and link-sate routing protocol, composed of different
instances for each VNO. A promising advantage of
the proposed architecture (in contrary to the previous
TDMA-based solutions) is the fact that it is in line
with the packet nature of the Internet.
Our future work will be directed towards the im-
plementation of VIMENO in real wireless devices.
This will allow testing and optimizing the proposed
solutions. Future enhancement of VIMENO is also
planned in order to make the architecture even more
suited to various possible requirements of wireless
network operators and services. The following net-
working challenges will be considered: design of al-
ternative routing solutions supporting the virtualiza-
tion concept, definition of the formula for the rout-
ing composite metric, distributed network manage-
ment and traffic shaping, wireless network bootstrap-
ping, appropriate load balancing, network reconfigu-
ration and rerouting, network resilience and surviv-
ability, security of signalling data, and quality of ser-
vice (QoS) provisioning. They will be designed tak-
ing the perspective of VNOs into account. Obviously,
an introduction of new functionalities will require re-
thinking of the currently proposed solutions and pro-
tocols. E.g., the introduction of QoS provisioning will
impact the settings of traffic shapers and schedulers.
ACKNOWLEDGEMENTS
This work has been carried out as a part of a project
financed by the Polish National Science Centre (deci-
sion no. DEC-2011/01/D/ST7/05166)).
REFERENCES
Al-Hazmi, Y. and de Meer, H. (2011). Virtualization of
802.11 interfaces for wireless mesh networks. In Proc.
of WONS 2011.
Azcorra, A., Banniza, T., Chieng, D., Fitzpatrick, J., Von-
Hugo, D., Natkaniec, M., Robitzsch, S., and Zdarsky,
F. (2009). Supporting carrier grade services over wire-
less mesh networks: The approach of the European
FP-7 STREP CARMEN. Commun. Mag., 47(4):14–
16.
Banchs, A., Serrano, P., Patras, P., and Natkaniec, M.
(2012). Providing Throughput and Fairness Guaran-
tees in Virtualized WLANs Through Control Theory.
Mob. Netw. Appl., 17:435–446.
Bin Ngadi, M. A., Ali, S., Abdullah, A. H., and Khokhar,
R. H. (2012). A taxonomy of cross layer routing met-
rics for wireless mesh networks. Journal on Wireless
Communications and Networking, 2012(177).
Braham, O. and Pujolle, G. (2011). Virtual wireless network
urbanization. In Network of the Future (NOF), 2011
International Conference on the, pages 31–34. IEEE.
Chandra, R. and Bahl, P. (2004). MultiNet: Connecting to
multiple IEEE 802.11 networks using a single wire-
less card. In Proc. of INFOCOM 2004.
Chowdhury, N. and Boutaba, R. (2010). A survey of net-
work virtualization. Comp. Net., 54:862–876.
Crisostomo, S., Sargento, S., Natkaniec, M., and Vicari,
N. (2005). A QoS architecture integrating mobile ad-
hoc and infrastructure networks. In Proc. of AICCSA
2005.
Dedecker, P., Hoebeke, J., Moerman, I., Moreau, J., and
Demeester, P. (2011). Network virtualization as
an integrated solution for emergency communication.
Telecommunication Systems, pages 1–18.
EIGRP (2005). Enhanced Interior Gateway Routing Proto-
col.
http://www.cisco.com/en/US/tech/tk365/
technologies_white_paper09186a0080094cb7.
shtml
.
IEEE 802.11 (2012). IEEE Standard for Information
technology—Telecommunications and information
exchange between systems—Local and metropoli-
tan area networks—Specific requirements—Part 11:
Wireless LAN Medium Access Control (MAC) and
Physical Layer (PHY) Specifications, March 2012.
IEEE 802.1Q (2011). IEEE 802.1Q-2011 IEEE Stan-
dard for Local and metropolitan area networks–Media
Access Control (MAC) Bridges and Virtual Bridged
Local Area Networks.
Lv, P., Wang, X., and Xu, M. (2012). Virtual access net-
work embedding in wireless mesh networks. Ad Hoc
Networks, 10(7):1362–1378.
mac80211 (2013). Linux wireless (IEEE-802.11) subsys-
tem.
http://linuxwireless.org
.
Matos, R., Sargento, S., Hummel, K. A., Hess, A.,
Tutschku, K., and de Meer, H. (2011). Context-based
wireless mesh networks: a case for network virtual-
ization. Telecommunication Systems, 51:1–14.
Rivera, A. D. and Zucci, W. (2010). Virtualization of wire-
less network interfaces Wi-Fi IEEE 802.11. In Proc.
of TELE-INFO 2010.
Shrestha, S., Lee, J., and Chong, S. (2008). Virtualization
and slicing of wireless mesh network. In Proc. of CFI
2008.
Smith, G., Chaturvedi, A., Mishra, A., and Banerjee, S.
(2007). Wireless virtualization on commodity 802.11
hardware. In Proc. of ACM WiNTECH 2007.
WINSYS2013-InternationalConferenceonWirelessInformationNetworksandSystems
214