ADVCOMP 2014 : The Eighth International Conference on Advanced Engineering Computing and Applications in Sciences

Mobile Edge Computing: Challenges for

Future Virtual Network Embedding Algorithms

Michael Till Beck, Marco Maier

Ludwig-Maximilians-Universitat Munchen
Munchen, Germany
{michael.beck,marco.maier}@ifi.lmu.de
AbstractMobile edge computing aims at reducing network
latency and network stress by deploying mobile applications at
the network edge. This paper proposes network virtualization in
the context of mobile edge computing networks as an enabler
for future, flexible and shared network infrastructures (IaaS).
Network virtualization leads to the Virtual Network Embedding
(VNE) problem, which aims at deploying virtual networks onto
a shared, physical infrastructure. This paper is a position paper
discussing new challenges for future VNE algorithms. To this
end, new parameters specific to the mobile edge computing
scenario are analyzed which are not considered by state-of-theart VNE algorithms. Furthermore, novel and edge-specific VNE
optimization objectives are derived.
KeywordsVirtual Network Embedding, Next-gen Cellular Networks, Edge Computing

I.

I NTRODUCTION

Starting with the introduction of the first smartphones, one

of the most apparent challenges for mobile network operators
is handling future bandwidth demands, which are expected to
further increase dramatically over the next years [1].
Mobile data traffic is predicted to continue doubling each
year. Video contributes heavily to overall mobile network
traffic. The usage of mobile applications (apps) accessing services deployed in the Internet is expected to further contribute
to this trend, resulting in a growth of around 12 times by
2018 [2]. This trend becomes even more remarkable as novel
mobile devices (like Google Glass and other, wearable devices)
and applications arise. Offering more and more hardware
capabilities, these devices pave the way for novel application
types like augmented reality [2].
Network operators spend enormous efforts to keep up with
these demands in order to satisfy the needs of their users,
providing low-latency network access. Upgrading base stations
and core network routers to higher-capacity equipment reduces
high utilization in the core network, but comes with significant
operational cost. Furthermore, network operators are impelled
to quickly integrate upcoming technologies (like Long Term
Evolution, LTE) at the edge of their networks, offering better
quality of experience. Higher bandwidth capacities at the
network edge, however, directly affect core network utilization
and require further investments.
Two technologies have recently been proposed as a remedy
to this dilemma: 1) Mobile edge computing and 2) network
virtualization. Mobile edge computing aims at reducing latency
by shifting computational efforts from the Internet cloud to
the mobile edge. Network virtualization increases management
flexibility of the mobile infrastructure and enables resource
sharing between multiple providers.
1) Mobile edge computing (MEC) is an emerging technology that is seen as an alleviating factor in this context
[3], [4], [5]. MEC aims at reducing both network latency
and resource demands by shifting computing and storage
capacities from the Internet cloud to the mobile edge. Instead
Copyright (c) IARIA, 2014.

ISBN: 978-1-61208-354-4

of uploading or downloading content generated or demanded

by the mobile user, mobile applications refer to a service
located close to the current position of the user. Services are
hosted on devices directly attached to base stations or smart
cells (i.e., macrocells, microcells, or picocells). These hosts are
also known as MEC servers and are operated by the mobile
infrastructure provider. The proximity of the MEC servers to
the mobile device not only takes load from the mobile core
but also increases responsiveness of mobile applications.
2) Network virtualization is commonly seen as a key technology for the Future Internet and has recently been proposed
in the context of mobile networks [1], [6], [7], [8]. Network
Virtualization enables sharing of physical network resources
like base stations, core routers, and MEC servers between
multiple network operators. Network virtualization enables
operators to fully isolate their (virtual) resources from those
hosted by others on the same physical device (data and control
plane). Network sharing not only reduces cost for deployment
of new hardware resources but also operational cost [1][6][8].
Increasing network management flexibilities, virtualization of
network resources is also seen as a key technology to mitigate
the ossification of the core protocols.
This paper proposes a fully virtualized MEC infrastructure. Virtualizing both the mobile core and the mobile edge
network enables infrastructure providers to shared resources
between several mobile operators (Infrastructure as a service,
IaaS), including computing and storage capacities of the MEC
servers servers, and enhances management flexibilities. One
major challenge of network virtualization is the embedding of
virtualized resources onto the physical network. This problem
is known as the VNE problem. VNE algorithms aim to embed
multiple Virtual Network Requests (VNRs) onto a shared
substrate network, enabling virtual network operators to share
a common substrate infrastructure flawlessly by assigning
sufficient resources. While virtualization is a technique that is
well-known both in mobile networks and computer networks,
the VNE problem has not been discussed in the context of
MEC so far. Therefore, this paper addresses this shortcoming
and identifies new research directions for future, MEC-specific
VNE approaches.
The remainder of this paper is structured as follows: In
section II, network topologies of MEC networks are explained. Furthermore, the VNE problem is formalized and
the formalization is extended with respect to the mobile
edge scenario. Section III-A introduces new VNE parameters
and Section III-B introduces VNE optimization objectives for
MEC networks. In section III-C, challenges for future VNE
approaches are discussed. Section IV discusses related work
and section V concludes the paper.
II.

V IRTUALIZATION OF M OBILE E DGE

I NFRASTRUCTURES

This section discusses mobile edge computing, motivates

network virtualization in the context of mobile networks, and
65

ADVCOMP 2014 : The Eighth International Conference on Advanced Engineering Computing and Applications in Sciences

formulates the VNE problem in the context of mobile edge

networks.
A. Mobile Edge Computing
Mobile edge computing is an emerging concept becoming
more and more feasible with the shift towards the LTE
wireless communication standard. LTEs new core network,
System Architecture Evolution (SAE), is an all-IP network
with a simplified architecture, allowing for greater flexibility
of the networks topology and more heterogeneous access
networks, integrating legacy systems (e.g., air interfaces of
GPRS or UMTS) and LTEs new Evolved Universal Terrestrial Radio Access (E-UTRA). Due to LTEs low-latency and
high-bandwidth radio access networks, deployment of new
computing resources at the mobile edge becomes a promising
approach for supporting novel latency-sensitive applications.
MEC servers provide computing, storage and bandwidth
capacity that is shared by multiple virtual machines installed
on top of them. Fig. 1 depicts the mobile edge computing
scenario. MEC servers, being owned and managed by the infrastructure provider, are directly attached to the base stations.
Traditionally, all data traffic originating at the data centers is
forwarded by Internet routers to the mobile core network. The
traffic is routed through the core network to a base station
which delivers the content to the mobile devices. In the mobile
edge computing scenario, MEC servers take over some or
even all of the tasks originally performed in a data centers.
Being located at the mobile edge, this eliminated the need of
routing these data through the core network, leading to low
communication latency.
Two different, but related usage scenarios have been proposed in the context of mobile edge computing: The first one
proposes mobile devices to delegate calculations to the MEC
servers (offloading of resource- or power-intense tasks) [5],
while the second one proposes application service providers
(ASPs) to deploy services traditionally hosted within data
centers on the MEC servers (Edge Deployment) [3], [4].
Both aim at making the edge of the mobile network smarter,
leading to a reduction of core network utilization and decreased
latency:
1) Offloading Some applications running on a mobile device are capable of offloading resource- or power-intense tasks
to MEC servers. Therefore, the mobile application invokes
additional services deployed at a virtual machine hosted on a
MEC servers. MEC servers are placed nearby, offer excellent
Internet connectivity, and are easily reachable by the mobile
device: In the best case, there is a one-hop communication
between mobile device and host, offering low-latency access.
The concept is expected to increase limited computing, storage,
or bandwidth capacities of mobile devices by referring to external, resource-rich resources. Another objective of offloading
is to reduce power consumption of a mobile device.
If no MEC server is available, the mobile device degrades
gracefully to a more distant MEC server, a remote Internet
cloud server, or use its own hardware resources [5], [9].
2) Edge Deployment In the edge deployment scenario,
network providers offer MEC capacities for the deployment of
ASP-operated virtual machines running at the edge. ASPs offer
additional services at the network edge, increasing responsiveness of their applications. Services or parts of services traditionally hosted in data centers are now shifted to the network
edge. Since traffic between MEC servers and mobile devices
has not to be routed through the core network, this leads to
decreasing core network utilization and lower communication
latency.
Copyright (c) IARIA, 2014.

ISBN: 978-1-61208-354-4

B. Network Virtualization
Network virtualization has been proposed both in the
context of computer networks and for mobile core networks
[3], [4], [5]. This paper proposes the application of network
virtualization techniques for the whole network infrastructure,
including network core, base stations, and MEC servers.
Network virtualization is proposed as a key technology to
overcome the ossification of core protocols, since it enables
the deployment of several, isolated virtual networks on top of
a shared physical infrastructure. Virtual networks are co-hosted
on a common substrate infrastructure and, since they are fully
isolated, are even capable of deploying different communication protocols (e.g., IPv4/v6 or proprietary protocols) on the
same substrate links.
Infrastructure providers (InPs) offer physical network resources to several mobile network operators. Operators specify
network topologies and hardware resource demands to be
deployed within the infrastructure of the InP. Operators are
usually external customers of the InP. This does not exclude,
however, that InP itself can also deploy networks on its
own on top of its infrastructure, renting spare resources to
other operators. The InP provides its physical resources to the
operators, ensuring that all network requests of the operators
are fulfilled. In this paper, fully-virtualized MEC networks are
proposed. This means that both the network core, and also the
network edge, i.e., MEC servers and base stations resources
provide virtualization capabilities. Network virtualization is a
useful technique in order to separate several internal networks
and to increase manageability. Furthermore, it enables the InP
to rent spare resources to other operators.
One important aspect in this area of research lies in the
embedding of virtual network entities to the physical (or to be
more general: the substrate) infrastructure. This is commonly
known as the Virtual Network Embedding (VNE) problem
[10]. Physical resources are limited and have to be shared
between the virtual network entities that are assigned to these
resources. This is depicted in Fig. 2: two network requests are
assigned to a substrate network. The VNE problem is divided
in two sub-problems: Virtual Node Mapping and Virtual Link
Mapping: Virtual nodes are assigned to substrate nodes offering sufficient resources. Virtual links are either assigned to a
single substrate link, or span a path of multiple links in the
substrate network, where each link offers sufficient resources.
This is shown in Fig. 2 for the virtual link demanding
100MBit/s bandwidth capacity. The VNE problem becomes
N P-hard when substrate nodes and links have finite resources
[10].
C. Problem Formulation
In this subsection, a formal description for the general VNE
problem as depicted in Fig. 2 is presented. This formal model
will then be enhanced with respect to MEC specific properties.
A substrate network S = (N, L) is modeled as a set of
substrate nodes N and a set of links L mutually connecting
some of the nodes. Similar to the substrate network, a VNR
is modeled as a collection of virtual nodes N i and links
Li . Substrate nodes and links offer resources R, assigned by
cap : N L 2R . Virtual nodes and links demand these
resources, formally described as demi : N i Li 2R .
The objective of a VNE algorithm is to embed several Virtual
Network Requests VNRs, denoting VNRi = (N i , Li ) as being
the i-th request. Virtual entities that are embedded onto a
substrate entity consume substrate resources they demanded.
Therefore, the VNE has to assure that a sufficient amount of
resources is provided by a substrate entity before a virtual
entity gets assigned to it. The embedding is modeled as a
66

ADVCOMP 2014 : The Eighth International Conference on Advanced Engineering Computing and Applications in Sciences

Internet
Backbone

Mobile
Core Network

Data Centers

ASPs

Gateway

Mobile Users

Base Stations
with co-hosted
Mobile Edge Computing Servers

CDNs

Fig. 1: Mobile Edge Computing: Deployment of MEC servers at the Edge of the Mobile Network
Virtual Network Request 2

Virtual Network Request 1

bandwidth: 150MBit/s
bandwidth: 100MBit/s
CPU: 2 GHz

or

te

a
str
ub

bandwidth: 1GBit/s

tw
Ne

Fig. 2: Virtual Network Embedding

Virtual Network Request

SGW

PGW
Region 2

Region 1

PGW
SGW

Internet

SGW

SGW
PGW

Region 1
SGW

Ledge and Lcore for substrate edge/core links, and, respeci

i
tively, Nedge
, Ncore
, Liedge , and Licore for the virtual nodes
and edges.
Edge nodes are nodes located at the edge of the network,
i.e., nodes providing network access to the mobile devices
(representing base stations) or computing and storage capabilities (representing MEC servers). Core nodes are all other
nodes (network core devices), e.g., routers, Serving Gateways
(SGW) (connecting the base stations to the core network) or
Packet Data Network Gateways (PGW) (connecting the core
network to the internet). A main characteristic of edge nodes
representing base stations is their relatedness to a specific
geographic region. Fig. 3 depicts the embedding of a VNR.
In this case, the VNR demands base stations with affiliated
MEC servers in two specific regions (region 1 and 2), thus
not allowing to, e.g., only use the two base stations with
MEC capabilities in region 1. Geographic constraints of edge
nodes are one of the new challenges for VNE in the MEC
scenario, and are reflected in the proposed VNE parameters
and optimization objectives which are discussed in the next
section.
III.

Substrate Network
Region 2

Fig. 3: Embedding of Mobile Edge Computing Networks

function fi : N i N assigning nodes of VNRi to the

substrate network, and a function gi : Li SN0 SN
assigning edges of the VNR to the substrate links (or a
combination of links, i.e., paths). The notation used here is
in line with the one introduced in [10].
The VNE problem in the mobile edge scenario is depicted
in Fig. 3: For this scenario, a differentiation between edge and
core nodes is needed. Therefore, the following new variables
are introduced: Nedge and Ncore for substrate edge/core nodes,
Copyright (c) IARIA, 2014.

ISBN: 978-1-61208-354-4

E MBEDDING V IRTUAL N ETWORKS IN M OBILE E DGE

C OMPUTING S CENARIOS

This section introduces the VNE problem in the context

of mobile edge computing. To this end, the VNE problem (1)
is extended with respect to edge-specific parameters and (2)
new optimization objectives are introduced. Furthermore, (3)
general challenges for future VNE algorithms in mobile edge
networks are discussed. To the best of our knowledge, these
parameters, optimization objectives, and mobile edge related
challenges have not been discussed in literature before.
A. New VNE parameters
VNE algorithms take different parameters of network resources into account for calculating network embeddings. As
an example for such parameters, substrate resources have individual capacities, and virtual resources have respective requirements (also known as demands). Various kinds of resources are
67

ADVCOMP 2014 : The Eighth International Conference on Advanced Engineering Computing and Applications in Sciences

assigned to nodes and links: For example, in traditional VNE

scenarios, CPU resources are assigned to nodes and bandwidth
resources to links. Such a strict distinction in node and link
parameters, however, is problematic, since some parameters
influence each other. For example, CPU utilization of a node
being part of a path between two nodes influences available
bandwidth on that communication path.
Fischer et al. propose a classification of VNE parameters: Primary/secondary and functional/non-functional [10].
Primary parameters like CPU resources are directly assigned to
a substrate resource. Secondary parameters, however, depend
on other primary parameters and these side-effects have to
be considered first. As an example, packet loss at a node
depends on the primary parameters CPU resource and memory
resource. Functional parameters like CPU-/bandwidth capacity
specify low-level functionality. Non-functional parameters are
high-level properties. For example, resilience and security are
both non-functional parameters.
Being two well-known parameters common to most existing VNE approaches, CPU and bandwidth also apply in
the context of MEC. Both core and edge nodes provide CPU
capacities. On core nodes, CPU capacity is consumed for tasks
like routing and mobility-specific tasks like authorization and
billing. Edge nodes, i.e., MEC servers, provide CPU capacity
for applications deployed by the ASPs. In the following, new
VNE parameters are introduced that are applicative to the MEC
scenario. Parameters are classified as being primary/secondary
and functional/non-functional.
MEC Coverage (primary & functional) Base stations provide wireless link capacity, connecting mobile devices to the
mobile network. The range of a typical macrocell base station
covers several dozens of kilometers. Other types of base
stations, small cells like microcells, cover much lower ranges.
Relay nodes are base stations providing enhanced coverage at
cell edges and hot-spot regions, covering the same region as
the main base station.
Coverage refers to the geographical region covered by the base
stations of the same mobile operator, i.e., the geographical
region in which mobile devices are able to communicate with
the operators core network. Coverage depends on lower level
characteristics such as physical location of the base station,
transmission power and environmental influences. Increasing
the transmission power of a base station increases the radius
of the covered region, but also decreases average bandwidth
available to the mobile users. Increasing coverage leads to
interferences with other base stations if frequencies are reused.
In the MEC scenario, coverage also influences availability of
resources provided by MEC servers due to the fact that MEC
servers are directly connected to base stations. Thus, increasing coverage also increases availability of MEC resources.
However, since more mobile users are able to reach the MEC
server, this leads to higher utilization of the MEC server and,
thus, to higher latency. Therefore, there is a tradeoff between
availability and utilization/latency of MEC resources.
MEC Server Storage (primary & functional) MEC servers
provide disk storage for virtual machines running on top of
them in order to, e.g., cache proximity-related data. This requirement is not considered by most existing VNE approaches.
Latency (secondary & functional) If the VNE algorithm
embeds a latency-sensitive link to a path spanning several
substrate links, the sum of all latency properties of these
links may never exceed the demanded maximum latency of
the corresponding virtual link. Latency is a critical factor
for several mobile applications. Therefore, operators define
an upper bound for reaching a MEC server from a certain
base station. Depending on this value, the VNE is either able
Copyright (c) IARIA, 2014.

ISBN: 978-1-61208-354-4

to choose between any nearby MEC server or is forced to

use substrate resources of a MEC server right at the given
base station. Communication latency is a secondary, functional
parameter, since it depends on CPU utilization and refers to
low-level functionality.
Regional Bandwidth Capacity (secondary & functional)
Cell capacity refers to available bandwidth resources provided
by one or more base stations within a region (up- and downlink). Bandwidth capacity is increased by operating multiple
geographically adjacent base stations, each covering a subset
of the same region, instead of just one macrocell. Furthermore,
LTE Advanced allows the parallel deployment of small cells
within the same region that is covered by a macrocell (enhanced Inter-cell interference coordination, eICIC), leading to
improved bandwidth capacity. In the MEC scenario, regional
bandwidth capacity is proportional to the amount of available
MEC server resources that are accessible within a geographical
region. Bandwidth capacity, being a secondary parameter, is
limited by bandwidth resources of deployed base stations
and by bandwidth resources of the links connecting the base
stations to the core network.
Regional MEC Computing and Storage Capacity (secondary & functional) Similar to regional bandwidth capacity,
CPU/storage capacity refers to CPU/storage resources available in a region. CPU/storage capacity compounds of the
resources of MEC servers directly connected to the base
stations deployed in that region and computing capacity of
logically adjacent MEC servers, i.e., MEC servers that are
either directly connected to base stations covering that region
or well-connected through the mobile network to the base
stations.
MEC Resources per Mobile Device (secondary & functional) Since all mobile devices in a single cell share resources
corresponding to the above parameters CPU capacity, storage,
bandwidth and latency, VNRs demand resources on a per-user
basis, e.g., a certain minimum amount of bandwidth per user
in order to achieve the intended quality of experience. As a
consequence, the VNE has to lower the average number of
users in certain cells, which is done by adjusting the coverage
of adjacent cells (i.e., to distribute the users among multiple
smaller cells). Vice versa, when resource requirements are
smaller than available resources, the VNE is able to utilize
a smaller number of base stations, omitting certain nodes
completely, allowing to (temporarily) shut down these nodes
for energy-saving purposes.
The next subsection discusses MEC-specific optimization
objectives for the VNE problem. Due to the fact that many
MEC-related parameters are secondary, i.e., they depend on
other, primary parameters , the interdependencies are emphasized.
B. New Optimization Objectives
VNE algorithms compute the optimal (or near-optimal)
embedding of a set of VNRs. An embedding is optimized with
respect to an optimization objective. As an example, such an
optimization criterion is, in order to reduce cost for the IsP, to
minimize the number of substrate resources that are utilized by
the embedded virtual networks. This section discusses various
VNE optimization objectives in the context of MEC.
Increase MEC Coverage One main objective for network
operators is to provide cell coverage to their mobile users.
Operators both aim at expanding geographical dimension of
cell coverage, and the capacity of their cells: The more
bandwidth resources are available in a region, the more users
are able to connect to their networks (and pay for using their
68

ADVCOMP 2014 : The Eighth International Conference on Advanced Engineering Computing and Applications in Sciences

services). Therefore, one objective for future VNE algorithms

is to share resources provided by the base stations optimally
between VNRs with respect to coverage and bandwidth. However, this is not limited to base stations: In MEC, this also
applies to resources provided by MEC servers. Cells providing
bandwidth resources also provide computing capabilites. These
capabilities have to be adequately shared between network
operators covering the same regions.
There is a tradeoff between power consumption and coverage,
since more resources are needed to cover larger areas or to
provide higher bandwidth capacities.
Reduce Latency Reducing communication latency is of
major importance for todays mobile network operators. VNE
algorithms are obliged to take this aspect into account by
embedding VNRs in a way that minimizes latency at the edge
and/or in the core network. Reducing latency at the edge,
i.e., between MEC servers and base stations, leads to small
response times between services hosted at the MEC servers and
mobile users. To reduce overall network delay, delay between
base stations and gateway nodes connecting the network core
to the Internet (PGWs, cf. Fig. 3) has to be considered by
future VNE approaches.
Provide QoS-compliant Embeddings Besides latency, future
VNE algorithms have to consider flexible QoS- and QoEconsiderations. VNRs either demand strict resource assignments which have to be reserved exclusively. As an example,
a VNR requests 1km cell capacity in a specific area and 10GB
of MEC server storage capacity. These resources are then
exclusively reserved for the VNR and thus can not be shared
with other VNRs. As an alternative, VNE algorithms should
balance utilization throughout the substrate network resources
in order to guarantee equivalent QoS for all VNRs. In this
case, VNRs do not request such strict assignments and the
VNE aims at providing well-balanced cell coverage and MEC
server capacities for all VNRs.
Maximize Regional MEC Capacities Regional bandwidth/
computing/ storage capacity refers to available resources provided by one or more base stations or MEC servers within
a region. Bandwidth is limited by resources of the base
stations, but also by the links connecting the base station to the
core network. Bandwidth capacity is increased by overlapping
coverage of geographically adjacent base stations. Computing
capacity is increased by utilizing logically adjacent MEC
servers, i.e., MEC servers that are well-connected through the
core network to a base station covering the region.
Provide Resilience / Fault Tolerance for MEC services
Resilience at the network edge is provided by a VNE by
allocating resources of additional, geographically adjacent base
stations covering the same region. If another base station fails,
this backup resource takes over. The VNE has to ensure that
the backup resource does not introduce any interferences with
other base stations. Backup MEC server capabilities are provided by allocating logically adjacent MEC server resources.
The path from the base station to the backup MEC server
should provide similar latency conditions as the old one.
Reduction of Power Consumption Only few VNE approaches aim at reducing power consumption of substrate
resources [10]. However, since they focus on data center
routers and servers, these approaches are only partly applicable
in the context of MEC. In the MEC scenario, new power
models and interdependencies have to be considered: Increasing transmission power of base stations extends coverage,
but also influence other base stations [11]. Decreasing power
consumption decreases coverage, impelling mobile devices to
re-register to other base stations due to poor connection quality
to the current BS. This leads to load shifting both at the edge
Copyright (c) IARIA, 2014.

ISBN: 978-1-61208-354-4

of the mobile network and, as a consequence, also in the

network core. Current energy-aware VNE approaches aim at
switching off as many substrate entities as possible in order
to decrease overall power consumption. However, in the MEC
scenario, this not only results in decreased coverage, but also
in reduction of MEC server computing and storage capacity. If
the connection of the base station to the core network is shut
down, the MEC server gets isolated. MEC servers connected
to base stations that are already in use should be preferred
instead of switching on MEC servers connected to inactive
core resources.
Therefore, there is a tradeoff that should be taken into account
by future VNE algorithms: Coverage of base stations, available
bandwidth, and MEC server capacity in this region vs. energy
efficiency. How sparse can network infrastructure be / how
many cloudlets can be switched off, while still ensuring enough
bandwidth and CPU capacities?
Provide Security at the Edge Due to security considerations, not all operators or ASPs want to share MEC servers
running their applications with other operators. Therefore,
virtual MEC servers of different providers are embedded onto
different substrate MEC servers in order to reduce the risk of
any malicious influence (e.g., denial of service attacks). The
challenge for new VNE approaches is to find an embedding
that considers security constraints by still ensuring latency/QoS
demands.
C. Challenges in Mobile Edge Networks
This subsection outlines several general challenges for
VNE algorithms in the context of large-scale, fully-virtualized
mobile networks.
Scalability As mentioned before, the VNE problem is N Phard. Therefore, most VNE approaches are based on heuristics.
This results in non-optimal solutions, but decreases the size
of the problem and leads to significant improvements with
respect to runtime. However, almost all algorithms rely on
a single, central node calculating the VNE. Centralization
hinders scalability for large-scale, dynamic networks. In fact,
most algorithms were evaluated with substrate networks spanning only few dozens or hundreds of nodes. Such settings
might be realistic in the context of middle sized networks
like testbeds. However, it is far away from national-wide or
transnational mobile networks. One might be puzzled by the
fact, therefore, why only few approaches are distributed: In
fact, almost all VNE algorithms are centralized and require
full knowledge of the substrate network topology and substrate
resources [10], [12]. In large-scale environments like real-life
mobile networks, current VNE algorithms are stretched to their
limits [13].
Mobility-Awareness of Embeddings Another VNE challenge in MEC networks is to provide mobility-awareness.
Mobile users move between neighbored cells, resulting in
additional handoff overhead and routing of traffic through
different paths. Therefore, embedding communication paths
of virtual base stations covering adjacent cells on similar core
network paths reduces variations in latency (jitter). A mobilityaware VNE algorithm aims at providing both physical and
logical proximity for virtual base stations covering adjacent
regions.
Utilization-Awareness of Embeddings In mobile networks,
both edge and core utilization fluctuates significantly depending on the time of day. Demand also varies due to big sports
events, new years Eve, etc. Davy et al. propose the usage of
user mobility models for shared network resources, predicting
aggregated movement patterns of mobile users. Furthermore,
popularity prediction methods for video content discussed,
69

ADVCOMP 2014 : The Eighth International Conference on Advanced Engineering Computing and Applications in Sciences

predict distribution of video content based on their expected

popularity [13]. Integrating such prediction methods is seen as
promising step towards utilization-awareness VNE algorithms.
Being static and centralized, most VNE algorithms cannot
cope with dynamic and flexible VNR requirements in largescale environments. Therefore, novel VNE algorithms should
be both dynamic and distributed in order to provide flexibility
and adaptability in large network scenarios.
IV.

[1]

[2]

[3]

R ELATED W ORK

This section discusses related work: Several papers related

to mobile network virtualization are mentioned. To the best
of our knowledge, there is neither related work on network
virtualization in the context of edge computing networks nor
on VNE algorithms in this context.
Virtualization-based isolation techniques in the context of
mobile networks are proposed as an enabling technology for
future, cost-efficient mobile networks, shared and operated by
multiple network operators. Different network infrastructure
sharing scenarios are discussed and sharing options are classified based on different business models [1].
A general virtualization-enabled network architecture is
proposed in [6]. The advantages of shared, hereogeneous
network infrastructures are emphasized and applications of
network virtualization in this context is discussed. An approach
towards the virtualization of LTE networks has been introduced
by [8]. Authors discuss the advantages of virtualizing LTE
infrastructures and elaborate on virtualization of the LTE air
interface.
An extensive and up-to-date classification of current VNE
parameters and objectives is given in [10]. Many VNE approaches have been proposed so far, focussing on embedding
objectives like cost-optimization, resilience etc. While some
objectives are related to the ones presented in this paper (e.g.,
resilience and security), none considers the special demands of
MEC networks. Since InPs and network operators usually aim
for a combination of multiple (possibly contrary) objectives
(e.g., reducing power consumption to a certain extend by also
providing a sufficient degree of network resilience), an in-depth
evaluation of these approaches in combination with the novel
VNE parameters and objectives presented in this paper is left
for future work.
V.

R EFERENCES

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

A. Khan, W. Kellerer, K. Kozu, and M. Yabusaki, Network sharing in

the next mobile network: TCO reduction, management flexibility, and
operational independence, Communications Magazine, IEEE, vol. 49,
no. 10, pp. 134142, 2011.
Ericcson, Ericcson mobility report june 2013, http://www.ericsson.
com/res/docs/2013/ericsson-mobility-report-june-2013.pdf [retrieved:
2014-01-28].
Intel and Nokia Siemens Networks, Increasing mobile
operators
value
proposition
with
edge
computing,
http://nsn.com/system/files/document/edgecomputingtechbrief 328909
002 0.pdf [retrieved: 2014-01-28].
IBM
Corporation,
Smarter
wireless
networks;
add
intelligence
to
the
mobile
network
edge,
http://public.dhe.ibm.com/common/ssi/ecm/en/wsw14201usen/WSW14
201USEN.PDF [retrieved: 2014-01-28].
M. Satyanarayanan, P. Bahl, R. Caceres, and N. Davies, The case for
vm-based cloudlets in mobile computing, Pervasive Computing, IEEE,
vol. 8, no. 4, pp. 1423, 2009.
M. Hoffmann and M. Staufer, Network virtualization for future mobile
networks: General architecture and applications, in Communications
Workshops (ICC), 2011 IEEE International Conference on. IEEE,
2011, p. 15.
E. Kudoh and F. Adachi, Power and frequency efficient virtual
cellular network, in Vehicular Technology Conference, 2003. VTC
2003-Spring. The 57th IEEE Semiannual, vol. 4. IEEE, 2003, pp.
24852489. [Online]. Available: http://ieeexplore.ieee.org/xpls/abs all.
jsp?arnumber=1208838
Y. Zaki, L. Zhao, C. Goerg, and A. Timm-Giel, LTE mobile network
virtualization: Exploiting multiplexing and multi-user diversity gain,
Mobile Networks and Applications, vol. 16, no. 4, pp. 424432, 2011.
D. Fesehaye, Y. Gao, K. Nahrstedt, and G. Wang, Impact of cloudlets
on interactive mobile cloud applications, 16th International Enterprise
Distributed Object Computing Conference, pp. 123132, Sep. 2012.
A. Fischer, J. F. Botero, M. T. Beck, H. De Meer, and X. Hesselbach,
Virtual network embedding: A survey, Communications Surveys &
Tutorials, IEEE, vol. 15, no. 4, pp. 18881906, 2013.
F. Richter, A. J. Fehske, and G. P. Fettweis, Energy efficiency aspects
of base station deployment strategies for cellular networks, in Vehicular
Technology Conference Fall. IEEE, 2009, p. 15.
M. T. Beck, J. F. Botero, A. Fischer, H. De Meer, and
X. Hesselbach, A distributed, parallel, and generic virtual network
embedding framework, in IEEE Intl Conf. on Communications (ICC).
IEEE, 2013. [Online]. Available: http://www.net.fim.uni-passau.de/pdf/
Beck2013a.pdf
S. Davy, J. Famaey, J. Serrat, J. L. Gorricho, A. Miron, M. Dramitinos,
P. M. Neves, S. Latre, and E. Goshen, Challenges to support edge-as-aservice, Communications Magazine, IEEE, vol. 52, no. 1, pp. 132139,
2014.

C ONCLUSION

This paper proposes network virtualization as a key technology for future mobile edge computing networks. The Virtual Network Embedding problem is analyzed in this context
and new challenges for future embedding algorithms are
discussed. To the best of our knowledge, no VNE approach
has been published so far considering the MEC-specific network parameters and optimization objectives presented here.
Therefore, the authors hope that this position paper provides
an initial step towards new VNE approaches. As a first step,
the authors are currently implementing a delay-aware VNE
algorithm that considers geographical constraints as well as
latency considerations.

Copyright (c) IARIA, 2014.

ISBN: 978-1-61208-354-4

70