A practical architecture for mobile edge computing

A Practical Architecture for Mobile Edge
Computing
Tejas Subramanya∗, Leonardo Goratti∗, Shah Nawaz Khan∗,
Emmanouil Kafetzakis†, Ioannis Giannoulakis‡, Roberto Riggio∗
∗FBK CREATE-NET, Trento, Italy; Email: {t.subramanya,lgoratti,s.khan,rriggio}@fbk.eu
†Orion Innovations, Athens, Greece; Email: mkafetz@orioninnovations.gr
‡NCSR Demokritos, Athens, Greece; Email: giannoul@iit.demokritos.gr
Abstract—Recently, mobile broadband networks are focused
on bringing additional capabilities to the network edge. For
instance, Mobile Edge Computing (MEC) brings storage and
processing capabilities closer to the mobile user i.e., at the radio
access network, in order to deploy services with minimum delay.
In this paper, we propose a resource constrained cloud-enabled
small cell that includes a MEC server for deploying mobile
edge computing functionalities. We present the architecture with
special focus on realizing the proper forwarding of data packets
between the mobile data path and the MEC applications, based
on the principles of SDN, without requiring any changes to
the functionality of existing mobile network nodes both in the
access and the core network segments. The significant benefits
of adopting the proposed architecture are analysed based on a
proof-of-concept demonstration for content caching application
use case.
Index Terms—5G, Mobile Edge Computing, Content Caching
I. INTRODUCTION
The growth in mobile data consumption has skyrocketed in
recent times due to the growing demand for more contextual
and immersive mobile experiences like Augmented Reality
and Virtual Reality. These applications have requirements in
terms of both latency and bandwidth that the traditional mobile
network architecture fails to meet. Mobile Network Operators
(MNOs) are trying to keep up with these demands by increas-
ing the cellular network capacity through the densification of
the Radio Access Networks (RAN). However, there is a strong
need to rethink some of the fundamental aspects of network
design to enhance mobile user experience. A key problem is
the large end–to–end delay between the end users and the
requested services as traditional services are deployed in data–
centres serving a large number of users in a highly centralized
manner. One promising solution is the emerging of the Mobile
Edge Computing (MEC) concept which provides a distributed
computing environment allowing applications and services to
be executed in close proximity to the actual end–users, thus
improving both time-to-response and end-to-end latency.
We present a SDN-enabled Mobile Edge Computing solu-
tion which places services at the network edge using cloud-
enabled small cells. The proposed architecture handles the
forwarding of data packets between the mobile data path and
the MEC applications. Using SDN concepts, a unified control–
plane interface has been developed to retrieve network context
information which is subsequently used to selectively steer
data traffic from the RAN nodes to the MEC server and vice–
versa.
The main contribution of this work is twofold. On one
hand, we present a modular SDN–enabled MEC architecture
that integrates with the existing LTE systems. We also present
a prototype realization of this architecture using a software-
based LTE system implementation. On the other hand, we
demonstrate the applicability of the proposed architecture to a
practical use case, namely content caching.
The rest of the paper is organized as follows. Section II
reviews related work on MEC, while Sec. III provides back-
ground information on LTE network architecture. The pro-
posed MEC architecture, its communication interfaces and
service development framework are discussed in Section IV.
Section V demonstrates the proof of concept by analyzing
various measurement results for content caching application.
Conclusions and future work are discussed in Section VI.
II. STATE-OF-THE-ART
The MEC related research has received considerable atten-
tion in recent times. In [1], the authors provide an architectural
blueprint for MEC and discuss the technical advantages it
offers by presenting a number of use cases. A taxonomy
of MEC along with its key attributes are discussed in [2].
The authors also present some promising real-time MEC
application scenarios.The challenges involved in commercial
deployment of MEC and the progress towards it are discussed
in [3]. In [4] and [5], the authors present different mobile
offloading techniques in cellular networks to enable low-
latency applications. Other similar concepts to enable edge
computing capabilities such as Fog Computing and Cloudlet
are presented in [6], along with a detailed comparison.
In this paper, we explore the possibility of leveraging SDN
and cloud technologies to enable flexible inter-working be-
tween the proposed MEC application framework and the LTE
mobile network, through MEC RAN Information Interface. It
should be mentioned here that the deployment of our proposed
architecture can be achieved without making any changes to
the functionality of access and core network.
III. BACKGROUND: THE LTE MOBILE NETWORK
In this section, we will briefly introduce the overall LTE
network architecture (see Fig. 1) and we will provide an
overview of the basic procedures in LTE. At a very high level,
an LTE network is composed of two major elements: the RAN
and the Evolved Packet Core (EPC). The RAN comprises one
logical node, the evolved NodeB (eNodeB), which connects
to the User Equipments (UEs). The EPC consists of many
logical nodes such as Mobility Management Entity (MME),
the Serving Gateway (SGW), the Packet Gateway (PGW) and
2017 IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN)
978-1-5386-3285-7/17/$31.00 ©2017 IEEE

Fig. 1: LTE network overview
the Home Subscriber Server (HSS). Each of these elements
are interconnected using standardized interfaces.
UE Attach Procedure. This consists of three phases: (i)
IMSI Acquisition: Once a UE establishes a radio link synchro-
nization with the eNodeB, it attempts to attach to the network
by sending an ’Attach Request’ message to the MME which
contains International Mobile Subscriber Identity (IMSI) in
it and thus MME obtains IMSI from this message. (ii) UE
Authentication and Security Setup: Once MME acquires IMSI,
various authentication and security procedures takes place
between UE and MME with the help from HSS. Therefore
the messages exchanged between them from here on are
encrypted and integrity-protected for secure communication.
(iii) Session Establishment or Data Bearer Setup: After se-
curity mechanisms and once MME obtains UE subscription
information from HSS, a default bearer is created for the UE,
by establishing tunnels between the eNodeB and the SGW and
between the SGW and the PGW. The tunnels are established
by exchanging Tunnel Endpoint Identifiers (TEID) for this
bearer among eNodeB, MME, SGW and PGW.
UE Data Transfer: Once a UE attaches to the network by
creating a default bearer through EPC, it can send/receive data
to/from Packet Data Networks using GPRS Tunneling Protocol
(GTP) [8]. In uplink, once eNodeB receives the data packets
from UE over air interface, it encapsulates the packet in GTP,
IP, UDP headers by setting the value of TEID field in GTP
header as expected by SGW and forwards the packet towards
SGW. The SGW removes the encapsulated headers and adds
its own GTP, IP, UDP headers and forwards the packet towards
PGW. The PGW removes the encapsulated header and adds its
own GTP, UDP, IP headers and forwards the packet towards
PDN. The similar process repeats in downlink data transfer
from PDN to UE.
UE Detach Procedure: Once UE is done using LTE
services, it may initiate a detach procedure by sending a
’Detach Request’ message towards MME. The MME removes
the UE context by releasing its bearers and sends a ’Detach
Accept’ response message to UE.
As we will see later, in order to transparently encapsulate
and decapsulate GTP traffic so as to reroute user plane traffic
towards RAN edge cloud, it is fundamental to correctly obtain
the radio network information.
IV. SYSTEM ARCHITECTURE
A high–level overview of the SDN–enabled MEC archi-
tecture presented in this paper together with its main com-
ponents is reported in Fig. 2. Our design is modular enough
with the following main components: MEC RAN Information
Interface (MRI), MEC Application Platform Services, MEC
Hosting Infrastructure and MEC applications that leverage
the underlying platform through Application Programming
Interfaces (API), and is in accordance with the ETSI MEC
Fig. 2: System Architecture
Fig. 3: Sequence diagram of the edge caching application
architecture as presented in [9]. For the moment, our design
and implementation is mature enough to identify the benefits it
offers in terms of latency, service delivery and user experience,
based on the content caching MEC application as an use case.
In the rest of this section we will describe the various com-
ponents of the proposed architecture. Moreover, we will also
describe the role of such components in the implementation
of a edge content caching application. The Sequence diagram
associated to this application is reported in Fig. 3.
A. MEC RAN Information Interface (MRI)
MRI acts as a RAN–specific interface between MEC ap-
plications and the underlying physical or virtual network by

providing a real–time insight into radio network information
of the mobile users, performance statistics of the cell and user
location awareness. It enables the MEC server to communicate
with other network entities, i.e., MME, S-GW, P-GW and
eNodeB, through RAN-specific control plane (S1-C and X2-
C) and user plane interfaces (S1-U and X2-U).
The MRI is implemented based on Tshark [10] (a terminal
based version of Wireshark), that monitors the interfaces
between eNodeB and EPC. The three types of information
collected by MRI include: (i) Control Plane Information,
by capturing and processing signalling messages between
eNodeB and MME, (ii) Data Plane Information, by processing
data plane packets between eNodeB and SGW, (iii) UE/eN-
odeB Low-level Information, relevant to Physical, Medium
Access Control (MAC), Radio Link Control (RLC) and Packet
Data Convergence Protocol (PDCP) layers of UE or eNodeB.
Based on our content caching use case, 1 to 4 in Fig.
3, illustrates on the radio network parameters retrieved during
UE Attach procedure (IMSI, Tracking Area Identity (TAI),
UE/eNodeB/SGW IP, eNodeB/SGW GTP TEID and Glob-
ally Unique Temporary Identifier (GUTI)), to make decisions
on transparently modifying and rerouting data traffic from
mobile network nodes to MEC applications and vice-versa.
The implementation of MRI can also be used for detecting
UE Detach events, handover events and bearer modification
events in the mobile network, which are not illustrated in this
paper due to limited space constraints. Some useful parameters
collected include Cell Radio Network Temporary Identifier (C-
RNTI), UE capability, bandwidth, carrier frequency, Mobility
state, QoS Class Identifier (QCI), Allocation and Retention
Priority (ARP), Guaranteed Bit Rate (GBR), Maximum Bit
Rate (MBR) and Packet Data Network (PDN) identity.
B. MEC Application Platform Services
It is the collection of essential services provided to mobile
edge applications hosted within the MEC server.
Radio Network Information Service (RNIS), is basically
a collection of radio network information related to users and
cells captured by MRI and maintained in a database. The
database can be accessed by other services and authorized
applications within the MEC server.
Analytics and Event Capture (AEC), analyses the control
plane information from RNIS and sends attach/detach/han-
dover event triggers to Traffic Offload Service along with the
necessary radio network information as seen in 5 in Fig. 3.
Traffic Offload Service (TOFS), is realized together with
the Lagopus virtual switch and Ryu SDN Controller, which has
been extended to support GPRS tunneling protocol. Addition-
ally, a Ryu application has been developed to provide state-
ful GTP/IP/UDP header encapsulation/decapsulation service
which is required to support MEC functionality. Based on the
triggers received from AEC, Ryu controller pushes flow rules
onto Lagopus switch using OpenFlow protocol for selectively
routing data traffic towards MEC application service chain as
seen in 6 from Fig. 3.
Based on our use case, four flow rules 7 to 10 as seen
in Fig. 3, are important to understand. For example, in 7
Lagopus matches on uplink packets, to check if the traffic
is GTP (UDP port=2152), originated from UE and if TEID
is same as that of SGW TEID. If so, Lagopus strips outer
UDP, outer IP and GTP headers on the matched packets and
forwards the traffic towards MEC application service chain by
changing the destination MAC address to that of next hop in
the path. In 8 , IP packets are matched to see if the packets
arriving from MEC application service chain are originated
from UE (source IP=UE IP), and if so, Lagopus performs
GTP, IP and UDP header encapsulation and forwards the traffic
towards EPC by changing the destination MAC address to that
of next hop in the path. Similarly, 9 and 10 in Fig. 3 matches
packets in downlink and performs respective routing actions.
TOFS is supplied to MEC applications in 2 ways:
• Pass-through mode [ 7 to 10]. In our use case, if the
user requested content is not available in local edge
cache, the caching application VNF can modify the data
traffic and pass it back to the original PDN connection
(3GPP bearer). In this case, caching application acts as a
transparent proxy to the user.
• End-point mode [ 7 and 10]. In our use case, if the
requested content from the user is already cached, the
data traffic is terminated by the caching application VNF
which acts as a server to the user.
C. MEC Applications
MEC applications are an individual or chain of service
VNFs belonging to a specific use case, that run on top of
the proposed architecture. We implemented content caching
application Squid as a VNF.
D. MEC Hosting Infrastructure
It provides virtual resources for computation, storage and
networking within the MEC server. It also provides connec-
tivity service among a chain of applications, services, 3GPP
network, local networks and external networks, by routing
traffic among them.
V. PERFORMANCE EVALUATION
The experiments were performed using Open Air Interface
(a software-based LTE system covering 3GPP complaint full
LTE protocol stack for both eNodeB and EPC). We used
commercial LTE smart phones with programmable SIM card
in them as mobile users. We used a low-power, low-cost,
advanced communication computer (Soekris net6501) as a
light-weight MEC server that includes Ryu controller, virtual
Lagopus switch and a caching application (i.e., Squid). We
measured the performance of individual components within
the MEC server in terms of processing time, latency and re-
sponsiveness, and illustrated the potential for reducing average
round trip time (RTT) and overall backhaul traffic by caching
content at the MEC server based on our approach.
A. MEC RAN Information Interface
We performed a series of 10 UE initiated attach and detach
events to measure the time taken by MRI to detect the UE
events. The average processing time to detect a UE attach
event is 1.4ms and to detect a UE detach event is 1ms, without
considering the Tshark packet parsing time (additional 600ms).
B. MEC Application Platform Services
Once the UE attaches successfully to the network, we
measure the total time taken by the AEC service and the Ryu
controller to install/uninstall traffic rerouting flow rules onto
Lagopus. The mean value for the 10 UE attach events is 16ms
with a variance of 5.55ms and standard deviation of 2.35ms.

A
pple
Bing
G
ithub
M
icrosoftW
ikipedia
0
100
200
300
RoundTripTime(ms)
’Original Response’ ’Cached Response’
(a) Round Trip Time.
Static Content
Local Server
YouTube
0
2
4
TotalTCPTransfers(MB)
’Original Response’ ’Cached Response’
(b) Cache Hits.
Fig. 4: Evaluation Results
C. RTT and Backhaul Congestion Improvement
The first experiment focuses on the different response times
the user gets when served by caching application in the MEC
server, instead of directly from the origin Web server. The
comparison among them is seen in Fig. 4a, which is obtained
by making a web request using Curl (a command-line tool) to
five popular web pages, from the LTE UE terminal (Huawei
USB dongle) located about 2 meters to the serving eNodeB.
The measurements have been repeated for 10 successive times.
The difference in average RTT is significantly lower with
caching, mainly because of the evasion of core network and
web server processing delay in the backhaul.
The second experiment focuses on the amount of web
content that can be cached to serve the users directly from local
cache instead of the origin web server. The analysis is shown
in Fig. 4b, which is based on three web request operations
performed by the LTE UE terminal with Squid as the caching
application. (i) User makes a GET request to receive static
content such as html pages, css scripts, javascripts, images
and binaries. The response is completley cached by Squid,
and any identical requests later on are served directly from
the local cache. (ii) Similarly, when a user requests a video
hosted on our local web server (Apache), the entire video is
cached by Squid. (iii) However, when a user requests a video
from some of the major media sites like YouTube or Google
Videos, Squid alone cannot cache the entire video, since this
traffic does not behave like simple HTTP web pages [11]. In
such a situation, Squid uses custom video caching solutions to
cache only the static content from these websites (25%). As
seen in Fig 4b, in all three cases, mobile backhaul traffic is
significantly reduced with caching. It is important to note that,
in this paper, we do not focus on the performance of different
caching algorithms.
VI. CONCLUSION
We leveraged SDN and cloud to present a ETSI complaint
modular MEC architecture for LTE mobile networks. The
architecture proposes three important components: MRI, to
communicate with the underlying radio network; MEC Ap-
plication development Platform, that provides a number of
value added services and APIs to application developers; and
MEC applications, for offering new personalized services to
specific customers. We also analyze the video content caching
use case as a proof of concept for the proposed architecture
by transparently intercepting and rerouting the traffic towards
MEC applications. In future, we plan to extend the proposed
architecture with additional value added services such as
policy control and service registry to support more low-latency
mobile edge applications.
ACKNOWLEDGEMENTS
Research and development leading to these results has re-
ceived funding from the European Framework Program under
H2020 grant agreement number 671596 (SESAME).
REFERENCES
[1] Y. C. Hu, M. Patel, D. Sabella, N. Sprecher, and V. Young, ”Mobile edge
computing a key technology towards 5G”, ETSI White Paper, vol. 11,
2015.
[2] A. Ahmed and E.Ahmed, ”A survey on mobile edge computing”, in Proc.
of IEEE ISCO, 2016, Coimbatore, India.
[3] H. Li, G. Shou, Y. Hu, and Z. Guo, ”Mobile edge computing: progress
and challenges”, in Proc. of IEEE MobileCloud, 2016, Oxford, UK.
[4] J. Cho, B. Nguyen, A.Banerjee, ”SMORE: Software-Defined Networking
Mobile Offloading Architecture”, in Proc. of ACM ATC, 2014, Chicago,
IL, USA.
[5] A. Banarjee, X. Chen, J. Erman, V. Gopalakrishnan, S. Lee, and J. Van
Der Merwe, ”MOCA: a lightweight mobile cloud offloading architec-
ture.”, in Proc. of the ACM MobiArch, 2013, Krakow, Poland.
[6] K. Dolu, S.K. Datta, ”Comparison of Edge Computing Implementations:
Fog Computing, Cloudlet and Mobile Edge Computing”, in Proc. of
Global IOT Summit, 2017, Geneva, Switzerland.
[7] 3rd Generation Partnership Project, ”Radio Access Network; Evolved
Universal Terrestrial Radio Access Network (E-UTRAN); Architecture
description”, 3GPP TS 36.401, 2015.
[8] European Telecommunications Standards Institute, ”General Packet Ra-
dio System (GPRS) Tunnelling Protocol User Plane (GTPv1-U)”, ETSI
TS 129 281, 2014.
[9] European Telecommunications Standards Institute, ”Mobile Edge Com-
puting (MEC); Framework and Reference Architecture”, ETSI GS MEC
003, 2016.
[10] Wireshark, ”Tshark Manual Pages”. [Online]. Available:
https://www.wireshark.org/docs/man-pages/tshark.html.
[11] Squid, ”Caching Dynamic Content”. [Online]. Available:
http://wiki.squid-cache.org/ConfigExamples/DynamicContent

A practical architecture for mobile edge computing

More Related Content

What's hot (18)

Similar to A practical architecture for mobile edge computing (20)

Recently uploaded (20)

A practical architecture for mobile edge computing