Conference Paper: Distributed cloud and de-centralized control plane A proposal for scalable control plane for 5G

Distributed cloud and de-centralized control plane
A proposal for scalable control plane for 5G
Amir Roozbeh
Ericsson Research, Cloud Technology
Stockholm, Sweden
Email: amir.roozbeh@ericsson.com
Abstract—5G is the next generation of mobile network.
The aim is to launch service starting in 2020. There are
many requirements on 5G, such as high capacity, low latency,
flexibility, and support for any-to-any communication. Cloud
technology, in the form of a distributed cloud (also known
as a network embedded cloud), is an enabler technology for
5G by providing flexible networks to meet different user
application requirements. On the other hand, Machine Type
Communication (MTC) is a primary application for 5G, but
it can add a high volume of control signaling. To manage the
expected high volume of control signaling introduced by MTC,
we identified the main control events that generate signaling
messages in the network, i.e., session management, hand over
management, and tracking area update. Then, we proposed
a decentralized core network architecture optimized for the
identified control events. The proposed control plane functions
are independent of each other in the sense that each function
can be executed separately. The control functions can utilize
a distributed cloud ( embedded in the 5G core network) to
manage the enormous amount of control signaling by handling
this signaling locally, i.e., close to the end user. Additionally, we
present an analysis of the control signaling performance for each
proposed control function. The result shows that it is beneficial
to move session management to data centers collocated with
the base station in the 5G network when there is high user density.
Keywords: 5G, decentralized control function, core network,
network enabled cloud
I. INTRODUCTION
Mobile communications have had a significant impact on
the way that people and devices interconnect. Historically
there has been an evolution of wireless technology in terms
of radio access technology, data rates, operational bandwidth,
and mobile network architecture every ten years [1].
First generation mobile communication aimed at connect-
ing people by combining communications and mobility in
one package utilizing a circuit switch network in 1980. The
(r)evolution in mobile communication began with adapting the
mobile architecture to support packet switch networking. This
lead to the second generation 2G (i.e., introduced in 1990s),
such as Global System for Mobil Communications (GSM),
and later 3rd generation (3G) mobile networks. 2G introduced
increased capacity and coverage, enabling additional services
beyond voice, such as SMS and Email. Later, 3G introduced
mobile broadband (i.e., up to 2 Mbps) by combining high-
speed mobile access with IP services. In recent years, wireless
technologies (r)evolution resulted in an all-IP architecture, i.e.,
fourth generation (4G) together with Long Term Evolution
Advanced (LTE Advanced). The term all-IP refers to the fact
that all data and signaling within the mobile network utilize
the Internet Protocol (IP) on the network layer.
Today there are wide ranges of requirements on the next
generation of mobile communication, i.e., fifth generation
(5G). These requirements include: higher data rates and in-
creased traffic capacity (i.e., 1000 time more capacity than 4G),
improved reliability, support for enormous numbers of devices,
and lower latency. As a result, 5G has to be more efficient and
scalable than 4G [2]. 5G is concerned not only with providing a
service for people but also serving any devices that may benefit
from being connected. In another word, the next generation of
mobile communication should enable anyone or anything to
access information and share data anywhere, anytime [3].
IOT is one of the important applications for 5G. In the
transition to a networked society and connected planet with
5G, a massive number of devices, which can be embedded
into the environment, will be connected to networks. The
concept of IOT combines different Machine-Type Commu-
nications (MTC) (i.e., Machine-to-Machine and Machine-to-
Human) with human communication (i.e., Human-to-Human
and Human-to-Machine). Wearable devices, smart homes,
smart parking places, smart and self-driving cars, and smart
sensors are some example applications of MTC. IOT devices
can be categorized based on many factors such as their
portability (i.e., fixed or mobile) or communication direction
(i.e., one-way or two-way communication). Most IOT devices
are characterized by their simplicity, the fact that they usually
transmit small amounts of data, and whether they transmit
frequent or only from time to time.
5G has a requirement to support hundreds of billions
of low-power connections in order to bring IOT to peoples
daily lives. The enormous number of IOT devices and IOT
application characteristics in a future communication ecosys-
tem will have an impact on future mobile communication
architectures. IOT and MTC will effect both the user plane
and control plane traffic that need to be handled by future 5G
networks. Management of billions of IOT devices is expected
to add considerable signaling loads on the control plane of
the network, but this may lead to higher response times and
increased congestion at network control plane entities. The
future 5G control plane should be designed to avoid congestion
and reduce latency of the control plane, while providing a
flexible framework which can adapt to user requirements in a
wide variety of different scenarios.

In this paper, we present a decentralized control plane
function for 5G to provide the scalability needed for IOT
control signaling traffic. In this proposal, the mobile network
control plane is divided into different functions based on the
primary control events expected in a mobile network. Each
of the proposed functions can execute in a decentralized or
centralized way, depending upon the network’s demands to
handle signaling traffic.
The reminder of this paper is organized as follows:
Section II presents an overview of LTE network and 5G
characteristics and concept. Section III describes the concept
of IOT and how this can effect signaling load. Section IV
gives a description of the proposed control plane architecture
for 5G. Section V gives an analytic model used to calculate
signaling load on different control plane functions in 5G.
Additionally, a numerical analysis is made based on the
analytic model and metrics from an LTE network as a
baseline of 5G. Finally, we conclude the paper in Section VII.
II. FIFTH GENERATION (5G)
The 5th generation of mobile network (5G) is in its early
stage and lack a standard architecture and protocol. As a result,
this paper uses Long Term Evolution (LTE) technology as a
baseline to identify the main control plane events that produce
signaling load. The assumption is that a future 5G architecture
will evolve from today’s All-IP LTE. In this section we provide
overview about the LTE network followed by a description of
the expected 5G characteristics and underlying concepts.
A. LTE NETWORK
The LTE network architecture separates the control plane
and user plane (or data plane) [4]. The control plane is respon-
sible for control and transmission of signaling information,
while the user plane is responsible for forwarding user traffic.
This flat architecture has two parts: the Evolved-Universal
Terrestrial Radio Access Network (E-UTRAN) and Evolved
Packet Core (EPC). E-UTRAN consists of evolved node-
BSs (eNodeBs) that are responsible for providing wireless
connectivity to the user entities (UEs). EPC consists of four
primary entities: Mobility Management Entity (MME), Home
Subscriber Server (HSS), Packet Data Network Gateway (P-
GW), and serving gateway (S-GW). EPC provides functions
for user plane and control signaling management. Fig. 1 shows
the LTE network architecture. The EPC also includes standard
component of IP networks, such as switches, DNS, and NTP
servers (although these are not shown in this figure).
An MME is a critical control node in the EPC and supports
the most important control plane functions, specifically mobil-
ity managements managing security when device attaches to
the access network, and tracking and paging to support devices
in idle mode. MME uses information provided by HSS to
authenticate devices and updates UE’s location information in
HSS as part of its mobility management.
HSS is a database that contains subscriber-related informa-
tion. HSS provides support functions related to call and session
setup, user authentication, access authorization, and mobility
management.
The P-GW and S-GW of the core network act as a gateway
for user plane. These two elements provide connectivity
Fig. 1. LTE architecture.
between the UE and external IP networks. The S-GW is
logically connected to a P-GW and provides the intercon-
necting between the radio access network and EPC. The P-
GW interconnects the EPC with external IP networks. The P-
GW offers several functions, such as IP address management,
policy control, and charging.
LTE network elements communicate with each other by us-
ing standard interfaces defined by the 3rd Generation Partner-
ship Project (3GPP). For example, eNodeBs are interconnected
with each other by means of the X2 interface [5], while S10
is the standard interface between two MMEs. There are four
main events that initiate signaling control in LTE network:
a UE-originated session, a UE-terminated session, Handover
(HO), and Tracking Area Update (TAU).
A UE-originated session occurs when a UE attempts to
establish a connection to the EPC in order to receive or
send data. The signaling messages involved in this procedure
are shown in Fig. 2. To initiate the communication link, UE
starts the Radio Resource Control (RRC) connection procedure
with an eNodeB to ask for resources from access network,
(in step 1). In step 2, the UE sends an ATTACH and PDN
CONNECTIVITY request to MME to set up an internet
connection (this is referred to as a Non-Access Stratum (NAS)
service). If the MME is unable to identify the user (for example
because the UE has just at powered on or at the time initial
access to the network), then the MME starts identification and
authorization as NAS common procedure (in step 3). This
process involves six messages between the UE and MME. The
MME updates the HSS with the UE’s location while asking for
this subscriber’s profile. Also, there is a negotiation between
MME and S-GW and P-GW to establish an initial bearer for
this UE (in steps 4 - 7). Next the MME sends a context setup
request to the eNodeB that results in a signaling message being
sent between the eNodeB and UE to establish a bearer between
these two entities (in steps 8 - 10). When the initial context
configuration is satisfactory, the MME configure the user plane
of S-GW and sends an ‘attach accept” to UE (in steps 11 -
15 ). At this point, the data flow between the UE and external
network is established. Finally, when the session is complete
the MME can release all bearers allocated for this session (in
steps 16 - 21).
A UE-terminated session occurs when the network has data
for an idle UE. In this situation, the MME starts by paging all

Fig. 2. UE-originated session signaling message flow.
eNodeBs (within a tracking area size that UE registered with
it) in order to notify the UE that there is pending data for it.
The remainder of the procedure is same as for UE-originated
session, as shown in Fig. 3.
Handover (HO) can occur when the UE is idle or in
connected mode. We ignore those cases when HO occurs
when UE is in idle mode or when a HO occurs within a
single eNodeB (i.e., when a UE changes its cell sector),
as in both of these cases the MME is not involved in the
signaling procedure [6]. During a HO the UE may stay with
a given MME (i.e., an intra-MME HO) or need to change
its MME (i.e., an inter-MME HO). Intra-MME HO occurs
when an MME serves multiple eNodeBs. In this scenario, a
UE can continue to use its S-GW (i.e., intra-MME/SGW HO)
or changes its S-GW (i.e., intra-MME/inter-SGW HO). The
source eNodeB (SeNB) triggers a HO based on measurements
reported by a UE. A summary of intra-MME/SGW HO
signaling flows and a description of this HO procedure is
depicted in Fig. 4. These scenarios assume that the a X2
Fig. 3. UE-terminated session signaling message flow.
Fig. 4. Intra-MME/SGW HO signaling flows.
interface between source and destination eNodeB exists∗
.
The intra-MME/inter-SGW HO signaling flows contain two
additional pairs of signaling messages. The first pair between
MME and source S-GW (SSGW) release the bearer, while
the second pairs between MME and target S-GW (TSGW)
establish the bearer.
The inter-MME HO happens when more than one MME
is available in the network. Upon HO the UE can stay with
a given S-GW (i.e., inter-MME/intra-SGW HO) or changes
its S-GW (i.e., inter-MME/SGW HO). The source MME (S-
MME) controls the source eNodeB (SeNB) and source S-
GW (S-SGW) [if applicable], while the target MME (T-MME)
controls target eNodeB (T-eNB) and target S-GW (T-SGW) [if
applicable]. The message flow and description of the procedure
for inter-MME HO is depicted in Fig. 5. These scenarios
assume that the HO initiation needs to use the S2 interface.
Tracking Area Update (TAU) happens when a UE detects
that it has entered a new tracking area. If so, the UE updates
the network with this new tracking area information. If this
new tracking area is served by same MME, then this MME
accepts the TAU and register a new location for this UE
without interaction with HSS. However, if the serving MME
changes (i.e., inter-MME TAU), then the network forwards the
TAU to target MME. Next, the target MME updates the HSS
database with this new location information and the identity of
the target MME. The HSS cancels the UE’s location in source
MME and send the UE’s subscription data to the target MME.
The Inter-MME TAU message flow is depicted in Fig. 6.
B. 5G
The 5th generation (5G) is the next generation of wireless
technology and is expected to be deployed in the post 2020
timeframe. Currently, there is a global discussion on the
∗If this interface does not exist, then the HO procedure is very similar to
the inter-MME HO except for the involvement of an MME coordinating the
HO signaling loads.

Fig. 5. Inter-MME/SGW HO signaling flows.
Fig. 6. TAU signaling flows.
definition of 5G [7]–[9]. NO recognized standards body has
yet defined 5G. There are several proposals for 5G network
architectures in literature [3], [14], [15]. These proposals
generally assume a network upgrade from an All-IP network
architecture, and utilize cloud technology and variety of
different radio access technologies.
There are many factors that are driving the development
of 5G such as: multi-gigabit per second (Multi-Gbps), Internet
of Things (IOT), ubiquitous access to mobile services, cloud
technology, software defined networks (SDNs), and increasing
service complexity (e.g., 3D video and gaming, augmented
reality, and self-driving cars). Study shows that telecommu-
nication services are increasingly to be virtualized and will
migrate to the cloud [10], [11]. While previous generations
of mobile networks were communication centric, 5G is about
both communication and computing. As a result 5G is expected
to extend the continuing revolution of moving the cloud to the
network in the form of a network embedded cloud.
To achieve higher data capacity, cell deployment will be
denser, i.e., gradually increasing up to ten times the density
of deployment in today’s networks [2]. Additionally, coop-
erative communication and joint processing techniques, such
as coordinated multi-point (CoMP), massive multiple-input-
multiple-output (MIMO), and joint radio resource scheduling
can be utilized to provide higher data rates. These methods
introduce additional cost due to interfering links and intensive
data exchange and computation. Additionally, the 5G core
network must be able to handle all of the control and user plane
traffic generated by the increasing large number of network
connected devices.
Based upon the different demands for communication
in the future it is unlikely that one network deployment
model be able to serve all use cases and scenarios in 2020.
As a result, flexibility to adapt to different scenarios with
different requirements will be an inherent part of future 5G
networks. One way to provide flexibility for future mobile
network deployment is leveraging cloud technology, specifi-
cally telecommunication cloud or network embedded cloud as
a distributed cloud architecture in which network providers add
data centers within their network to offer cloud services [12].
The distribution of cloud deployment in operator networks can
range from medium size data centers at a back office to Nano
data centers coexisting with a Base Station (BS). Operators
can exploit this cloud to offload the computing requirement of
mobile networks (e.g., cloud-radio access network (C-RAN)
or to virtualize mobile core network functions) [13].
III. INTERNET OF THINGS (IOT)
In a future networked society, every thing that benefits from
a connection will be connected. As a result, IOT communi-
cation such as MTC, vehicle-to-vehicle (V2V), Machine-to-
Machine (M2M), smart objects and sensors can be embedded
in almost everything everywhere. IOT can be described as
an application that communicates with other IOT devices
or servers within the cloud (either cloud inside or outside
an operator’s network) using a communication network and
without human interaction. IOT devices can use any type
of network for their communication, such as wireless local
area networks, mobile networks including UMTS, LTE, and
future 5G depending on the required QoS, cost, and network
support. As mobile networks already have significant coverage
in order to services to humans, they might also be used to
carry IOT traffics [16]. However, current mobile networks were
not designed for MTC, hence 5G aims to support the IOT by
meeting the requirements for this communication.
Monitoring information from embedded sensors, remote
device configuration, and triggering alarms based on data
received from IOT devices are examples of IOT applications.
IOT devices can have a fixed location (e.g., sensors measuring
humidity in environment or other sensors inside a smart home)
or be mobile (e.g., sensors embedded within a car, a human
body, or other moving objects).
Data generated by IOT devices differs from data generates
by humans. Additionally, not all MTC applications have the

same characteristics [17]. IOT data can be small in size but
frequently transmitted. IOT may deploy billions of devices
(i.e., many more than the number of humans). The increasing
numbers of devices utilizing a mobile network is a primary
concern and may lead to problems such as congestion and
overhead both on the data and control planes. Congestion can
occur both in the radio access and core network. Although
each IOT device may send and receive a small amount of
data, the cumulative traffic from all of these devices can lead
to congestion. Additionally, IOT devices also generate control
and signaling traffic (even while they only transfer a small
amount of data) due to their need to attach or re-attach to
the network, and this control traffic can negatively effect core
network control entities.
There is an ongoing discussion regarding the prerequisite
for the network to meet MTC demands [17], [18]. Additionally,
There efforts have been made to address congestion control
for IOT and MTC, especially for LTE network. J. Wang,
et al. [19] proposed TCP-FT as an enhancement for TCP
to reduce congestion. Another suggestion was group based
traffic management [20], where in each group only one IOT
device communicated directly to the mobile network, hence
reducing the control signaling overhead in mobile network.
The time controlled policy introduced by 3GPP [21] is another
suggestion, in which a device only can communicate with a
network if it is not in a blocked period. The intention of most of
these proposals is to reduce congestion caused by IOT devices
in the radio access network. This paper proposes a solution
to reduce the load and probability of congestion in the core
network control entities for 5G.
IV. PROPOSED SOLUTION
This proposal assumes that the 5G architecture will not be
limited by current flat LTE architecture. Fig. 7 describes the
proposed architecture for future 5G. In this proposal, The core
network consists of two categories of nodes: control nodes and
gateway nodes. Control nodes are responsible for processing
and handling control signaling, while gateway nodes are
responsible for handling user plane traffic. The proposed 5G
control node characteristics are
Fig. 7. Proposed 5G core architecture.
1) A control node is virtualized and realize all required
control functions, such as (re)attachment handling,
mobility management, authorization, and authentica-
tion.
2) Control node functions are independent and can
run independently, in the same or separate physical
locations.
3) Each of control function can be executed in a different
physical location while maintaining a parent-child
relationship.
4) Each control functions can act as a parent, a child, or
both. The child acts the same as its parent and is able
to update its parent if required. As a result, logically
from UE’s point of view, the parent and child looks
like a single entity.
5) Control functions communicate with each other when
needed.
One suggestion to manage control signaling traffic in order
to avoid congestion in the 5G core network is placing those
control functions that make high demands upon the network
due to users, as a child function close to UE’s current radio
access point. Combining the telecommunication cloud and
the proposed 5G control node characteristics are the primary
enablers for this proposal. In this way, control functions can
be launched as a child in data centers within the core network
or data centers co-located with radio base stations. An SDN
can be employed to provide a communication between parent
and child control functions, i.e., enabling the child to migrate
to an other location.
To achieve this goal, the design and implementation of
the control node should be done such that it can satisfies the
proposed characteristic of the 5G core network. The benefit
of this approach is that by distributing control functions as
a children running close to UEs, the signaling load can
be processed and handled locally. Additionally, this reduces
the response time to control function requests. Moreover,
the parent control functions are unaffected by a amount of
signaling traffic that might be generated by large numbers
of UE, such as IOT devices. Finally, this approach reduces
the control plane’s bandwidth consumption, while adding
additional capacity to the core network to handle user plane
traffic.
Fig. 8 a flow chart describing how this solution works.
Assumes that fp
m is a control function M running at a control
node as a parent. Also assume that DC1 exists in or near
the BS that serves UE1. When “function M” is needed by
UE1, and if function Ms child (fc
m) exists close to BS for
UE1, then fc
m serves the UE’s request and updates fp
m [if
required]. If fc
m does not exist, then the request is routed
to fp
m at the main control node. Now fp
m serves the request
and launches fc
m close to BS (i.e., inside DC1). The primary
control node decides whether fm should execute as a child
(i.e., close to UE1) or not. This assume that there will
be collaboration between network management and cloud
management. There are ongoing activities such as EU FP7
Project UNIFY that addresses collaboration between the cloud
and network management [22].
V. SIGNALING LOAD ANALYSIS
This section gives a mathematical model that can be used
to calculate the control node(s) signaling load for a future

Fig. 8. De-centralized control plane function procedure.
mobile communication platform. For this purpose, we assume
that there are four control plane events that initiate signaling
control in LTE (as described earlier in SectionII-A) remains in
future 5G. Another simplifying assumption is that each IOT
device supports one application.
Let λo denotes the average number of originated session
by an IOT application per second and λt denotes the average
number of session terminating at an IOT device per second.
If λo and λt are independent events, the probability of IOT
device being active in a terminated session (PAT ) when the
IOT device wishes to send data is:
PAT = λt × γt (1)
Where γt is the average duration of originated session (in
seconds). Further, let “ρ” represents the density of IOT devices
per km2
, “A” represents coverage area by each cell (in km2
),
and “C” denotes the total number of cells in an operator
network’s. Additionally, introduce “K” a distribution factor for
each control function. If K = 1, then a central control function
in the network manages all related control events within a
network, while K = C denotes a fully distributed approach
where control functions coexist with each BS. The number
of message processed by a control function (i.e., messages
entering or leaving the control function) due to IOT originated
sessions is given by
Lo =
Mo × λo × ρ × A × C × (1 − PAT )
K
(2)
where Mo denotes number of control messages processed by
control function(s). It worth considering that when a UE is
in CONNECT mode in a UE-terminated session, there is no
signaling overhead to set up a connection for originating a
session to send data. Also, if a UE is in CONNECT mode in
a UE originated session, there is no signaling overhead to set
up a connection for receiving data. As a result, for a UE in
idle state number of messages processed by the control node
(entering or leaving) due to UE-terminated session is given by
Lt =
Mt × λt × ρ × A × C × (1 − PAO)
K
(3)
where Mt denotes the number of control messages processed
by control function(s) to set up a connection for UE-terminated
session and PAO denotes the probability that an IOT device
is connected in a UE-originated session when data arrives for
an IOT device. From Fig.3 Mt = Mo + Ca + 1 for a LTE
network where Ca represents the number of eNodeBs per
tracking area∗
. Also, based on the independent assumption of
λo and λt, PAO can be calculated as
PAO = λo × γo (4)
where γout is the average duration of originated session
(in seconds). The load due to session management can be
calculated as
Lsm = Lo + Lt (5)
To calculate control node load due to hand over classical
Fluid-Flow Mobility [23] is employed to estimate a UE
HO rate. Based on this model for a circular region with a
population density of ρ, an average velocity ¯v, and region
diameter of D, the average number of site crossings per unit
time is Navg = ρπD¯v. The centralized approach of HO
management function (i.e., K = 1) presents a scenario where
a central HO function in the network manages all HOs within
a network and no changing of HO management node occurs
during HO. In this case, the total number of message per hour
at the control function due to HO is given by:
Lh
K=1 = [Mcsg
ncc × Pcsg + Mncsg
ncc × (1 − Pcsg)] × NavgPA × C
(6)
where Pcsg is S-GW relocation probability, Mcsg
ncc denotes the
number of HO signaling messages processed by the control
function when there is no change in control node and no
change in S-GW, Mncsg
ncc denotes number of HO signaling
messages processed by the control function when there is
no change in control node but S-GW relocation is required,
and PA denotes probability of device being in CONNECTED
mode. PA can be well approximated by PAT + PAO.
When K = C we have full distribution of HO management
entity. Also, we assume that co-located with each BS is an
embedded data center to host a control function for HO
management. We also assume that the rate of UEs leaving
a BS equals to the rate of UEs joining a BS as a result of
uniform user density. In this procedure there is possibility that
∗We assumes the paging message will not be lost on the back-haul or air
interface and there is no need for paging re-transmission.

a UE stays with same S-GW or S-GW relocation also occurs
during the HO procedure. As a result, the load on the HO
control function is sum of load on HO management function
as a source control function (Lsc) (i.e., due to UEs leave the
cell) and the loads on HO management function as a target
control function (Ltc) (i.e., due to UEs joining a cell). These
are given by
Lh
K=C = [(Mcsg
Lsc
+ Mcsg
Ltc
) × Pcsg + (Mncsg
Lsc
+ Mncsg
Lsc
) × (1 − Pcsg)] × Navg × PA
(7)
where Mcsg
Lsc
denotes the number of HO signaling messages
of the source HO function when S-GW relocation is required
and Mncsg
Lsc
denotes the number of HO signaling messages of
the source HO control function when the UE maintains the
same S-GW. Additionally, Mcsg
Ltc
denotes the number of HO
signaling messages on the target HO control function when
S-GW relocation is required and Mcsg
Ltc
denotes the number
of HO signaling messages on the target HO control function
when the UE maintains the same S-GW.
When 1 < k < C the coverage region divides to K areas∗
where each area has its own HO management entity. In this
case, the signaling load on the HO management entity in each
area is the sum of the load due to HO inside that area (i.e., HO
without changing HO management function) and the load due
to HO between two areas (i.e., HO cases when the UE changes
HO management function). The signaling load on each HO
management function can be approximated as
Lh
1<K<C = Lh
K=1 × ( 1
K − C
K ) + Lh
K=C × C
K
(8)
Another event that produces signaling load is TAU. If
K represents the distribution factor of tracking area, then
Ca = C
K . In this case, the total number of signaling messages
generated on the network due to tracking area update is
approximated as
LT AU = [(Mcsm
T AU ×Pcsm+(Mcsm
T AU ×(1−Pcsm)]×Navg×
√
C ∗ K
(9)
where Mcsm
T AU denotes the number of signaling messages for
intra-session management TAU, Mcsm
T AU denotes the number
of messages for inter-session management TAU, and Pcsm
represents the probability of changing session management
function when changing tracking area region. Pcsm can be
approximated as K
K when K ≤ K and to 1 for other cases.
VI. NUMERICAL ANALYSIS
This section presents the numerical analysis of the control
signaling load for both centralized and de-centralized control
approach. The assumption in future 5G is that the number
of signaling message processed at control functions for each
controlling event is equal to messages proceed at an LTE MME
due to same events (as discussed in section II-A). In this
analysis we assumed a region size of 1000 Km2
, and λo =
∗We assumes coverage region divides to equal pieces.
3λt = 1 per minute. Additionally, based on 5G UE terminal
target capacity (i.e., 10 Gbps) and the assumption of only a
very small amount of data being transmitted by IOT device
(i.e., assumed 500 Kb for up-link and 1500 Kb per down-link)
session duration can be calculated as γt = 3×γo = 1.5e−06.
finally, we assumed the average speed of devices ¯v = 20 km/h.
Equations 2 and equation 3 shows that user density and
the number of BSs connected to one session management
function are two factors that effect session management load
due to UE originated and UE terminated scenarios. As a
result, in future 5G deployments with denser cell deployment
and high user density, the load on the session management
function is expected to increase. One approach for reducing
this load would be to decrease the area covered by a session
management function (i.e., to reduce the number of BSs
attached to a single session management function). This
reduction can be a factor of co-locating session management
functions at each BS, i.e., session management distribution
factor K = C (as we discussed in section IV). Fig. 9 shows
how user density and the distribution factor of the session
management function effect the session management load. In
this figure we assumed that the size of a tracking area was
Ca = 10.
Table I gives examples of the number of messages pro-
cessed at each session management function due to UE
originated and UE terminated scenario in units of millions
of messages per hour. It is worth mentioning that increasing
distribution factor of session management function will not
effect the total number of signaling message in the network.
The total number of signaling message generated in the
network for different values of ρ is shown in the first column
of Table I when the K = 1.
TABLE I. NUMBER OF MESSAGES PROCEED AT THE SESSION
MANAGEMENT FUNCTION BASED ON SAMPLE VALUE OF K AND ρ IN
UNITS OF MILLIONS OF MESSAGES PER HOUR.
Distribution factor K =
1 100 200 300
ρ =
1000 556 5.56 2.78 1.853
4000 2224 22.24 11.12 7.413
8000 4448 44.48 22.24 14.827
‘Inter HO management function” describes a scenario
Fig. 9. Session management function load per hours based on session
management distribution factor K and user density ρ. The ”Z” axis uses
logarithm scale.

Fig. 10. HO management load based on HO management distribution factor
K and user densityρ. The ”Z” axis uses a logarithm scale.
wherein after HO the UE changes HO management function,
while ‘intra HO management function” is an HO within a
single HO management function coverage area. As discussed
previously, inter HO management function scenarios produce
greater load than intra HO management function scenarios. On
the other hand, increasing UE density or dense deployment of
BSs will increase HO rate in the mobile network resulting
in higher load. In a fully distributed HO management function
approach, all HOs are inter HO management function, but each
HO management function handles fewer HO. Fig. 10 shows
how increasing user density and the distribution factor of the
HO function affects the HO management function loads.
Table II gives an example of the number of messages
processed at each HO management function and the total HO
signaling load generated on the network when the user density
ρ = 8000 UE/Km2
. From this table, we can observe that
increasing the distribution factor of HO management function
will reduce the load on each HO management function but,
at the cost of increasing the total signaling load generated
on the network. However, the total number of messages per
hour is negligible. The high capacity of future 5G networks
and low amount of data transmitted in each communication
results in low value of PA (i.e., 8.340e − 09 based on our
assumptions). The low value for PA indicates a low probability
of a device being in a connected state when HO occurs, hence
the signaling between UE and HO management function are
unneeded.
TABLE II. NUMBER OF MESSAGES PER HOUR PROCESSED AT EACH
HO MANAGEMENT FUNCTION AND THE TOTAL LOAD GENERATED AT
NETWORK DUE TO HO WHEN THE ρ = 8000 AND BASED ON SAMPLE
VALUES OF K IN UNITS OF MESSAGES PER HOUR.
Distribution factor K =
1 100 200 300
Load at each
HO management
function
1.318 0.031 0.019 0.015
Total load gener-
ated due to HO
in network
1.318 3.093 3.910 4.538
The tracking area information will be utilized by session
management entities when there is incoming data for an idle
UE. Tracking area size affects the load generated because of
the need to send TAU (i.e., when a UE ﬁnds itself in new
Fig. 11. Number of signaling message proceed at each session management
function due to TAU based on different distribution factor for session
management K and and tracking area K . The ”Z” axis are in logarithm
scale.
Fig. 12. Number of signaling message proceed at each session management
function due to TAU based on different distribution factor for tracking area
K when session management K = 150.
tracking area it must send a TAU) and also the signaling load
due to the UE terminated scenario. In this part of the analysis,
we assume that the agent for handling TAU is co-located with
the session management functions. From equation 9, it is clear
that increasing user density results in linear growth of load
due to TAU. There are two more factors that can effect on
TAU: distribution factor of session management functions (K)
and the distribution factor of the tracking area (K ). These
two factors determine the number of BSs in one tracking area
and the number of tracking areas supported by one session
management. Fig. 11 shows how these two factors affect the
number of messages processed because of TAU.
Table III gives an example of the number of messages
processed at each session management function due to TAU

in units of millions messages per hour. It is worth mentioning
that increasing K result in more tracking area but with smaller
size thus increasing TAU load as the user more frequent needs
to send a TAU. K < K means that one TA will be covered
by numbers of session management. As a result, the load of
TAU will split between those session management functions
that result in slow growth in TA management function when
K increase. K > K means that each session management
function cover several TAs. As a result, the load on each
session management function will be the sum of the load
generated by each TA, thus leading to a rapid growth in
session management load when K increase. Fig. 12 depicts
this behavior when the session management distribution factor
K = 150.
TABLE III. NUMBER OF MESSAGES PROCEED AT SESSION
MANAGEMENT FUNCTION DUE TO TAU BASED ON SAMPLE VALUE OF
SESSION MANAGEMENT DISTRIBUTION FACTOR K AND TA DISTRIBUTION
FACTOR K IN UNITS OF MILLIONS MESSAGES PER HOUR.
K =
1 100 200 300
K =
1 53.808 0.538 0.269 0.179
100 1506.622 5.381 2.690 1.794
200 2175.261 12.067 3.805 2.537
300 2688.325 17.198 6.370 3.107
On the other hand, increasing the distribution factor of TAU
(i.e., results in smaller Ca) decreases the load on each session
management functions due to UE terminated sessions. As a
result from a session management function point of view the
maximum value possible for the TA distribution factor results
in lower load due to UE terminated scenarios. Based on the
assumption of co-locating the session management function
and TA management function to find appropriate value for
TA distribution factor the aggregation of UE termination and
TAU signaling load should be considered. Table IV shows
the number of messages processed at the session management
function due to UE originated and UE terminated scenario
based on sample values of K and K .
TABLE IV. AGGREGATION OF MESSAGES PROCEED AT SESSION
MANAGEMENT FUNCTION DUE TO TAU AND UE ORIGINATED AND UE
TERMINATED BASED ON SAMPLE VALUE OF SESSION MANAGEMENT
DISTRIBUTION FACTOR K AND TA DISTRIBUTION FACTOR K IN UNITS OF
MILLION MESSAGES PER HOUR.
K =
1 100 200 300
K =
1 5 61.474 0.615 0.308 0.205
100 1507.985 5.394 2.698 1.799
200 2176.592 12.081 3.811 2.541
300 2689.647 17.211 6.377 3.111
Fig. 13 shows how changing K and K effects on the total
number of message processed on the network due to handling
TAU. It worth mentioning that increasing the distribution factor
of TAU increase the total load generated due to TAU on
network. The pattern of increase is the same as shown Fig.
12. On the other hand, increasing the session management
distribution factor decreases the total TAU load on the network.
When K = K , the total load generated on the network due to
TAU will remain constant. Table V gives examples of the total
number of messages generated on the network due to TAU.
A. Discussion
Two factors are relevant when we are debating the dis-
tribution factor value for each control function, i.e., the
TABLE V. TOTAL NUMBER OF SIGNALING MESSAGE GENERATED
WITHIN NETWORK BASED ON SAMPLE VALUE OF SESSION MANAGEMENT
DISTRIBUTION FACTOR K AND TA DISTRIBUTION FACTOR K IN UNIT OF
MEGA MESSAGE PER HOUR.
K =
1 100 200 300
K =
1 53.801 53.808 53.808 53.808
100 1506.622 538.080 538.080 538.080
200 2175.261 1206.718 760.959 760.959
300 2688.325 1719.783 1274.024 931.980
signaling load on each function and total load generated within
the network for each control events. Reducing the load at
each control function by distributing those functions enables
network operators to better scale their network in order to
manage the huge number of signaling messages generated by
different types of users in a future 5G network. The second
factor that we should consider is the total number of messaged
generated on the network. A more highly loaded network
indicates a need for more compute resources and capacity both
of which will increase operational costs.
Increasing the “session management function” distribution
factor results in a division of session management signaling
messages to more session management functions without
adjusting the total number of session management signaling
messages in the network. As a result, increasing this factor
when user density increases is a reasonable action.
The high capacity of future 5G networks and limited data
transmitted by IOT devices affects most of the HO scenarios as
the UE is likely to stay in idle mode. In this case, a low amount
of HO signaling is expected (i.e., even in a network with very
high UE density) as for the idle UEs no signaling messages are
required between the UE and the HO management function.
As a result, the distribution of HO management function does
not add considerable value for the network operators. future
IOT device that are not concerned with battery consumption
can choose ‘always-on” ∗
approach that raise signaling load
because of HO. On the other hand, for devices with ”always-
on” approach session management function signaling load
will decrease as there is no paging and session management
required. Further study is needed to analyze the side effect of
the always-on approach, but this is out of the scope of this
paper.
The TA distribution factor determines the size and number
of TAs covered by each session management. A larger value
for the TA distribution factor will result in lower load for UE
terminated scenarios, while increasing the aggregated load due
to the TAU on the network.
VII. CONCLUSION
This paper discusses a novel control plane architecture for
future 5G to provide a flexible and scalable network. The
proposed network architecture consists of independent control
function responsible for different control events in the network.
The control functions can leverage cloud technologies (e.g.,
network embedded cloud) and run in a virtualized environment
to implement a flexible and scalable control plane function
distributed within 5G mobile networks. This approach aims to
enable 5G networks to manage the enormous amount of control
∗This approach keeps continuously UEs in a connected state.

Fig. 13. Total number of signaling message proceed at network due to TAU based on different distribution factor for tracking area K and distribution factor
fo session management K.
signaling produce by IOT and MTC. We utilized 3GPP LTE
to identify the required control plane events, and we assumed
that 5G takes evolutionary path from LTE. We have shown that
moving the control plane session management function close
to UEs at a data center co-located with the BS is beneficial in
terms of control signaling load management.
REFERENCES
[1] A. M. Mousa, “Prospective of Fifth Generation Mobile communica-
tions,” International journal of next-generation computing (IJNGC),
vol. 4, no. 3, eptember 2012.
[2] A. Osseiran, V. Braun, T. Hidekazu, P. Marsch, H. Schotten, H. Tullberg,
M. Uusitalo, and M. Schellman, “The foundation of the mobile and
wireless communications system for 2020 and beyond: Challenges,
enablers and technology solutions,” in Vehicular Technology Conference
(VTC Spring), 2013 IEEE 77th, June 2013, pp. 1–5.
[3] E.Dahlman, G.Mildh, S.Parkvall, J.Peisa, J.Sachs, and Y.Selen, “5G
Radio Access,” Ericsson review, vol. 91, 2014.
[4] 3GPP, “Evolved Universal Terrestrial Radio Access (E-UTRA) and
Evolved Universal Terrestrial Radio Access (E-UTRAN); Overall
description; Stage 2,” 3rd Generation Partnership Project (3GPP), TS
36.300, 09 2014. [Online]. Available: http://www.3gpp.org/DynaReport/
36300.htm
[5] 3GPP, “Evolved Universal Terrestrial Radio Access Network (E-
UTRAN); X2 general aspects and principles,” 3rd Generation
Partnership Project (3GPP), TS 36.420, 06 2014. [Online]. Available:
http://www.3gpp.org/DynaReport/36420.htm
[6] 3GPP, “Evolved Universal Terrestrial Radio Access (E-UTRA);
User Equipment (UE) procedures in idle mode,” 3rd Generation
Partnership Project (3GPP), TS 36.304, Sep. 2014. [Online]. Available:
http://www.3gpp.org/dynareport/36304.htm
[7] G. P. Fettweis, “A 5G Wireless Communications Vision,” Microwave
Journal, pp. 24–36, December 2012.
[8] R. Baldemair, E. Dahlman, G. Fodor, G. Mildh, S. Parkvall, Y. Selen,
H. Tullberg, and K. Balachandran, “Evolving wireless communications:
Addressing the challenges and expectations of the future,” Vehicular
Technology Magazine, IEEE, vol. 8, no. 1, pp. 24–30, March 2013.
[9] Mobile and wireless communications enablers for the twenty-
twenty information society, METIS. [Online]. Available: https:
//www.metis2020.com/
[10] I. Albarran Munoz and M. Parras Ruiz De Azua, “Telecommunication
services migration to the cloud : Network performance analysis,”
Master’s thesis, KTH, Communication Systems, CoS, 2012,Available:
http://kth.diva-portal.org/smash/get/diva2:524296/FULLTEXT01.pdf.
[11] G. Caryer, T. Rings, J. Gallop, S. Schulz, J. Grabowski, I. Stokes-
Rees, and T. Kovacikova, “Grid/cloud computing interoperability,
standardization and the Next Generation Network (NGN),” in
Intelligence in Next Generation Networks, 2009. ICIN 2009. 13th
International Conference on, Oct 2009, pp. 1–6.
[12] A. Roozbeh, A. Sefidcon, and G. Maguire, “Resource Monitoring in
a Network Embedded Cloud: An Extension to OSPF-TE,” in Utility
and Cloud Computing (UCC), 2013 IEEE/ACM 6th International
Conference on, Dec 2013, pp. 139–146.
[13] P. Rost, C. Bernardos, A. Domenico, M. Girolamo, M. Lalam,
A. Maeder, D. Sabella, and D. Wubben, “Cloud technologies for flexible
5G radio access networks,” Communications Magazine, IEEE, vol. 52,
no. 5, pp. 68–76, May 2014.
[14] S. Hossain, “5G Wireless Communication Systems,” American Journal
of Engineering Research (AJER), vol. 02, pp. 344–353, 2013, e - ISSN
: 2320 - 0847.
[15] A. J. Pratik Sule, “Architectural Shift from 4G to 5G Wireless Mobile
Networks,” International Journal of Computer Science and Mobile
Computing, vol. 03, pp. 715–721, 2014, iSSN : 2320 088X.
[16] B. Emmerson, “M2M: The Internet of 50 Billion Devices,” WinWin
Magazine, pp. 19–22, Jan 2010, Available: http://m2m.com/docs/
DOC-1043/.
[17] 3GPP, “Service requirements for Machine-Type Communications ;
Stage 1,” 3rd Generation Partnership Project (3GPP), TS 22.368, 06
2014. [Online]. Available: http://www.3gpp.org/DynaReport/22368.htm
[18] 3GPP, “Feasibility Study on remote management of USIM application
on M2M equipment,” 3rd Generation Partnership Project (3GPP), TR
33.812, 06 2012. [Online]. Available: http://www.3gpp.org/DynaReport/
33812.htm
[19] J. Wang, J. Wen, J. Zhang, and Y. Han, “A demonstration of a new
TCP congestion control algorithm over LTE and other challenging
networks,” in Workshop on Wireless of the Students, by the Students,
for the Students, ser. S3 ’10 Workshop, 2010.
[20] K.-R. Jung, A. Park, and S. Lee, “Machine-Type-Communication
(MTC) Device Grouping Algorithm for Congestion Avoidance of
MTC Oriented LTE Network,” in Security-Enriched Urban Computing
and Smart Grid, ser. Communications in Computer and Information
Science, T.-h. Kim, A. Stoica, and R.-S. Chang, Eds. Springer
Berlin Heidelberg, 2010, vol. 78, pp. 167–178. [Online]. Available:
http://dx.doi.org/10.1007/978-3-642-16444-6 22
[21] 3GPP, “System improvements for Machine-Type Communications
(MTC),” 3rd Generation Partnership Project (3GPP), TR 23.888, 09
2012. [Online]. Available: http://www.3gpp.org/DynaReport/23888.htm
[22] A. Csaszar, W. John, M. Kind, C. Meirosu, G. Pongracz, D. Staessens,
A. Takacs, and F.-J. Westphal, “Unifying Cloud and Carrier Network:
EU FP7 Project UNIFY,” in Utility and Cloud Computing (UCC), 2013
IEEE/ACM 6th International Conference on, Dec 2013, pp. 452–457.
[23] D. Lam, D. C. Cox, and J. Widom, “Teletraffic modeling for personal
communications services,” Comm. Mag., vol. 35, no. 2, pp. 79–87,
Feb. 1997. [Online]. Available: http://dx.doi.org/10.1109/35.565664

Conference Paper: Distributed cloud and de-centralized control plane A proposal for scalable control plane for 5G

More Related Content

What's hot (20)

Similar to Conference Paper: Distributed cloud and de-centralized control plane A proposal for scalable control plane for 5G (20)

More from Ericsson (20)

Recently uploaded (20)

Conference Paper: Distributed cloud and de-centralized control plane A proposal for scalable control plane for 5G