17 HPSR Economic Analysis 5g Superfluid
17 HPSR Economic Analysis 5g Superfluid
17 HPSR Economic Analysis 5g Superfluid
Abstract—We target the evaluation of a Superfluid 5G network [7]. In particular, RFBs can be arbitrarily decomposed in other
from an economic point of view. The considered 5G architecture RFBs, while VNFs in the ETSI model cannot be decomposed
has notably features, such as flexibility, agility, portability and in other VNFs. Moreover, the RFBs can be mapped into
high performance, as shown by the H2020 SUPERFLUIDITY
project. The proposed economic model, tailored to the Superfluid different SoftWare (SW) and HW execution environments
network architecture, allows to compute the CAPEX, the OPEX, (see [6]), while the ETSI model focuses on mapping VNFs
the Net Present Value (NPV) and the Internal Rate of Return into Virtual Machines (or Containers) in traditional cloud
(IRR). Specifically, we apply our model to estimate the impact infrastructures.1
for the operator of migrating from a legacy 4G to a 5G network. In this context, several questions are arising such as: Is the
Our preliminary results, obtained over two realistic case studies
located in Bologna (Italy) and San Francisco (CA), show that 5G Superfluid network sustainable from an economic point of
the monthly subscription fee for the subscribers can be kept view? What is the cost of upgrading current 4G cells sites
sufficiently low, i.e., typically around 5 [USD] per user, while to support 5G services? What is the monthly subscription
allowing a profit for the operator. fee to be set in order to guarantee a net profit for the
operator? The goal of this paper is to answer these questions.
I. I NTRODUCTION In particular, we derive an economic model of the Superfluid
According to different studies (see e.g. [1]), the traffic in network architecture firstly proposed in [6]. We then evaluate
cellular networks is going to notably increase in the forthcom- the model over two representative scenarios. Our preliminary
ing years, due to a constant rise of the number of subscribers, results, obtained over two realistic case studies, demonstrate
their mobility, and the very high bandwidth required by future the feasibility of the proposed solution. Specifically, we show
applications, such as tactical Internet and high definition video that the architecture becomes profitable for the operator when
services. In this context, current 4G network infrastructure may the monthly subscription fee per user is higher or equal than 5
not be able to face the aforementioned challenges. To solve this [USD], thus making the Superfluid solution a viable approach
issue, operators and researchers have started the investigation from an economic point of view.
of new technologies, under the umbrella of 5G, which are Focusing on the related work, the cost modeling for an
expected to turn into reality by 2020. Such technologies SDN/NFV based mobile 5G network is proposed in [8].
include, e.g., the massive adoption of Multi User Multiple However, no indication about the subscription fee that the user
Input Multiple Output (MU-MIMO) [2], cell densification [3], should pay is provided. Moreover, the evaluation is conducted
softwarization of network devices [4], and the possibility to only over synthetic cases, and not realistic ones like in this
exploit commodity HardWare (HW) to run virtual resources work. The work in [9] details the life-cycle cost modelling
[5]. for NFV/SDN based mobile networks. However, the analysis
Among the different projects focusing on 5G, SUPERFLU- considers 4G, not 5G, networks. The total cost of ownership
IDITY [6] (funded by the EU through the H2020 program) for a 5G network is instead investigated in [10]. However, the
targets the design of a flexible, agile, portable and high setup of the physical HW devices, as well as the possibility
performance 5G network architecture. The core of the project to use softwarized elements, are not considered. Finally, the
is the introduction of a Superfluid approach in which both ser- closest paper to our work is [11], in which the authors have
vices and network functions are decomposed into softwarized performed a preliminary evaluation of a Superfluid network
components, named as Reusable Functional Blocks (RFBs), under different Key Performance Indicators (KPIs) over a
which are deployed on top of physical 5G nodes. RFBs have simple synthetic scenario. In this work, we go four steps
notable features, including: i) RFBs chaining to realize more further by introducing the following original contributions:
complex functionalities and provide the service required by • defining an economic model to estimate the costs and the
users; ii) platform independence via softwarized functions profits achieved by the operator;
that can be run on different HW equipments; and, iii) high • considering two representative case studies from realistic
performance and high flexibility, by deploying the RFBs where
1 In this work we focus on RFBs types that can be mapped in VNFs
and when they are really needed (hence the Superfluid attribute
of the ETSI model. The evaluation of other RFBs features (such as the
of the architecture). The RFB concept is a generalization of the decomposition of RFBs in smaller RFBs, and the mapping of RFBs to different
Virtual Network Function (VNF) concept proposed by ETSI software environments) will done as future work.
EPC
Macro Cell
In this context, the Superfluid vision aims to move from
Small Cell the current architectural approaches, which are based on
monolithic network components/entities and their interfaces, to
a solution where network functions can be programmatically
5G-Nodes composed using RFBs, which are dynamically deployed in
the 5G nodes, allowing a continuous real-time optimization
of the network. More in detail, the main idea behind the
RFB concept is the decomposition of high-level monolithic
functions into reusable components. A RFB is a logical
entity that performs a set of functionalities and has a set
(a) Physical Infrastructure
of logical input/output ports. In addition, the RFB concept
MU-MIMO is used to model the allocation of service components to an
Radio Links
execution platform, with the proper mapping of resources.
Base Band
User
Traffic Signal User “A”
We refer to this approach as RFB Operations. On the other
MEC BBU RRH hand, it is also used to explicitly model the composition of
User “B” RFBs to realize a service or a component. We refer to this
(b) RFB Connections approach as RFB Composition. More in detail, the RFBs can
be composed in graphs to provide services or to form other
Fig. 1. 5G Superfluid architecture. RFBs (therefore a RFB can be composed of other RFBs). The
RFB (de)composition concept is applied to different execution
environments in which the RFB can be deployed and executed.
4G deployment, the first one located in Bologna (Italy), All these features make the RFB concept a generalization of
and the second one in San Francisco (CA); the classical VNF approach, which is instead based on a rigid
• evaluating the impact of upgrading the existing 4G cell
mapping (which can not be further decomposed) in Virtual
sites to 5G ones; Machines run over a traditional cloud infrastructure.
• evaluating the impact of expanding the current set of
Thanks to the fact that the RFBs are fully virtualized
deployed cells to meet the 5G service requirements.
resources, they can be dynamically moved across the nodes
The rest of the paper is organized as follows. Sec. II reports to satisfy the KPIs of the network operator. An RFB performs
the 5G Supefluid network architecture description. Sec. III specific tasks in the network architecture, such as processing
details our economic model. The considered scenarios and the HD video to users, or performing networking and physical
the setting of the different components of the architecture layer tasks. In addition, each RFB consumes an amount
are described in Sec. IV. Sec. V reports the results of the of physical resources on the hosting 5G node. As physical
economic analysis over the considered scenarios. Finally, resources we consider the processing capacity (that will be
Sec. VI concludes our work. simply referred to as capacity in the rest) and the memory
occupation (in short referred to as memory).
II. 5G S UPERFLUID N ETWORK A RCHITECTURE
Similarly to [11], we consider the following RFBs types:
We briefly review the 5G Superfluid network architecture.
We refer the reader to [11], [6] for a more detailed description. • Mobile Edge Computing (MEC) RFB;
In brief, the 5G Superfluid network model is composed of a set • Base Band Unit (BBU) RFB;
of nodes, a set of links and a set of users. The nodes are used • Resource Radio Head (RRH) RFB.
to deploy either Small Cells (SCs), Macro Cells (MCs), or to More in depth, each RRH RFB performs physical layer
realize the core network elements of the so called Evolved operations. In particular, the RRH module handles a set of
Packet Core (EPC). Each node is connected to the rest of the Radio Frequency (RF) channels with users and the correspond-
network by means of a path of physical links. Moreover, each ing baseband channels with the BBU RFB. The amount of
node is composed of Dedicated HardWare (DHW) equipment resources required by an RRH RFB depends on the type of
and Commodity HardWare (CHW). In addition, each user can deployed cell (denoted with “Type 1” for a MC and “Type
be connected to the network by means of a cell. For simplicity, 2” for a SC). The operations required by the RRH RFB are
the EPC elements are collapsed into a single site in our model. run on DHW equipment of the node. Focusing on the BBU
Fig. 1(a) reports an example of the considered physical RFB, this module acts as an interface between the RRH RFB
system infrastructure, which is composed of different SCs, one and the MEC one. Specifically, the BBU RFB exchanges a
MC and one EPC site. In this scenario, each site corresponds baseband signal with the RRH RFB, and an amount of IP
to a 5G node. The figure reports also the coverage areas of the traffic with the MEC module. Its functions are run on both the
cells (which are represented by hexagonal layouts for the sake DHW and CHW elements installed in the node. In our work,
of simplicity). The service area, i.e., the area where the users we assume two types of BBU RFB. Specifically, a BBU RFB
are located, is assumed to overlap with the coverage area of of Type 1 is able to serve an RRH RFB of Type 1, while a
the macro cell. BBU RFB of Type 2 is able to serve an RRH RFB of Type 2.
D-HW C-HW where CAP EX M C and CAP EX SC are the total CAPEX
BBU BBU for deploying the Macro Cells (MCs) and Small Cells (SCs),
BaseBand Processing respectively. CAP EX M C is defined as:
Tasks Tasks
CAP EX M C = NCM C C CHW −M C + C DHW −M C
"
[USD]
(2)
RRH MEC
where NCM C is the number of MCs, C CHW −M C is the com-
modity HW cost for an MC, and C DHW −M C is the dedicated
Fig. 2. RFB mapping on CHW and DHW equipment of a 5G node.
HW cost for an MC. Similarly, we define CAP EX SC as:
CAP EX SC = NCSC C CHW −SC + C DHW −SC
"
[USD]
(3)
Finally, each MEC RFB is responsible for providing the HD where NCSC is the number of SCs to be deployed, C CHW −SC
video distribution service to the users. A practical example of is the commodity HW cost for a SC, and C DHW −SC is the
a MEC RFB is a cache serving a set of videos to users.2 In dedicated HW cost for a SC.
general, this module is able to serve a large amount of traffic, Focusing on the OPEX, we consider the yearly electricity
and consequently a subset of the users spread over the service consumption of the cells and the yearly scheduled maintenance
area. The MEC functionalities are run on CHW installed in operations costs (which may include, e.g., HW/SW upgrades).
the node. In particular, the total electricity costs depend on the power
Focusing then on the interactions between the RFBs, these consumption of the cells, which varies based on the current
modules are organized in logical chains. Specifically, each load ρt . In our work, we assume that the time is discretized
MEC RFB is logically connected to a BBU RFB, which, in in time slots, with a fixed duration, denoted with δt . The
turn, is connected to an RRH RFB and consequently to a set load of each cell ρt in each time slot varies then between
of users. Fig. 1(b) reports an example of RFBs chain and the a minimum and a maximum value, denoted with ρmin and
exchanged information between the modules and the users. In CHW −SC
ρmax , respectively. In addition, let us denote with Pmax
addition, the connection between a pair of RFBs in the chain CHW −SC
and Pmax the power consumption of the CHW and
can be direct, i.e., both RFBs are located on the same physical DHW equipment when ρt = ρmax , respectively. Similarly, let
5G node, or indirect, i.e., the RFBs are located on two separate us denote with Pmin CHW −SC CHW −SC
and Pmin the MC power
nodes. In this latter case, the information flows on an external consumption when ρt = ρmin . The total yearly electricity
physical link. Finally, RRH RFBs are able to setup a radio costs C E−M C of the MC at year i can be expressed as:
link with users, by exploiting the Multi User Multiple Input X ρ −ρ
E−M C MC MC t min MC
Multiple Output (MU-MIMO) technology. Ci = Pmax − Pmin · + Pmin ·δt ·c [USD]
t
ρmax − ρmin
We now consider the placement of the RFB over the (4)
MC
" CHW −M C CHW −M C
MC
Superfluid architecture. In this context, RRH RFBs can be where
" CHW −MP max = P max + P max , P min =
C CHW −M C
placed only in nodes connected to the antennas of the Radio Pmin + Pmin , and c is the hourly electricity
Access Network (RAN). On the contrary, BBU RFBs can cost for one [kWh] of energy. In a similar way to Eq. 4, we
be pooled in other nodes (i.e., by exploiting the Cloud-RAN define CiE−SC as the total yearly electricity costs of the SCs
paradigm). Finally, MEC RFBs can be potentially deployed in (not reported here due to the lack of space).
every node of the network. The total OPEX of the MCs for year i is then defined as:
Finally, Fig. 2 reports a scheme of a 5G node, including the
OP EXiM C = NCM C (CiE−M C + CiM −M C ) [USD] (5)
CHW and the DHW equipment. Specifically, the node in the
example hosts one MEC RFB in the CHW, one RRH RFB where CiE−M C is computed from Eq. 4, and CiM −M C is the
in the DHW and one BBU RFB split between the CHW and yearly scheduled maintenance cost for a MC. In a similar way,
DHW equipment. we define the total OPEX of the SCs as:
III. E CONOMIC M ODEL OP EXiSC = NCSC (CiE−SC + CiM −SC ) [USD] (6)
In order to assess the overall profitability of the considered where CiE−SC is again computed as Eq. 4 (with the SC
architecture, we need to compute the different costs and parameters), and CiM −SC is the yearly scheduled maintenance
profits experienced by the operator. Specifically, the costs are cost for a SC. The total OPEX for the operator is simply the
divided into CAPEX and OPEX components. More in depth, sum of the previous two terms:
the CAPEX represents the initial investment, which can be
OP EXiT OT = OP EXiM C + OP EXiSC [USD] (7)
expressed as:
In addition, the operator earns profits from the users.
CAP EX T OT = CAP EX M C + CAP EX SC [USD] (1)
Specifically, we assume that each user has to pay a monthly
2 MEC functionalities are going to be implemented in current cellular subscription fee, denoted with F . The total yearly profit
networks, thus bringing notably advantages. For example, the work in [12] P ROF ITiT OT from users is then denoted as:
shows that MEC may trigger a high utilization of server resources, coupled
with low latency times. P ROF ITiT OT = NUT OT · 12F [USD] (8)
We then denote the net cash flows experienced by the 44.56
operator on each year as: Macro Cell
( Small Cell
−CAP EX T OT if i = 0 44.54
CFi =
P ROF ITiT OT − OP EXiT OT if 0 < i ≤ L
44.52
(9)
Latitude
where L [years] is the lifetime of the architecture. 44.5
Given the knowledge of CFi , a useful metric to assess the
economic benefit is the Net Present Value (NPV), which is 44.48
defined as:
L
X CFi 44.46
NPV = [USD] (10)
i=0
(1 + η)i
44.44
11.25 11.3 11.35 11.4
where η is the discount rate, i.e., the return (in percentage) that Longitude
could be earned with a classical financial investment (such as
(a) Bologna Deployment
bank funds, loans, etc.). In particular, if the NPV is larger than
0, then the operator has a monetary advantage in building the
5G architecture rather than choosing to invest the money in 37.81 Macro Cell
financial activities. Small Cell
Finally, another metric to assess the deployment efficiency 37.8
is the Internal Rate of Return (IRR), which is defined as the
37.79
discount rate η ∗ making the NPV equal to 0. More formally,
Latitude
Users Cells
Area A
T OT MC SC T OT
Residents Visitors Total (NU ) NC NC Total (NC )
2 2 2
Bologna 123 [km ] 2413 [1/km ] 19 [1/km ] 299136 56 24 80
San Francisco 36 [km2 ] 7124 [1/km2 ] 600 [1/km2 ] 278064 1027 602 1629
TABLE IV
usually employ directional, sectorial antennas, which we place DHW F EATURES
at one of the vertices of the convex area, chosen to minimize
Feature Small Cell (SC) Macro Cell (MC)
the average distance between the base station and its served
Max. Processing 240 Gbps 640 Gbps
users. Capacity
Fig. 3(b) reports the obtained 4G deployment. Interestingly, DHW −SC
Pmax =1180 [W] DHW −M C
Pmax =2180 [W]
Power Consumption
in this case a very large number of 4G MCs and SCs is DHW −SC
Pmin =145 [W] DHW −M C
Pmin =269 [W]
deployed. Similarly to the Bologna case, SCs tend to be CAPEX cost C DHW −SC
=9828 [USD] C DHW −M C =29484 [USD]
deployed in the area with the highest user density, i.e., the
one encompassing most of offices and commercial buildings.
NPV [USD]
7
3 10
Bologna
2 6
10 Bologna − Upper Bound
San Francisco
1 San Francisco − Upper Bound
5
10
0 0 2 4 6 8 10 12 14 16 18 20
Bologna Bologna − Upper Bound San Francisco San Francisco − Upper Bound Monthly Subscription Fee [USD]
Scenario
IRR [%]
6 2
10
Bologna
4
1
10 Bologna − Upper Bound
San Francisco
2 San Francisco − Upper Bound
0
10
0 2 4 6 8 10 12 14 16 18 20
0
Bologna Bologna − Upper Bound San Francisco San Francisco − Upper Bound Monthly Subscription Fee [USD]
Scenario
(b) IRR
(b) Yearly OPEX
Fig. 5. Net Present Value (NPV) and Internal Rate of Return (IRR) over the
Fig. 4. Total CAPEX and yearly OPEX breakdown over the different different scenarios.
scenarios.