highlighting the advantages brought by MEC, also including data

aggregation and compression techniques.

Performance Evaluation of a
I. INTRODUCTION
Satellite Communication-Based
When looking at the use of the Internet, the landscape
MEC Architecture for IoT of applications and communication protocols has changed
substantially in the last years. Nonreal-time information
Applications exchange represented most of the Internet traffic in the past,
with predominance in the use of web technology and file
transfers (e.g., HTTP, e-mail, FTP, and peer-to-peer solu-
tions). Nowadays, real-time interactions are fundamental
for a whole new set of applications, from web gaming to
MICHELE LUGLIO VoIP and chat services, from conferencing platforms to
University of Rome “Tor Vergata,”, Rome, Italy cloud and Internet of Things (IoT) services. In particular,
massive IoT applications show a relevant growth and are
MARIO MARCHESE , Senior Member, IEEE
already representing a significant portion of Internet traf-
FABIO PATRONE , Member, IEEE
fic, as discussed for instance in the latest Cisco Annual
University of Genoa, Genoa, Italy
Internet Report [1]. Many kinds of sensor-based services
CESARE ROSETI fall in this category and are applied in a set of scenarios,
FRANCESCO ZAMPOGNARO such as home automation, smart cities, support to logistics,
University of Rome “Tor Vergata,”, Rome, Italy wearable health devices, industrial automation, environ-
mental sensing, and critical infrastructures monitoring and
protection. In all these environments, it is often important
New scenarios and use cases are raising following the birth of the
to consider large coverage and high mobility requirements.
fifth generation of mobile communications. The Internet of Things An example may be a sensor network which spans across
(IoT) is one of the main use cases which are growing, leading to a many administrative domains and covers remote areas.
massive amount of data that need to be exchanged throughout the A massive growth and spread of devices, such as sensors
Internet. Satellite communication networks are essential in remote and and actuators, has been envisioned for 2021 and beyond,
isolated environments and can support fully connected environments
by offloading the terrestrial infrastructure concerning delay–tolerant
with a consequent big mass of data that need to be collected,
traffic flows. However, satellite network resources are limited and processed, and, in some cases, displayed to users remotely
expensive, so they need to be carefully used in order to avoid waste connected to the Internet.
and satisfy the required user performance. The multi-access edge Satellite-based approaches, such as the space–air–
computing (MEC) concept can be exploited in this context to allow ground integrated networks [2], are considered as viable
data preprocessing at the edge, i.e., close to the users, so reducing
the amount of data that has to traverse the backhaul satellite link
solutions thanks to their ubiquitous coverage and broad-
and, in some cases, reducing data delivery times. This article analyses cast/multicast capabilities. Considering the IoT context and
the performance of a satellite architecture in the IoT framework the associated connectivity requirements, satellite access
can play a significant role in enabling IoT services [3].
Satellite communications (SatCom) networks are partic-
Manuscript received 27 October 2021; revised 2 May 2022; accepted 5
ularly suitable to offer connectivity to a high number of
August 2022. Date of publication 17 August 2022; date of current version
11 October 2022. devices located in wide areas, complementing terrestrial
infrastructure thanks to the large coverage. They can support
DOI. No. 10.1109/TAES.2022.3199330
cross-the-borders IoT scenarios properly managing their
Refereeing of this contribution was handled by F. Granelli. available network resources to guarantee the required qual-
This work was supported by ESA through VIBeS project under Grant ity of service (QoS) [4]
4000122991/18/U.K./ND. In particular, Cioni et al. [3] emphasize the role of satel-
lites in the foreseen spread of IoT devices and applications
Authors’ addresses: Michele Luglio, Cesare Roseti, and Francesco Zam-
pognaro are with the Electronics Department of the University of in the future 5G ecosystem focusing on the challenges re-
Rome “Tor Vergata,” 00133 Rome, Italy, E-mail: ([email protected]; lated to the massive machine-type communication (mMTC)
[email protected]; [email protected]); Mario Marchese service provision, air interface, and network aspects related
and Fabio Patrone are with the Electrical, Electronics and Telecommu- to the system design. Energy efficiency is another important
nication Engineering and Naval Architecture Department (DITEN), Uni-
aspect to take into account. A network coding (NC)-based
versity of Genoa, 16124 Genoa, Italy, E-mail: ([email protected];
[email protected]). (Corresponding author: Fabio Patrone.) approach has been proposed in [5] to properly manage the
available network resources and offer reliable IoT multicast
services in a 5G/satellite network. Key network challenges
This work is licensed under a Creative Commons Attribution 4.0 License.
For more information, see https://creativecommons.org/licenses/by/4.0/
for satellite exploitation in future 5G and IoT satellite-based
sensor networks are also presented in [6].

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 58, NO. 5 OCTOBER 2022 3775
 Assuming a massive amount of data generated by The rest of this article is organized as follows.
an IoT sensor network composed of thousands of IoT Section II highlights the roles that SatCom networks can
devices, the overall volume of raw data can lead to the have in the IoT applications. Section III presents the
rise of several issues. An example is the security and data used MEC-enabled satellite architecture and the considered
privacy issue, which involves both legal [7] and technical network configurations. The obtained results are reported in
matters [8] related to different aspects, such as how the Section IV. Finally, Section V concludes this article.
network is managed and how data are stored and processed.
Another example is related to how the network resources
are efficiently exploited to offer the required service and II. ROLE OF SATELLITES IN IOT APPLICATIONS
avoid unpleasant situations, such as the saturation of the Satellites can have two main alternative roles in a typical
satellite link bandwidth. satellite/terrestrial integrated network for IoT applications:
The presence of multiaccess edge computing (MEC) direct-access (fronthaul) or backhaul.
functionalities close to the sensor network allows saving In the first case, sensors are directly linked to the satel-
SatCom resources and offering enhanced performance to lites, which are the first hop of the communication path
the users thanks to smart management of large set of data between IoT devices and users. Third Generation Partner-
by enabling local processing and in-network functions (i.e., ship Project defined six different reference scenarios [24]
microservices) tailored on the specific use case. changing the altitude of the considered satellites (GEO and
The MEC concept, formerly mobile edge computing, non-GEO), the complexity of the carried satellite payload
has been introduced with the main aim to move the intelli- (transparent or regenerative), and the flexibility of the gen-
gence, or at least part of it, from the network core (central- erated beams (steerable or moving beams) for direct-to-
ized datacenters and cloud platforms) to the network edge satellite access (fronthaul). The most promising approaches
(distributed microdatacenters and edge platforms). MEC are mainly considering low earth orbit (LEO) and very low
contributes shaping the next generation network architec- earth orbit satellites, even if intrinsic challenges have still to
ture and infrastructure [9], [10], thanks to its applicability in be overcome, such as the heterogeneity of the IoT solutions
different scenarios and use cases, including IoT [11], [12]. and the high energy consumption of IoT devices. Some
Concerning MEC application in satellite networks, dif- ongoing activities are investigating these issues to assess the
ferent possible network architecture configurations, chal- feasibility of direct-to-satellite access, such as [25] and [26].
lenges, and open issues have been discussed [13]. Possible In the second case, satellites can be adopted as backhaul
solutions have been implemented and tested focusing on to the Internet or to a dedicated service center. Sensors’
the integration within the 5G environment [14], based on data are collected within an edge cloud in the terrestrial
different kinds of satellites (including nanosatellites [15]), access network and then made available through a satellite
and aimed at improving the user perceived QoS [16], [17]. data network backhaul by using IP-based protocols. High
IoT applications implementing satellite edge computing throughput satellites and very high throughput satellites
solutions are the object of [18], [19], [20], also considering GEO platforms operating at Ka or higher frequency bands
the severe resource limitations that affect IoT devices. are the main options. They can considerably reduce the com-
Latency and energy consumption reductions are the munication costs providing large amounts of bandwidths,
main aims of the studies [21], [22], [23], which also adopt making satellites a viable solution to enhance the overall
machine learning techniques to make resource management network capability.
more effective. The employment of communication satellites for IoT
This article addresses the integration of satellite and can be beneficial for multiple applications. Several solu-
terrestrial networks to interface IoT sensors and efficiently tions have been developed considering the possible different
enable IoT services through the Internet. The considered roles of satellites and tested in different application sce-
solution includes a satellite backhaul link based on a geosta- narios [27]. Satellites operating as backhaul support have
tionary earth orbit (GEO) satellite, to offload the terrestrial been considered for the smart city [28] and the road safety
network of delay-tolerant data, and the employment of the and autonomous ship scenarios [29]. Smart agriculture [30],
MEC paradigm, to achieve and optimize a fully integrated smart grid [31], and maritime IoT [32] are three other typical
and global IoT service, i.e., to offer an architectural and op- use cases where the use of satellites has been considered for
erational solution able to guarantee the typical performance IoT applications and has been proven to be effective, not
requirements of IoT applications worldwide, overcoming only for the lack of communication infrastructures based
limitations of different kinds, such as geographical and data on different technologies.
management on vast scale. This solution is assessed in order From the application viewpoint, data generated by de-
to quantify and highlight the performance improvement vices need to be transferred to receivers by using a suit-
obtained implementing the MEC concept to enhance the able protocol. Two main communication models can be
satellite transmission, by both performing data compression employed: “publish/subscribe” or “client/server,” depend-
to save satellite bandwidth, as well as distributing and ing on the use case. Although in principle, it is possible
aggregating data in an efficient way in relation of the domain to exchange such data in non-IP/proprietary formats and
where it is useful (i.e., local or remote). associated proprietary transport and application protocols,

3776 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 58, NO. 5 OCTOBER 2022
it is worthwhile to consider the IP-based approach. IP allows reference scenario to help collect data from the sensors.
enabling interoperability and reusability, leveraging well However, we decided to not include them since our main
established and freely available IP-based applications and focus is on the backhaul link which, especially for rural and
protocols. The two main IP-based solutions are Message remote areas, can be more realistically assumed as a satellite
Queuing Telemetry Transport (MQTT) and Constrained link, while UAVs can be more useful for the access link. Be-
Application Protocol (CoAP). sides, the presence of UAVs would not affect the application
MQTT [33] was originally an IBM proprietary protocol of our proposed MEC functionalities and the effect on the
now managed as a standard by the International Organiza- shown performance would be negligible, because the main
tion for Standardization. MQTT uses a publish/subscribe improvement comes from a local management of data.
model and requires an MQTT broker to manage and route Fig. 1 shows the considered reference network archi-
messages among MQTT nodes through point-to-point TCP tecture, which was validated within the European Space
permanent connections. MQTT has not been designed for Agency (ESA) funded project M2MSat.1 A large and
“constrained” devices, so power saving optimization and widespread set of IoT devices, called sensor equipment
tradeoffs among complexity/efficiency and bandwidth us- (SE), are the data sources of this sensor network scenario.
age are usually not considered. They are linked to a base station located in the remote area
CoAP [34] is a lightweight IP-based protocol for sensor through wired or wireless links. The base station aims to
data communication. It is part of the Constrained RESTful collect the data generated by the SEs in order to send them
Environments (CoRE) IETF working group and it aims to to the Internet through the satellite backhaul link. An IoT
realize a REpresentational State Transfer (REST) archi- cloud service, located on the Internet, collects and stores all
tecture in a suitable form for the most constrained nodes. SE data and offers data access to authorized users, called
CoRE is a very efficient RESTful protocol based on the user equipment (UE). UEs can be located on the Internet (on
classic client/server model. Even if DLTS, SMS, and TCP the right-hand side of the satellite link, called core side) or
bindings are defined as transport for CoAP, UDP-based connected directly from within the service provider network
implementations are the most commonly used. (on the left-hand side of the satellite link, called edge side).
Studies have been carried out to evaluate the employ- In this way, the satellite offers a backhaul solution that can
ment of MQTT and CoAP in the satellite backhaul case [35]. completely replace the terrestrial connectivity to the cloud
Efficient configurations for both MQTT and CoAP have services.
been highlighted and some modifications have been pro- It is assumed, as well for the reference service, that not
posed. For example, MQTT and CoAP employment to all sensor data have the same scope and priority. In fact,
support sensor data exchange on a satellite random access it is possible to consider several use cases in which some
channel based on the DVB-RCS2 standard has been inves- remote data collected into a remote area can just be used for
tigated in [36]. local data processing, whereas only alarms or downsampled
values are delivered toward a central node. Improve satellite
III. MEC-ENABLED IOT ARCHITECTURE resource management and computation offloading is the
main aim of most studies on this topic, such as [37] and [38].
A. Reference Architecture Description In this context, MEC can foster satellite bandwidth
The locations where MEC functionalities can be typ- reduction through the implementation of dedicated func-
ically implemented in satellite networks are three: on the tionalities, such as data aggregation and local processing at
ground close to the users, on-board satellites, and on the the edge, with a consequent lower cost due to the reduced
ground close to the satellite ground stations. Implement- amount of data that need to traverse the satellite link. The
ing MEC on-board satellites require regenerative payload proposed reference architecture may in fact include storage,
satellites with largely available computation and storage computation resources, and data local breakout functional-
resources and an internal structure designed to make their ities, which are part of the network operator infrastructure,
allocation flexible and scalable. Companies are planning to but not necessarily present at the edge of the network and
launch this kind of satellites, which will be available in the not enabled for MEC. For instance, thanks to the presence
near future. However, a first step to deploy and test satellite of network functions and components for the termination
MEC networks could be to exploit the already deployed of the IoT protocol stack within the network operator’s core
transparent payload satellites. It is important to quantify the network, a possible edge computing infrastructure and the
achievable performance improvement in different possible associated components can be conceived and enabled in
IoT use cases with different and variable network and a straightforward way. These dedicated components can be
traffic flow features, such as number and density of IoT optimized for IoT and deployed dynamically, e.g., as virtual
devices, distribution of generated data flows, and minimum network functions, allowing to manage and terminate the
performance requirements. local data-application protocols before the transmissions
We consider exploiting the MEC concept and a satellite over satellite, as discussed in the next section.
link to enable IoT scenarios in remote areas. In particular,
we focus on GEO-based communication systems, such as
Inmarsat, configured as backhaul. Unmanned aerial vehi-
cles (UAVs) could also be considered and included in the 1 [Online]. Available: https://artes.esa.int/projects/m2msat

LUGLIO ET AL.: PERFORMANCE EVALUATION OF A SATELLITE COMMUNICATION-BASED MEC ARCHITECTURE 3777

 Fig. 1. Reference hybrid network to enable IoT cloud services via satellite.

Fig. 2. IoT application level network configuration. Case 1: IoT MQTT native sensors, no edge computing. All MQTT clients access the core MQTT
broker, edge MQTT clients requests and responses need to traverse the satellite link.

B. Possible Network Configurations and MEC and core sides (edge and core MQTT clients, respec-
Operations tively), request data to the MQTT broker. Besides,
no additional functionalities dedicated to satellite
In relation to current standardization activities and on
bandwidth optimization and data preprocessing are
the basis of the state of the art and research projects [25],
included.
[39], [40], having Fig. 1 as a reference, we introduce three
2) Case 2 (see Fig. 3): SEs are IoT non-MQTT sensors.
possible configurations, leveraging data local breakout. The
They get access to the network through different
local breakout means that the operator network is supporting
interfaces (e.g., LoRaWAN) equipped with their own
the split of data sessions for the sensor data, delivering such
application layer protocols. MEC is partially added:
data in different configurations according to the enabled
an edge MQTT broker is introduced at the edge side
MEC options (if any). The goal is to present a progressive
with a twofold aim. On the one hand, to “convert”
performance improvement first due to the enabling of MEC,
the generated data in the MQTT format in order
and then the enabling of satellite-oriented ad hoc optimiza-
to let them be stored in MQTT brokers and, con-
tions, which are tailored on the data type and meaning. In
sequently, allow MQTT clients to access them. On
particular data compression, aggregation, suppression, and
the other hand, to provide local data storage at the
differential distribution will be introduced as edge opera-
edge side. To guarantee the compatibility with the
tions in the last configuration.
core MQTT broker, the edge MQTT broker has to
1) Case 1 (see Fig. 2): SEs are IoT MQTT sensors. Due be configured to act as a transparent proxy to the
to the large number of involved sensors and users, we target core MQTT broker, i.e., redirecting the traffic
consider the use of MQTT because it scales better to it after locally storing a copy of each content.
with a large population of nodes, thanks to a broker Edge MQTT clients can directly ask the edge MQTT
node which manages the data exchange between data broker to provide the interested contents so reducing
sources and users. Edge computing is not included in the amount of data that need to traverse the backhaul
this case. IoT MQTT sensors generate data destined satellite link in both directions: MQTT requests from
to the centralized MQTT broker (core MQTT broker) edge MQTT clients to the core MQTT broker and
located on the core side and hosted by cloud services. MQTT answers from the core to the edge side. Also
MQTT clients, which are located both on the edge in this case, no additional functionalities aimed at

3778 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 58, NO. 5 OCTOBER 2022
Fig. 3. IoT application level network configuration. Case 2: IoT non-MQTT sensors, edge MQTT computing in the IoT base station. Presence of an
edge MQTT broker to store information made available to the edge MQTT clients that, in this way, do not need to access the core MQTT and so
forward their request through the satellite link.

Fig. 4. IoT application level network configuration. Case 3: IoT non-MQTT sensors, edge MQTT computing in the IoT base station. Presence of an
edge MQTT broker as in Case 2 + additional edge functionalities to improve satellite bandwidth utilization.

satellite bandwidth utilization improvement and data

preprocessing are included. MEC functionalities are
limited to the presence of the edge MQTT broker.
3) Case 3 (see Fig. 4): SEs are IoT not-MQTT Sensors,
and the network architecture includes both the core
and edge MQTT brokers as in Case 2. The edge com-
puting component also enables a data compression
and data aggregation functionality. In detail, MEC
functionalities include also: 1) local processing, such
as downsampling and duplicated values suppression,
to perform data compression; 2) data aggregation,
such as aggregating different values by collecting
a given amount of messages over time in a single
Fig. 5. Testbed configuration by using Linux-based VMs.
MQTT message to reduce the protocol header over-
head. This type of MEC functionalities are already
proposed in [41]. As a step ahead with respect to this Machines (VMs), as shown in Fig. 5. VM3 acts as a router
work, we perform an accurate performance evalua- to emulate a GEO very small aperture terminal SatCom
tion by using real MQTT software implementation connection. The considered MEC functionalities are imple-
on a Linux-based testbed. mented as a set of Docker-based applications in VM2. The
Docker image deployed to enable the edge MQTT broker is
IV. PERFORMANCE EVALUATION based on the Mosquitto image.2 Additional configurations
A. Testbed Setup are added to the network in VM2 to ensure that MQTT

The testbed used to perform the validation over the

proposed scenarios is composed of a set of Linux Virtual 2 [Online]. Available: https://hub.docker.com/_/eclipse-mosquitto

LUGLIO ET AL.: PERFORMANCE EVALUATION OF A SATELLITE COMMUNICATION-BASED MEC ARCHITECTURE 3779

 TABLE I
Configuration Variables and Reference Values Used in the Tests

Fig. 6. Satellite link bandwidth utilization for Case 1: MEC is not

applied.

traffic can be transparently intercepted and locally pro- core MQTT broker and then go back to the edge MQTT
cessed by the edge MQTT broker without requiring sensor clients owing to the edge MQTT brokering functions; 2)
reconfigurations. data compression and aggregation functions. Compression
At the core side, the MQTT broker is configured in VM4 is very important for all real-time and very-low latency ser-
and acts as the reference backhand on the cloud to collect vices where sensor data must be quickly available avoiding
and dispatch sensor data to the clients, which can be located unnecessary data forwarding through the high-latency satel-
both at the edge side (VM1) and at the core side (VM5). lite link; aggregation is significant for delay-tolerant data.
Clients are coded in C by using open-source Mosquitto The interesting aspect is to quantify these improvements in
libraries. Sensors are modeled with scripts that generate data a realistic environment.
messages with uniform distribution and settable mean value We have performed real-time tests whose duration is
for data size and intergeneration interval. Data messages 10 min for each of the three cases defined in Section III and
from sensors then can be either directly included within configured as detailed in Table I. The two considered output
MQTT messages by the SEs (Case 1) or encapsulated in performance are given as follows.
MQTT messages by the edge MQTT broker before trans-
1) Message delivery time: The time elapsed between the
mitting them to the Cloud (Cases 2 and 3).
data message generation by the SE and its reception
In our tests, we define four different sensor categories:
by the interested MQTT client.
real-time, nonreal-time, local-only, and global sensors.
2) Satellite link bandwidth utilization, also called
Real-time sensors generate time critical data (e.g., alarms)
throughput: The amount of data that traverse the
that should be delivered as fast as possible without addi-
satellite uplink (from edge side to core side) and
tional processing. On the contrary, nonreal-time sensors
downlink (from core side to edge side).
provide data whose validity is acceptable also in the case of
deferred or partial delivery (e.g., a noncritical temperature In each test, the MQTT clients gradually join the net-
value). Both real-time and nonreal-time sensor data can be work, which leads to about 200 s of transient time before
of local-only type, i.e., they will be required only by edge reaching the regime values of the satellite link bandwidth
MQTT clients, or global type, i.e., they will be required by utilization. For each of the three cases, we show the overall
both edge and core MQTT clients. satellite link bandwidth utilization and the distribution of the
Table I reports the reference values of the configuration message delivery separate for edge and core MQTT clients.
variables. Such values are chosen to get realistic configura- The first test, referred to Case 1, allows introducing
tions. The basic reference is the IoT scenario reported in the a benchmark reference for the proposed application. The
Ericsson mobility report [42] with additional assumptions satellite link bandwidth utilization is reported in Fig. 6.
to implement the satellite scenario. We can see an almost symmetric use of the channel since
all sensor data must reach the only MQTT broker available
B. Assessing the Proposed Optimizations in the network (the core MQTT broker at the core side)
The obvious expectation is that enabling the edge and go back to the edge clients. The values for the average
MQTT broker as an edge function will reduce the amount throughput after the initial ramp time are 47.9 kb/s for the
of traffic forwarded through the satellite link. Enabling the uplink and 48.6 kb/s for the downlink. The throughput over
additional MEC functionalities should reduce even more time provides further insights on the traffic pattern: after
the impact of sensor traffic on the satellite link, due to: 1) the gradual entry of the sensors, we can see wide variations
the reduction of the amount of data that have to reach the that must be taken into account to design and dimension

3780 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 58, NO. 5 OCTOBER 2022
Fig. 7. Delivery time distribution of data destined to the edge and core Fig. 9. Delivery time distribution of data destined to the edge and core
MQTT clients for Case 1. MQTT clients for Case 2.
to about 50% (from 47.9 to 24.6 kb/s), while the number of
sensors is reduced to 32% (from 5000 to 3400, which are
both real-time and nonreal-time “global” sensors, since the
data generated by the “local-only” sensors do not need to
traverse the satellite link but are received and stored only by
the edge MQTT broker). This additional reduction is due
to the implementation of the MQTT bridge by Mosquitto,
and in particular, due to the edge MQTT broker that groups
more data messages in bulk before sending them to the core
MQTT broker through the MQTT bridge. The confirmation
of this behavior is given by the data delivery time distribu-
tion shown in Fig. 9.
In particular, for the core MQTT clients [see Fig. 9(a)],
the distribution is much shifted toward higher values (with
an average of 587 ms). This happened because sensor mes-
Fig. 8. Satellite link bandwidth utilization for Case 2: MEC is applied
sages are queued and aggregated with other messages at
introducing an edge MQTT broker to handle local data.
the socket/TCP level. On the other hand, data destined to
edge MQTT clients [see Fig. 9(b)] are now available locally,
real systems. Even if the sensors transmit at random times, significantly reducing the data delivery time to a few ms and
a nonflat channel utilization is obtained. providing the main desired benefit of MEC at the edge.
Fig. 7(a) and (b) shows the delivery time distribution of This second case shows how bandwidth reduction can
data requested by the core MQTT clients and by the edge be achieved by adding an MQTT MEC function without
MQTT clients, respectively. specific optimizations, at the cost of slightly increasing the
Looking at these figures, it is possible to appreciate core side data delivery time.
the impact of the satellite link on the time required to get Since some sort of socket-level aggregation was already
data from the MQTT broker without the help of any MEC performed in Case 2, enabling also specific data aggregation
functionality at the edge. The mean value of the delivery and compression mechanisms at the edge side (Case 3) leads
time is 325 ms from Fig. 7(a) and 868 ms from Fig. 7(b), to marginal advantages, lower than what initially expected.
where a satellite double hop is always required to get data. The obtained results are shown in Figs. 10 and 11.
When the local processing of MQTT messages is en- In this third case, the edge MQTT broker is aware of the
abled at the edge side through dedicated MEC functionali- real-time or nonreal-time nature of the global sensor data, as
ties (Case 2), the first aspect worth noting is the significant well as of the global or local-only scope of the messages. In
reduction of the satellite link utilization reported in Fig. 8. this way, it is able to perform selective additional process-
The downlink is now almost unused since local sensor ing. In detail, the data aggregation functionality involves
data are handled by the edge MQTT broker without the the aggregation of all nonreal-time data messages to be sent
need to access the core MQTT broker. The overall value through the MQTT bridge, received within a time window
of about 1 kb/s is related to the flow of ACK packets of fixed duration (1 s in our tests), in one single message,
destined to the sensors. However, concerning the uplink, the reducing the overall protocol header overhead. The data
reduction is not proportional to the reduction of the amount compression functionality includes: 1) discharge of dupli-
of transmitted sensor data. The uplink bandwidth is reduced cate values—if a sensor generates the same data multiple

LUGLIO ET AL.: PERFORMANCE EVALUATION OF A SATELLITE COMMUNICATION-BASED MEC ARCHITECTURE 3781

 delays and enqueuing times for the MEC processing and
MQTT are still present, and are interesting for a large-scale
deployment of the proposed system in real networks.
The following parameters have been changed, keeping
all the other parameters set as reported in Table I:
1) NCC = 100 (higher number of core MQTT clients);
2) NCE = 0 (no edge MQTT clients);
3) NS = 50 000 (10x sensors, keeping the same per-
centage of real-time and nonreal-time, global and
local-only sensors defined in Table I);
4) rl = nl = 0 (no local-only sensors).
Table II includes the obtained mean results of the ad-
ditional tests for each of the proposed MEC configura-
Fig. 10. Satellite link bandwidth utilization for Case 3: MEC is applied, tions. The mean results of the tests, already described in
including also data compression to reduce satellite traffic. Section IV-B, labelled as default, have been included too
for comparison.
Concerning the increase of NCC , there is no significant
change in the results compared to the default configura-
tion due to the publisher–subscriber nature of the MQTT
protocol. The amount of sensor data collected at the core
MQTT broker through the satellite link does not depend on
the number of core MQTT clients. Therefore, the results of
this configuration are not included in Table II.
Without edge MQTT clients interested in sensor data
(NCE = 0 and, consequently, no edge side data delivery
time results), there is a significant reduction of the satellite
downlink bandwidth utilization for Case 1. Since there is
no MEC processing, the absence of edge MQTT clients
reduces the amount of traffic in the core to edge direction
since there is no more requests. For Cases 2 and 3, there
is no significant difference in the satellite link bandwidth
utilization since all traffic at the edge is handled locally,
Fig. 11. Delivery time distribution of data destined to the edge and core
regardless of the number of interested edge MQTT clients.
MQTT clients for Case 3.
Besides, in Cases 2 add 3, the amount of traffic delivered
through the MQTT bridge from the edge to the core side is
times, such as the same temperature value, the data are the same. For this reason, also core side data delivery time
just sent once, i.e., only the data variations are sent instead is not affected.
of always sending each data; 2) data reformatting—data In the second additional test, we introduce a high degree
message format is changed from text to binary in order to of congestion by increasing the number of sensors by ten
add some lossless compression. These functions, colocated times (NS = 50 000). The uplink becomes quickly congested
with the edge MQTT broker, are so able to reduce redundant in all cases, but the introduction of MEC provides a substan-
information and aggregate sensor data in a much more tial difference. Without MEC, also the satellite downlink
compact form by using larger TCP packets. These additional is highly occupied, with consequent very high mean data
actions lead to a further satellite uplink utilization reduction delivery time values both at core and edge sides. Several
of 12%, with an average value of 21.1 kb/s. At the same time, sensors suffered also disconnections due to the excessive
core side data delivery time is not so much affected, with packet loss and TCP retransmissions due to retransmission
an average value increased to 647 ms. timeout expiration. As soon as MEC is enabled, the core
side performance is guaranteed, obtaining an edge side
C. Further Testing Configurations
data delivery time mean value comparable with the default
We performed additional tests to determine the effec- configuration. Case 3 shows a further better performance of
tiveness of the MEC architecture considering alternative the core side data delivery time. This specific configuration
configurations. These configurations have been designed allows assessing the reaction of the system in case of con-
with the aim of presenting some limit/stress cases, in which gestion: the increase of sensors from 5000 to 50 000 does
we want to show the scalability of the service, and the valid- not lead to a proportional increase of the data delivery time,
ity of the approach when the number of local-only sensors which instead increases of about 100 times. This conclusion
or clients is reduced. Furthermore, even when local data are is not applicable, as already discussed, for the edge side data
managed at the edge and distributed only locally, processing delivery time with MEC enabled, which is not affected.

3782 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 58, NO. 5 OCTOBER 2022
 TABLE II
Mean Results From the Additional Performed Tests

Finally, without local-only real-time and nonreal-time possible link disruptions between the IoT base station and
sensors (rl = nl = 0), we can observe a proportional re- satellites on the shown performance in all the considered
duction of the satellite link bandwidth utilization in Case 1. cases, so to highlight the improvement achievable with
Concerning Cases 2 and 3, there are no significant changes the proposed MEC functionalities also in the case of LEO
since the local-only traffic flows do not traverse the satellite satellite backhaul links.
link. This test allows in particular to confirm the effective-
ness of data local management. ACKNOWLEDGMENT
The execution of additional runs showed the main ben-
Responsibility of the content resides with the authors.
efit of the MEC-based approach in several configurations.
According to the specific applications and associated re-
quirements in terms of satellite link utilization and latency, REFERENCES
a service operator can decide which is the best architectural
configuration depending on the numerical results shown in [1] Cisco, “Cisco annual Internet report (2018-2023),” San Jose,
CA, USA, Cisco, White Paper, 2020. [Online]. Available:
this article. https://www.cisco.com/c/en/us/solutions/collateral/executive-pers-
pectives/annual-internet-report/white-paper-c11-741490.html
V. CONCLUSION [2] J. Liu, Y. Shi, Z. M. Fadlullah, and N. Kato, “Space-air-ground
integrated network: A survey,” IEEE Commun. Surveys Tuts., vol. 20,
The huge amount of raw IoT data could stress the limited no. 4, pp. 2714–2741, Fourth quarter 2018.
and expensive SatCom resources, which have to be carefully [3] S. Cioni, R. De Gaudenzi, O. D. R. Herrero, and N. Girault, “On the
managed. Equipping the edge of the network with local pro- satellite role in the era of 5G massive machine type communications,”
cessing MEC functionalities, such as data compression and IEEE Netw., vol. 32, no. 5, pp. 54–61, Sep./Oct. 2018.
[4] N. Kato et al., “Optimizing space-air-ground integrated networks
aggregation, allow improving end-to-end performance and by artificial intelligence,” IEEE Wireless Commun., vol. 26, no. 4,
QoS in communications when satellite links are exploited. pp. 140–147, Aug. 2019.
This article has analyzed a SatCom-based MEC network [5] F. Chiti, R. Fantacci, and L. Pierucci, “Energy efficient commu-
architecture where a GEO satellite acts as a backhaul node nications for reliable IoT multicast 5G/Satellite services,” Future
between an IoT sensor network and the Internet. A dis- Internet, vol. 11, no. 8, pp. 164–176, 2019.
[6] M. Bacco et al., “Networking challenges for non-terrestrial networks
tributed MQTT framework is defined where local instances exploitation in 5G,” in Proc. IEEE 2nd 5G World Forum, 2019,
of MQTT brokers and data aggregation and compression pp. 623–628.
techniques are deployed as Virtual Functions at the edge of [7] L.-A. Turner and H. Jahankhani, “An investigation into an approach
the network. Performance evaluation is carried out through to updating the governance of satellite communications to enhance
an ad hoc designed testbed comparing the obtained perfor- cyber security,” in Proc. Cybersecur. Privacy Freedom Protection
Connected World, 2021, pp. 23–33.
mance with three different network configurations in terms [8] M. Lin et al., “Integrated 5G-satellite networks: A perspective on
of the amount of data that need to traverse the satellite physical layer reliability and security,” IEEE Wireless Commun.,
link and data delivery time. The employment of MEC vol. 27, no. 6, pp. 152–159, Dec. 2020.
functionalities allows obtaining a significant reduction of [9] Q.-V. Pham et al., “A survey of multi-access edge computing in 5G
the satellite link bandwidth utilization and of the achievable and beyond: Fundamentals, technology integration, and state-of-the-
art,” IEEE Access, vol. 8, pp. 116974–117017, 2020.
data delivery time in most of the considered cases. The use [10] S. Kekki et al., “MEC in 5G networks,” Sophia Antipolis, France,
of data aggregation techniques leads to a higher delivery ETSI, White Paper, 2018.
time in some cases, which can affect delay-tolerant data, so [11] P. Porambage, J. Okwuibe, M. Liyanage, M. Ylianttila, and T. Taleb,
implying a careful design. “Survey on multi-access edge computing for Internet of Things real-
Considering the rapid growth of LEO SatCom systems, ization,” IEEE Commun. Surveys Tuts., vol. 20, no. 4, pp. 2961–2991,
Fourth quarter 2018.
a second performance evaluation will be performed [12] T. K. Rodrigues, K. Suto, and N. Kato, “Edge cloud server deploy-
considering an additional reference scenario based on a ment with transmission power control through machine learning for
LEO satellite constellation. This further analysis aims to 6G Internet of Things,” IEEE Trans. Emerg. Topics Comput., vol. 9,
show the impact of aspects, such as satellite handover and no. 4, pp. 2099–2108, Oct.–Dec. 2021.

LUGLIO ET AL.: PERFORMANCE EVALUATION OF A SATELLITE COMMUNICATION-BASED MEC ARCHITECTURE 3783

[13] R. Xie, Q. Tang, Q. Wang, X. Liu, F. R. Yu, and T. Huang, “Satellite- [33] OASIS, “Message queuing telemetry transport (MQTT) version
terrestrial integrated edge computing networks: Architecture, chal- 3.1.1,” 2015. Accessed: Aug. 18, 2022. [Online]. Available: http:
lenges, and open issues,” IEEE Netw., vol. 34, no. 3, pp. 224–231, //docs.oasis-open.org/mqtt/mqtt/v3.1.1/mqtt-v3.1.1.pdf
May/Jun. 2020. [34] Z. Shelby, K. Hartke, C. Bormann, and B. Frank, “The constrained
[14] L. Yan et al., “SatEC: A 5G satellite edge computing framework application protocol (CoAP),” 2014. Accessed: Aug. 18, 2022. [On-
based on microservice architecture,” Sensors, vol. 19, no. 4, 2019, line]. Available: https://www.rfc-editor.org/rfc/rfc7252
Art. no. 831. [35] R. Soua, M. R. Palattella, and T. Engel, “IoT application protocols
[15] B. Denby and B. Lucia, “Orbital edge computing: Nanosatellite optimisation for future integrated M2M-Satellite networks,” in Proc.
constellations as a new class of computer system,” in Proc. 25th Int. IEEE Glob. Inf. Infrastructure Netw. Symp., 2018, pp. 1–5.
Conf. Architectural Support Program. Lang. Operating Syst., 2020, [36] M. Bacco, P. Cassarà, M. Colucci, and A. Gotta, “Modeling re-
pp. 939–954. liable M2M/IoT traffic over random access satellite links in non-
[16] Z. Zhang, W. Zhang, and F.-H. Tseng, “Satellite mobile edge comput- saturated conditions,” IEEE J. Sel. Areas Commun., vol. 36, no. 5,
ing: Improving QoS of high-speed satellite-terrestrial networks using pp. 1042–1051, May 2018.
edge computing techniques,” IEEE Netw., vol. 33, no. 1, pp. 70–76, [37] F. Wang, D. Jiang, S. Qi, C. Qiao, and H. Song, “Fine-grained
Jan./Feb. 2019. resource management for edge computing satellite networks,” in
[17] T. De Cola, M. Marchese, M. Mongelli, and F. Patrone, “A unified Proc. IEEE Glob. Commun. Conf., 2019, pp. 1–6.
optimisation framework for QoS management and congestion con- [38] C. Qiu, H. Yao, F. R. Yu, F. Xu, and C. Zhao, “Deep Q-learning
trol in VHTS systems,” IEEE Trans. Veh. Technol., vol. 69, no. 10, aided networking, caching, and computing resources allocation in
pp. 11619–11631, Oct. 2020. software-defined satellite-terrestrial networks,” IEEE Trans. Veh.
[18] J. Wei, J. Han, and S. Cao, “Satellite IoT edge intelligent comput- Technol., vol. 68, no. 6, pp. 5871–5883, Jun. 2019.
ing: A research on architecture,” Electronics, vol. 8, no. 11, 2019, [39] ESA ARTES AT Project, “VIBES - Implementation of Virtualised
Art. no. 1247. network functions (VNFs) for broadband satellite networks,” 2018.
[19] Y. Wang, J. Yang, X. Guo, and Z. Qu, “Satellite edge computing for Accessed: Aug. 18, 2022. [Online]. Available: https://artes.esa.int/
the Internet of Things in aerospace,” Sensors, vol. 19, no. 20, 2019, projects/vibes
Art. no. 4375. [40] ESA ARTES AT Project, “M2MSAT - Demonstrator of light-weight
[20] D. Chen et al., “Resource cube: Multi-virtual resource management application and transport protocols for future M2M applications,”
for integrated satellite-terrestrial industrial IoT networks,” IEEE 2017. Accessed: Aug. 18, 2022. [Online]. Available: https://artes.
Trans. Veh. Technol., vol. 69, no. 10, pp. 11963–11974, Oct. 2020. esa.int/projects/m2msat
[21] N. Cheng et al., “Space/Aerial-assisted computing offloading for [41] M. R. Palattella, R. Soua, A. Stemper, and T. Engel, “Aggregation
IoT applications: A learning-based approach,” IEEE J. Sel. Areas of MQTT topics over integrated satellite-terrestrial networks,” ACM
Commun., vol. 37, no. 5, pp. 1117–1129, May 2019. SIGMETRICS Perform. Eval. Rev., vol. 46, no. 3, pp. 96–97, 2019.
[22] G. Cui, X. Li, L. Xu, and W. Wang, “Latency and energy optimiza- [42] Ericsson, “Massive IoT in the city,” Mobility Report, 2016. Ac-
tion for MEC enhanced SAT-IoT networks,” IEEE Access, vol. 8, cessed: Aug. 18, 2022. [Online]. Available: https://www.ericsson.
pp. 55915–55926, 2020. com/en/mobility-report/articles/massive-iot-in-the-city
[23] Y. Cao, Y. Shi, J. Liu, and N. Kato, “Optimal satellite gateway place-
ment in space-ground integrated network for latency minimization
with reliability guarantee,” IEEE Wireless Commun. Lett., vol. 7,
no. 2, pp. 174–177, Apr. 2018.
[24] The Third Generation Partnership Project (3GPP), “Solutions for
NR to support non-terrestrial networks (NTN),” Tech. Rep. 38.821
v16.1.0, Jun. 2021.
[25] ESA ARTES AT Project, “NB-IoT4Space - 3GPP Narrow-Band
Internet-of-Things (NB-IoT) user sensor integration into satellite,” Michele Luglio received the Laurea degree in
2020. Accessed: Aug. 18, 2022. [Online]. Available: https://artes. electronic engineering and the Ph.D. degree
in telecommunications from the University of
esa.int/projects/nbiot4space
Rome “Tor Vergata,” Rome, Italy, in 1990 and
[26] A. Guidotti et al., “Architectures and key technical challenges for 5G 1994, respectively.
systems incorporating satellites,” IEEE Trans. Veh. Technol., vol. 68, From August to December 1992, he was a
no. 3, pp. 2624–2639, Mar. 2019. Visiting Staff Engineer with Comsat Laborato-
[27] M. Centenaro, C. E. Costa, F. Granelli, C. Sacchi, and L. Vangelista, ries, Clarksburg, MD, USA. From 1995 to 2004,
“A survey on technologies, standards and open challenges in satellite he was a Research and Teaching Assistant with
IoT,” IEEE Commun. Surv. Tut., vol. 23, no. 3, pp. 1693–1720, the University of Rome “Tor Vergata.” In 2001
Third quarter 2021. and 2002, he was a Visiting Professor with the
[28] R. Giuliano, F. Mazzenga, and A. Vizzarri, “Satellite-based capillary Computer Science Department, University of California Los Angeles, Los
Angeles, CA, USA, to teach Satellite Networks class. Since 2004, he has
5G-mMTC networks for environmental applications,” IEEE Aerosp.
been an Associate Professor of telecommunications with the University
Electron. Syst. Mag., vol. 34, no. 10, pp. 40–48, Oct. 2019. of Rome “Tor Vergata.” He teaches “Internet via Satellite” and “Telecom-
[29] M. Höyhtyä, T. Ojanperä, J. Mäkelä, S. Ruponen, and P. Järvensivu, munications Fundamentals”. He works on designing innovative satellite
“Integrated 5G satellite-terrestrial systems: Use cases for road safety communications systems for multimedia services, both mobile and fixed.
and autonomous ships,” in Proc. 23rd Ka Broadband Commun. Conf., He coordinates the Laboratory of the Satellite Multimedia Group, Univer-
2017, pp. 16–19. sity of Rome Tor Vergata, he is the President of the University spin-off
[30] J. V. Y. Martnez, A. F. Skarmeta, M. A. Zamora-Izquierdo, and RomARS, and was the President of the Research Interuniveristy consor-
A. P. Ramallo-Gonzlez, “IoT-based data management for smart tium NITEL. He is author of more than 170 international publications.
agriculture,” in Proc. 2nd Int. Conf. Embedded Distrib. Syst., 2020, His research interests include network protocols, resource management,
heterogeneous networks, and 5G development with particular regard to
pp. 41–46.
satellite systems.
[31] K. Sohraby, D. Minoli, B. Occhiogrosso, and W. Wang, “A review Dr. Luglio was the General Co-Chair of IEEE ISNCC 2018, ISAECT
of wireless and satellite-based M2M/IoT services in support of smart 2019, IEEE ISNCC 2018, ISAECT 2019, ISAECT 2020, and ISAECT
grids,” Mobile Netw. Appl., vol. 23, no. 4, pp. 881–895, 2018. 2021. He is a Member of the Editorial Board of China Communications
[32] T. Wei, W. Feng, Y. Chen, C.-X. Wang, N. Ge, and J. Lu, “Hy- and IAENG Engineering Letters. He is an Associate Editor for Space
brid satellite-terrestrial communication networks for the maritime Communications section of Frontiers in Space Technologies. He was
Internet of Things: Key technologies, opportunities, and challenges,” affiliated with the Italian Space Agency and has been the Italian expert
IEEE Internet of Things J., vol. 8, no. 11, pp. 8910–8934, Jun. 2021. delegate and Co-Chair of the advisory committee 5JAC of ESA.

3784 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 58, NO. 5 OCTOBER 2022
 Mario Marchese (Senior Member, IEEE) re- Cesare Roseti graduated (cum laude) in TLC en-
ceived the “Laurea” degree (cum laude) and the gineering and the Ph.D. degree in space systems
Ph.D. degree in telecommunications from the and technologies from the University of Rome
University of Genoa, Genoa, Italy, in 1992 and “Tor Vergata,” Rome, Italy, in 2007.
1997, respectively. Since 2012, he has been an Assistant Pro-
From 1999 to January 2005, he was with fessor with the University of Rome “Tor Ver-
the Italian Consortium of Telecommunications gata,” teaching technologies for the communi-
(CNIT), Parma, Italy, where he was the Head cations and broadcast technologies. He is the
of research. From February 2005 to January Co-Founder of RomARS srl, Rome, Italy, an in-
2016, he was an Associate Professor, and since novative startup dealing with advanced telecom-
February 2016, he has been a Full Professor with munication technologies. He is technical and/or
the University of Genoa, where, since December 2020, he has been Rector’s scientific responsible of several international research projects addressing
Delegate to Doctoral Studies. He is the author of the book Quality of Service converged satellite and 5G networks, and he actively participated to several
over Heterogeneous Networks (John Wiley Sons, 2007), and has authored funded international research programmes, with the role of WP leader.
or coauthored more than 300 scientific works, including international He is the coauthor of about 95 scientific publications for international
magazines, international conferences, and book chapters. His research conferences, workshops, and journals.
interests include networking, quality of service over heterogeneous net-
works, software-defined networking, satellite networks, network security,
critical infrastructure security and intrusion detection systems.

Francesco Zampognaro received the M.Sc. de-

gree in telecommunication engineering from the
University of Rome “La Sapienza,” Rome, Italy,
in 2004, and the Ph.D. degree in space systems
Fabio Patrone (Member, IEEE) received the and technologies from University of Rome “Tor
M.Sc. degree in telecommunication engineer- Vergata” Rome, Italy, in 2010.
ing and the Ph.D. degree in telecommunications He is currently a Researcher with the Uni-
from the University of Genoa, Genoa, Italy, in versity of Rome “Tor Vergata,” and a Professor
2013 and 2016, respectively. in space systems at University Marconi, Rome,
He is currently an Assistant Professor with the Italy. He teaches with Linux Management and
Satellite Communications and Heterogeneous Configuration Laboratory, MATLAB for Signal
Networking Laboratory, University of Genoa, theory and Satellite Systems classes. He is Co-Founder of RomARS
Genoa, Italy. His research interests include rout- srl, Rome, Italy, an innovative startup dealing with advanced telecom-
ing, scheduling, and congestion control algo- munication technologies. His research interests include satellite systems,
rithms in satellite, vehicular, and sensor net- supporting many research funded projects, and study and simulation of
works, and the employment of networking technologies, such as network satellite systems, in particular DVB-RCS, covering protocols optimization,
function virtualization and software-defined networking for the integration services provision and QoS, virtualization, security, resource allocation,
of these networks with the terrestrial infrastructure within 5G. integrated/hybrid architectures, and 5G.

Open Access provided by ‘Università degli Studi di Genova’ within the CRUI CARE Agreement

LUGLIO ET AL.: PERFORMANCE EVALUATION OF A SATELLITE COMMUNICATION-BASED MEC ARCHITECTURE 3785