A Framework for Security Monitoring of Real IoT Testbeds
Vinh Hoa La
1
, Edgardo Montes de Oca
1
, Wissam Mallouli
1
and Ana R. Cavalli
12
1
Montimage, 39 rue Bobillot, 75013 Paris, France
2
SAMOVAR, CNRS, Telecom SudParis, Paris-Saclay University, 9 rue Charles Fourier, 91011 EVRY, France
Keywords:
Intrusion Detection, Anomaly Detection, 6LoWPAN, Wireless Sensor Networks, IoT, Security Monitoring.
Abstract:
Internet of Things (IoT) has been acknowledged as a novel transformation technology because of its wide
range of applications in various domains, namely connected agriculture, industrial control, smart buildings
and home automation. It promises innovative business models and improved user experience. However, the
devices are prone to failures and malicious attacks on account of their resource-constrained characteristics. In
this paper, we present a framework for security monitoring of IoT systems. It is based on MMT-IoT, which is a
reactive monitoring tool to be deployed in a running IoT environment to address malicious behaviors, failures
and attacks. In this paper we also present the experiments conducted on two practical IoT-6LoWPAN testbeds.
The preliminary results confirmed the efficiency of the proposed solution.
1 INTRODUCTION
Computer security, also known as cyber-security or IT
security, has been an emerging topic for decades. It is
expected to attract even more attention due to the in-
creasing reliance on computer systems in many differ-
ent domains. Computer systems here are not limited
to servers, desktops or laptops but also include smart
devices (e.g. smart-phones, connected objects, sensor
devices). The pervasiveness of these systems goes to-
gether with the growing of cyber-attacks in both vol-
ume and sophistication. According to a study made
by Symantec
1
in 2015, nearly one million new mal-
ware threats are released every day. Two-thirds of In-
ternet users have been victims of cyber-crime, with
more than 1.5 million new victims every day.
Additionally, the incredible growth of Internet
and wireless networks, based on technologies such
as Bluetooth and Wi-Fi and the concept Internet of
Things (IoT), promise to make future networks be-
come Internet of Every Things. There are nowadays
about 15 billion of IoT devices and they are esti-
mated to be 50 billion connected devices in 2020,
according to a report by Cisco and DHL (Macaulay
et al., 2015). As a representation, Wireless Sensor
Networks (WSNs) have been attracting a lot of inter-
est from both the research community and the pub-
lic. However, the resource-constrained characteris-
1
https://www.symantec.com/security-center/threat-
report
tics of physical objects in those networks presumably
limit the design and development of security proto-
cols. Whilst, sensor nodes, which usually operate in
remote, unattended and even harsh environments, are
prone to failures and malicious attacks.
In the last years, the research on IoT/WSNs was
mainly focused on how to make the concept of IoT
realistic and practical. In other words, most of the
IoT research projects have been trying to qualify this
technology by standardizing the communication pro-
tocols, ameliorating the performance of the IoT sys-
tems, optimizing the resource consumption, etc. Se-
curity is always considered as an important issue but
difficult to achieve thoroughly because it seems con-
tradictory with the system’s performance due to the
resource constraints of IoT devices.
To date, there are a number of research works on
the subject of IoT security. However, they mostly con-
centrate on designing secure communication proto-
cols, light encryption, authentication, data freshness
(avoiding packet injection), etc. Recently, researchers
are paying more and more attention on monitoring
in general and intrusion detection in particular for
IoT/WSNs. However, many existing approaches are
still at the design level and not yet implemented.
In this paper, we propose a framework for secu-
rity monitoring of IoT systems based on the MMT-
IoT tool. The tool allows capturing and analysing
the traffic generated by the IoT devices, as well as
visualising the findings. The solution has been de-
La, V., Montes de Oca, E., Mallouli, W. and Cavalli, A.
A Framework for Security Monitoring of Real IoT Testbeds.
DOI: 10.5220/0010578506450652
In Proceedings of the 16th International Conference on Software Technologies (ICSOFT 2021), pages 645-652
ISBN: 978-989-758-523-4
Copyright
c
2021 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
645
ployed and tested on two FED4FIRE+
2
platforms: w-
iLab.t
3
(provided by imec) and Log-a-Tec
4
(provided
by the Jo
ˇ
zef Stefan Institute - JSI). These experiments
have been decisive for raising the maturity level of the
MMT-IoT solution and, thus, provide a clearer view
of its market potential.
The rest of this paper is organized as follows. Sec-
tion 2 gives a presentation of the proposed framework.
It includes an introduction of MMT-IoT, as well as
the adaptations performed to make the tool applica-
ble for the FED4FIRE+ testbeds. These testbeds are
described in Section 3. The experimental results of
testing MMT-IoT in the two testbeds are presented in
Section 4. Finally, we conclude our study and identify
necessary future work in Section 5.
2 A FRAMEWORK FOR THE
MONITORING OF IoT
NETWORKS
2.1 MMT-IoT a Security Monitoring
Tool for IoT Networks: General
Architecture
Montimage has developed an extensible monitor-
ing framework called Montimage Monitoring Tool
(MMT) (Wehbi et al., 2012; Mallouli et al., 2012). It
has been conceived as a modular approach to analyse
security properties of conventional networks (both
wired and wireless) by means of extracting statisti-
cal information from the network protocols and feed-
ing it to an engine to perform a temporal logic-based
analysis. Montimage has adapted this technology to,
respectively, the Cyber Physical Systems (CPS/IoT)
and 5G networks, considering the particular require-
ments of these networks.
The MMT-IoT (La et al., 2016; La and Cavalli,
2016) has thus been developed to be used with the
MMT-Probe software. In general, the main goal of
the MMT-IoT solution is designed to avoid perform-
ing heavy operations on the IoT devices, leaving the
security analysis for the traditional MMT-Probe so-
lution. Since the latter is a Linux-based tool, it is
implicitly constrained to the protocols that the Linux
kernel is capable of handling. In particular, the IEEE
802.15.4 Protocol (IoT-specific Layer 2 protocol) is
not natively supported by the Linux kernel. In this
case, when Linux protocol stack tries to identify the
2
https://www.fed4fire.eu/
3
https://doc.ilabt.imec.be
4
http://www.log-a-tec.eu/
layer 2 protocol, it will not understand the frame for-
mat and, in consequence, discard the packets they
reach the Linux network drivers. To avoid this, an
abstraction layer needs to be inserted between the IoT
traffic and MMT-Probe, so the latter will be able to
capture the traffic from any traditional Linux inter-
face.
Figure 1: MMT-IoT general architecture.
Figure 1 shows how the MMT-IoT technology is
capable of extracting the information from the IoT
protocols. In order to correctly adapt this approach
(designed initially for traditional Ethernet networks)
it was required to split the network extractor (sniffer)
in two parts: the MMT-IoT Sniffer and the MMT-
IoT Bridge. The former is the IoT endpoint that is
in charge of sniffing the packets and forwarding them
to a more powerful machine through a USB line. The
latter recovers the transferred packets from the USB
line and injects them (encapsulated using the ZEP
protocol) in the loopback interface of the machine,
making the packets ready to be analysed by MMT-
Probe (in charge of parsing the protocol communi-
cations and analysing them) and visualised by the
MMT-Operator (in charge of collecting the informa-
tion provided by the probes and visualising statistics
and alarms).
The described technology is the core of the pro-
posed IoT framework, that is tested in the context of
this paper to determine its performance and scalabil-
ity, as well as to evaluate the effectiveness of the at-
tack/anomaly detection on IoT networks.
2.2 Adaptation of MMT-IoT to the
Zolertia Re-Mote Devices
As mentioned before, Montimage required the de-
ployment of its software on real IoT devices in or-
der that it can be used in the FED4FIRE+ platforms.
To this end, Montimage selected the Zolertia Re-Mote
devices
5
for its ease of deployment and its wide range
of support for IoT operating systems.
Figure 2 presents the general architecture of the
experiments performed using MMT-IoT. The number
of the Zolertia devices acting as the clients may vary,
5
https://zolertia.io/
ICSOFT 2021 - 16th International Conference on Software Technologies
646
Figure 2: General architecture for the experiments.
depending on the objective of the experiment or the
availability of the devices. In summary, the following
adaptations have been performed in order to facilitate
the application of MMT-IoT:
Border Router firmware: The Border Router is the
edge device placed between the IoT network and
the traditional IP network. In the context of this
work, the Border Router connects with a Linux-
based machine that acts as the gateway collecting
the sensed data sent by the IoT 6LoWPAN devices
(i.e. UDP clients), and transferring it to a dedi-
cated UDP Server via a USB line tunslip6 con-
nection. Tunslip
6
is a tool used to bridge IP traf-
fic between a host and another network element
over a serial line. The firmware is for forming a
6LoWPAN network and routing the messages to
the server.
Client firmware: The clients and the Border
Router self-organise among themselves to form a
6LoWPAN network. The normal clients are con-
figured to report sensed data every 10 seconds
7
,
whilst the attacker client behaves interchangeably
in one of the three modes (Normal, DoS attack
and Dead modes) changing every 60 seconds.
Sniffer: The sniffer includes a peripheral firmware
capturing all network frames and streaming them
to the host and a python script reading network
packets captured by the peripheral, converting
them to the PCAP format and piping them to the
Linux-based machine via the USB line.
MMT-IoT: A number of adapta-
tions/modifications have been integrated to
MMT so that the tool is able to work with IEEE
802.15.4/6LoWPAN traffic. This included devel-
oping new plugins (for parsing these packets),
6
https://github.com/contiki-
os/contiki/blob/master/tools/tunslip6.c
7
Note that values are purely illustrative and depend on
the business application
enabling the configuration of the network stack,
specifying new security rules, and designing new
dashboards.
3 SET-UP OF THE
EXPERIMENTS
In the context of these experiments, Montimage used
two FED4FIRE+ testbeds to assess and test the proto-
type and get feedback to improve the solution to reach
TRL6
8
. In particular, Montimage pursued two prin-
cipal objectives with these experiments:
Analyse the performance and the scalability of the
MMT-IoT solution in a real IoT scenario (repre-
sented in Figure 2),
Perform security analysis in IoT networks by
means of detecting well-known attack types.
3.1 w-iLab.t Testbed
The w-iLab.t platform allows its user to run remote
experiments in a fully automated way. Because
of limited interference from outside, repeatable ex-
perimentation is enabled, which allows the experi-
menters to compare results between different exper-
iment runs. Montimage has created credentials on the
FED4FIRE+ portal, with which all testbeds in the fed-
eration can be accessed. For this experiment it was
necessary to reserve all the available nodes (embed-
ded PC of type Intel NUC) from the “Datacenter”
floor of the testbed (as shown in Figure 2). Every
node in the w-iLab.t testbed has at least one Zoler-
tia Re-Mote sensor connected over USB. This sensor
node can be accessed after logging in to the embed-
ded PC. In addition to the reservation, to access these
nodes (over SSH) it was required to design an exper-
iment using the jFED-Experimenter tool. Before do-
ing this, we defined the two scenarios corresponding
to the two main objectives of the experiments run on
the w-iLab.t:
A main objective of using the w.iLab.t platform
is to perform scalability tests of the MMT-IoT
software. In this sense, we needed a big num-
ber of nodes that are capable of generating test
traffic that will be captured by a central sniffer.
More precisely, 21 clients were flashed with nor-
mal (i.e. reporting the data every 10 seconds)
or attacker (i.e. behaving interchangeably in the
Normal mode, DoS attack mode and Dead mode)
firmware so that we could increase the bitrate of
8
TRL: Technology Readiness Level
A Framework for Security Monitoring of Real IoT Testbeds
647
sensed data on the air to determine the limit of the
sniffer in terms of capturing traffic. In addition,
the MMT-IoT could be tested to determine if it
could detect the DDoS (Distributed Denial of Ser-
vice) attack when multiple nodes act as attackers.
Another objective is performing security tests on
IoT networks. In this sense, we modified the
firmware flashed on the IoT device that gener-
ates the malicious traffic so that it can be anal-
ysed and detected by the MMT-IoT. Indeed, 7 nor-
mal clients and 1 attacker were used. MMT-IoT is
tested to see if it could capture all the traffic with
no packet lost, extract the statistics, visualise them
using graphs, as well as detect 3 types of anoma-
lies:
DoS attack: Instead of reporting every 10 sec-
onds like the normal nodes, the attacker sends
sensed data 100 times faster, i.e. 10 mes-
sages/second;
Dead node (node failure): A normal node that
should send sensed data, but that stopped re-
porting (e.g. due to a failure).
Incorrect FCS (Frame Check Sequence): The
FCS field contains a number that is calculated
by the source node based on the data in the
frame. This number is added to the end of a
frame that is sent. When the destination node
receives the frame, the FCS number is recalcu-
lated and compared with the FCS number in-
cluded in the frame. If they are different, the
frame will be considered malformed (intention-
ally or not) or modified between the route from
the source node to the destination node. During
the period of the DoS attack mode, the attacker
sends messages with incorrect FCS. In real-
ity, an incorrect FCS can signify a malformed
packet (e.g. due to a misconfiguration or an er-
ror in the implementation), a jamming attack
(i.e. the attacker abuses the network by generat-
ing frames that should be ignored) or a message
manipulation attack (i.e. the attacker intercepts
and modifies a frame’s content).
Each scenario was executed (i.e. triggering of the at-
tack, start of the sniffing process along with the secu-
rity analysis) for at least 5 minutes, in order to gener-
ate enough traffic that could be later analysed.
3.2 Log-a-Tec Testbed
Montimage prepared a set of equipment forming a
“small” IoT 6LoWPAN network. It includes four Zol-
ertia RE-Motes, one Raspberry Pi and the accessories
(cables, USB hub, power adapters, etc.) to constitute:
A border router mote that acts as the gateway col-
lecting the sensed data from other motes and for-
wards the reports via the USB line to the server
deployed in the Raspberry Pi.
A sniffer mote that runs in a passive manner: lis-
tening for air traffic, capturing and piping it via
the USB line connected to the Raspberry Pi.
A normal client mote that reports the data every
10 seconds.
An attacker mote that behaves interchangeably in
the three modes described before (Normal, DoS
attack and Dead mode).
A Raspberry Pi that feeds the motes in terms
of batteries, hosting the server dealing with the
sensed data and receiving the sniffed traffic which
is then analysed by MMT-IoT.
Figure 3: The distribution of the Log-a-Tec Testbed.
The devices were sent to JSI to be placed near the
Log-a-Tec Testbed whose distribution is depicted in
Figure 3 (on the ground and in the office building).
The devices of JSI at the yellow dots are configured to
form a 6LoWPAN (G. Montenegro and Culler, 2007;
Vasseur et al., 2011) network on the same channel
(channel 15) as Montimage’s Zolertia Re-Motes. The
devices at the position 4 and 5 were offline at the mo-
ment of the experiments. Montimage’s devices and
the Border Router of JSI’s network were set up at dif-
ferent places:
Outdoor: at the position 6,
Outdoor: in a park, on a table between the position
3 and 5,
Indoor: in the office at the position 9.
ICSOFT 2021 - 16th International Conference on Software Technologies
648
4 EXPERIMENTAL RESULTS
In each experiment, the traffic captured is analysed
by the MMT-Probe to extract the statistics, to ver-
ify the security violations defined in MMT-Security
rules, and to visualise the obtained results on a MMT-
Operator dashboard. The current version of the dash-
board permits displaying the traffic bitrate, the most
active nodes, the most active links, the topology, the
security alerts and their distribution. In the context of
this work, anomalies were detected based on specific
symptoms, such as the following:
DoS attack: MMT-IoT raises an alert about a pos-
sible DoS attack if it sees a node reporting data
more than 4 times in one second. The values “4
times” and “one second” are configurable and de-
pend on the behaviour of the real use case.
DDoS attack: MMT-IoT raises an alert about a
possible DDoS attack if it sees two nodes report-
ing at least twice in a second. Similarly to the
item above, “two nodes” and “twice a second” are
configurable parameters depending on the real use
case.
Dead node (node failure): MMT-IoT raises an
alert about a possible dead node if it sees a node
that should send sensed data but does not. For in-
stance, should send a report every 10 seconds but
stops for at least 20 seconds.
Incorrect FCS: MMT-IoT calculates the FCS and
compares it with the FCS included in the packet
of the sender. A difference triggers an alert of an
incorrect FCS that might be related to a miscon-
figuration, an implementation error, a malformed
packet, a jamming attack, or a manipulation attack
in which the payload was modified intentionally.
4.1 w-iLab.t Testbed
Regarding the scalability and security tests with 27
clients, we discovered that 32 Kbps was the maximum
data bit rate that can be handled by the sniffer (shown
in Figure 5). Above this limit, the sniffer started drop-
ping some frames. In theory, by default the Zolertia
Re-Mote is capable of working at 50 Kbps data rates.
In practice, the limit in the context of our experiments
was found to be lower. This can be explained by
the physical factors that can affect the transmission,
namely the antenna, the power, the noise, etc., and by
the fact that such IoT devices are not always stable.
The DoS attack can be alerted as shown in Figure
4 but can also be intuitively observed by just looking
at Figure 5. The throughput greatly increased when
the DoS attack took place. Of course the variation
Figure 4: DDoS alerts in the test with 27 clients.
between normal operation and operation under attack
depends on the use case that could imply low or high
number of communications, and low or high variabil-
ity in the traffic.
Regarding the security tests with 7 clients and 1 at-
tacker, MMT-IoT raised the alerts correctly when the
attacker changed from the Normal mode to the DoS
attack and Dead mode, as shown in Figure 6 and 7
respectively. The messages containing incorrect FCS
were also detected, as demonstrated in Figure 8. The
attacker could be identified not only based on the se-
curity alerts but also by looking at the traffic volume
generated. For example, as displayed in Figure 9, the
attacker is the most active node.
4.2 Log-a-Tec Testbed
As mentioned in the previous sections, our IoT de-
vices were placed at different positions near the Log-
a-Tec testbed. However, regardless of the position
of JSI’s Border Router and Montimage’s devices, the
sniffer captured both traffic generated by Montim-
age’s Zolertia Re-Motes and by JSI’s 6LoWPAN net-
work. This was determined by the fact that there were
only 4 Zolertia Re-motes but the sniffer captured the
traffic of at least 10 nodes. In addition, after captur-
ing the traffic and having it analysed by MMT-IoT,
we observed that there was, in general, more traffic
captured due to two reasons:
The sniffer “S” (together with Montimage’s
equipment) is placed in a central position (at po-
sition 6 or between 3 and 5 in Figure 3),
The sniffer is placed close to the Border Router
“BR” of JSI’s network.
As a result, the least traffic was observed when “S”
was indoor at the position 9 (behind metallized glass)
and “BR” was at 6 (Figures 10). There were only 10
nodes found generating the traffic at a rather low data
rate (less than 1 Kbps). Meanwhile, most traffic was
A Framework for Security Monitoring of Real IoT Testbeds
649
Figure 5: Traffic bitrate observed in the test with 27 clients.
Figure 6: DoS alerts in the test with 7 clients and 1 attacker.
Figure 7: Dead node alerts in the test with 7 clients and 1
attacker.
witnessed when both “S” and “BR” were on the table
between 3 and 5 (Figures 11). 22 nodes were seen
generating the traffic with the bitrate of up to 3Kbps.
Regarding the security aspect, MMT detected the
misbehavior of the attacker node in all the performed
experiments(as shown in Figure 12). The attacker was
identified when it triggered the DoS attack or when it
acted as a dead node and did not report any data at all.
The messages containing the incorrect FCS were also
Figure 8: Incorrect FCS alerts in the test with 7 clients and
1 attacker.
Figure 9: Most active nodes in the test with 7 clients and 1
attacker.
alerted.
Most surprisingly, although Montimage prepared
only one Zolertia Remote acting as the DoS attacker,
the DDoS attack was still detected. This means that
there was another node of IJS’s testbed sending data
ICSOFT 2021 - 16th International Conference on Software Technologies
650
Figure 10: Active nodes observed when “S” was at 9 and
“BR” was at 6.
Figure 11: Most active nodes observed when both “S” and
“BR” were between 3 and 5.
at a very high rate. For example, as demonstrated in
Figure 13, there were two nodes reporting data twice
in several milliseconds which were alerted as a prob-
able DDoS attack.
This finding is explained by the fact that the de-
vices in Log-a-Tec communicate using the 6TiSCH
standard
9
where devices re-transmit the packets if the
acknowledgment for the sent packets is not received.
In this case, it sends 8 packets in a row within a short
interval of time, since the time slot of the 6TiSCH is
only 10ms. It could happen that a device leaves the
9
https://datatracker.ietf.org/wg/6tisch
network without informing the other devices so that if
one tries to reach it, no ACK will be received. This
behavior was captured because it violated the security
policy defined for the experiment, but the detection
rule can be changed to avoid generating DDoS detec-
tion alarms for these cases.
5 CONCLUSIONS
The network traffic analysis results presented in this
paper confirm that the proposed framework based on
the MMT-IoT tool behaves as expected in a real IoT
environment. This allowed us to perform an initial
validation of the technology and use the methods de-
veloped to plan a more extensive validation and stress
testing in the future. With this we aim to further
increase the TRL of the MMT-IoT solution, adding
value to the whole MMT suite and expanding the do-
mains of applicability of the Montimage Monitoring
Tool framework. In comparison with other similar
IDS (Intrution Detection System) tools proposed for
IoT/ 6LoWPAN-based WSNs (e.g., SVELTE (Raza
et al., 2013)), MMT-IoT attempts to passively moni-
tor the network based on the sniffer without createing
additional traffic that might be costly in the IoT sys-
tems.
Nevertheless, MMT-IoT’s scalability, as men-
tioned above, depends on the capacity of the sniffer.
Therefore, a more powerful/ dedicated node should be
used for performing the sniffing. In addition, by in-
specting the source code of the sniffer, the processing
time should be proportional to the size of the packet,
due mainly to the copying of memory and writing to
the USB line. It is natural to think that bigger packets
will reduce the number of captured packets, suggest-
ing an inverse relationship between these variables.
Considering this, it is not clear which will be the opti-
mal point that maximises the throughput, considering
that the size of the packet and the number of pack-
ets captured could have a potentially inverse relation-
ship. Also, this leads to a possible evasion: the at-
tacker may saturate the sniffer itself by injecting, for
instance, “big packets”. This needs to be confirmed
by more experimentation and will probably lead to the
development of new evasion avoidance techniques.
ACKNOWLEDGEMENTS
This paper is supported by European Union’s Horizon
2020 research and innovation programme under grant
agreement No 732638, project FED4FIRE+.
A Framework for Security Monitoring of Real IoT Testbeds
651
Figure 12: Anomalies detection in the test with Log-a-Tec testbed.
Figure 13: DDoS attack detection in the test with Log-a-Tec
testbed.
REFERENCES
G. Montenegro, N. Kushalnagar, J. H. and Culler, D. (2007).
Transmission of ipv6 packets over ieee 802.15. 4 net-
works. IETF Tech. Rep. RFC 4944.
La, V. H. and Cavalli, A. R. (2016). A misbehavior node
detection algorithm for 6LoWPAN Wireless Sensor
Networks. In Proceedings of 36th IEEE Interna-
tional Conference on Distributed Computing Systems
(ICDCS 2016), Second IEEE International Workshop
on Security Testing and Monitoring (STAM 2016).
La, V. H., Fuentes, R., and Cavalli, A. R. (2016). A Novel
Monitoring Solution for 6LoWPAN-based Wireless
Sensor Networks. In Proceedings of 22nd Asia-Pacific
Conference on Communications (APCC 2016).
Macaulay, J., Buckalew, L., and Chung, G. (2015). Internet
of Things in logistics. Cisco and DHL report.
Mallouli, W., Wehbi, B., and de Oca, E. M. (2012). On-
line Network Traffic Security Inspection Using MMT
Tool. In Systems Testing and Validation Workshop
2012, pages 23–31.
Raza, S., Wallgren, L., and Voigt, T. (2013). SVELTE:
Real-time intrusion detection in the Internet of Things.
Ad Hoc Networks, 11(8):2661–2674.
Vasseur, J. P., Agarwal, N., Hui, J., Shelby, Z., Bertrand, P.,
and Chauvenet, C. (2011). RPL: The IP routing pro-
tocol designed for low power and lossy networks. In
Internet Protocol for Smart Objects (IPSO) Alliance,
(April):20.
Wehbi, B., Montes de Oca, E., and Bourdelles, M. (2012).
Events-Based Security Monitoring Using MMT Tool.
In IEEE Fifth International Conference on Software
Testing, Verification and Validation (ICST), 2012,
pages 860–863.
ICSOFT 2021 - 16th International Conference on Software Technologies
652