Moving Target Defense Router: MaTaDoR
Berkan Ufuk
1,2 a
and Mehmet Tahir Sandikkaya
1 b
1
Computer Engineering Department, Istanbul Technical University, TR34469, Turkey
2
Aselsan Inc., Ankara, Turkey
Keywords:
Moving Target Defense, TCP-authentication Option, Transparent Routing, Message Authentication, DoS,
Honeynets, Honeypot.
Abstract:
The continuous increase in network attacks and the complexity of the available offensive technologies enforces
novel defensive mechanisms. Moving Target Defense (MTD) is a recent family of approaches for network
defense. This study proposes MaTaDoR, which utilizes message authentication akin to TCP Authentication
Option (TCP-AO) in a MTD setting to mitigate a wide range of attacks, including Denial of Service (DoS). The
purpose of MaTaDoR is averting unauthenticated packets from reaching protected assets. When many other
MTD approaches aim to delay adversaries, MaTaDoR strictly protects networked assets from unauthenticated
access. MaTaDoR is transparent, stateless and scalable. The efficiency of this combination is demonstrated
by the results of a simulation. The proposed approach is capable of blocking every DoS packet with an
insignificant trade-off increase in end-to-end delay.
1 INTRODUCTION
Although there are numerous network defense mech-
anisms, the number of attacks is increasing. New de-
fense mechanisms evolve hand in hand with new at-
tack strategies. Generally, the motivation behind a de-
fense mechanism is to make an attack relatively costly
for an adversary. Moving Target Defense (MTD)
is one of the mechanisms in order to protect a net-
worked service from attacks. MTD aims to confuse
adversary by manipulating network attributes in order
to render attacks infeasible. This study contributes
by proposing eager message authentication for net-
works that utilize MTD. The proposed solution is es-
sentially similar to chaffing and winnowing (Rivest
et al., 1998). Host-independent message authentica-
tion chaffs stray TCP connections. TCP Authenti-
cation Option (Touch et al., 2010) has common us-
age in networks linked with Border Gateway Protocol
(BGP). The hypothesis is that considerable amount of
unnecessary requests could be prevented prior to es-
tablishing connections if MTD with an authentication
header (abbreviated as MaTaDoR Moving Target
Defense Router) is utilized. Selection and elimination
of packets are possible in the beginning of the TCP-
handshake step by inspecting how the SYN packets
are constructed. Therefore, only genuinely message-
a
https://orcid.org/0000-0002-1416-2866
b
https://orcid.org/0000-0002-9756-603X
authenticated SYN packets would be apt to be pro-
cessed and other packets would be directed to a decoy
network. The proposed mechanism resides in a trans-
parent box to route the traffic to either the genuine or
the decoy network that have equivalent structure. The
deployment of the proposed mechanism is shown in
Figure 1.
The mechanism is effective against all kinds of
unauthorized access attempts. DoS, where such be-
havior is observed frequently, forms a valid demon-
stration. Therefore, the effectiveness of the hypoth-
esis is shown by simulating the mechanism with re-
generated real-world network traffic which is known
to include DoS attacks.
MaTaDoR is a router, rather than a gateway. It
does not keep the state of the connections that are
passing through so that how the unwanted connec-
tions could be relayed or dropped before handshakes
are completed. Therefore, MaTaDoR itself is resistant
against overflows. Since the transparent entry point in
the proposed mechanism acts as a router, it does not
form a single point of failure. Scalability is achieved
by adding or removing MaTaDoR boxes. Central key
management is not necessary as MaTaDoR is state-
less.
The proposed approach is compared with ongoing
research in Section 2. Section 3 sketches a typical use
case of the proposed solution. Re-generation of the
publicly available real-world traffic for the conducted
Ufuk, B. and Sandikkaya, M.
Moving Target Defense Router: MaTaDoR.
DOI: 10.5220/0011306500003283
In Proceedings of the 19th International Conference on Security and Cryptography (SECRYPT 2022), pages 649-654
ISBN: 978-989-758-590-6; ISSN: 2184-7711
Copyright
c
2022 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
649
Legitimate client with
correct message
authentication key
tries to connect
genuine server at
address 1.2.3.4
Adversary without correct
message authentication
tries to connect genuine
server at address 1.2.3.4
MaTaDoR boxes
are transparent
routers that can
be scaled easily
Genuine web server
with IP address 1.2.3.4
Decoy web server with
IP address 1.2.3.4
Legitimate client s messages
are relayed to the genuine
subnet
Adversary s messages are
relayed to the decoy subnet
GENUINE SUBNET
DECOY SUBNET
INTERNAL PROTECTION DOMAIN
Figure 1: MaTaDoR is able to distinguish the legitimate
packets from malicious ones even though their observable
behavior is the same for an external observer.
simulation and reproducible experimental steps are
explained in Section 4. The results of the conducted
simulations for the proposed cost-effective mecha-
nism are tabulated in Section 5. Section 6 concludes
the paper with a summarizing discussion.
2 RELATED WORK
Hong, et al. classifies MTD techniques into three cat-
egories: Shuffle, Redundancy and Diversity (Hong
and Kim, 2015). Shuffling moves or re-arranges the
resources to be protected. Redundancy replicates the
resources to enhance availability. Diversity places dif-
ferent variants of a resource with the same function-
ality. According to this taxonomy, proposed mecha-
nism is a diversity mechanism as the adversary is di-
verted to a non-functional decoy.
Chaffing and winnowing is a technique for
sharing confidential information in an untrusted
medium (Rivest et al., 1998). Chaffing and win-
nowing was not designed as a MTD approach. It is
designed to provide confidentiality through message
authentication instead of encryption. The proposed
mechanism was inspired by chaffing and winnowing
as message authentication is used as an effective filter
to distinguish legitimate or malicious traffic.
Bogosyan et al. uses MTD to do defend against
spoofing and replay attacks to Controller Area Net-
work (CAN) bus during electric vehicle charging pro-
cess (Bogosyan et al., 2020). This study also estab-
lishes redundancy MTD by multiplying resources.
Jenkins et al. design a MTD algorithm for real
time networks in space systems (Jenkins et al., 2021).
This study makes use of authentication bytes and dis-
tributes correct shuffling keys to authenticated clients.
This study designs a shuffling MTD.
Park, et al. propose Ghost MTD, where pre-
shared one time bit sequences are distributed and
non-conforming sequences are directed to some de-
coy module (Park et al., 2020). Park, et al.s approach
resembles the proposed mechanism; however, distri-
bution and management of one time bit sequences for
each host seems impractical.
Kampanakis et al. implement Moving Target
Defense mechanism in a Software-Defined Network
(SDN) environment, then evaluate the performance
of the system with Cisco’s One Platform Kit (Kam-
panakis et al., 2014). Even though the authors effec-
tively use SDN’s flexibility, it is difficult to adopt this
approach to traditional networks.
Openflow Random Host Mutation (OF-RHM) is
also based on SDN (Jafarian et al., 2012). Virtual ad-
dresses of network resources are continuously mod-
ified by the SDN controller via Openflow protocol.
This is an example of a shuffling mechanism. The
IP configuration of the network domain is constantly
changing. On the other hand, the proposed mecha-
nism has two different domains which have the same
configuration from an outsider’s point of view.
Network-based MTD (NMTD) constantly
changes the network configuration such as IP ad-
dresses and ports that are used by the protected
applications (MacFarland and Shue, 2015). This
setup requires a centralized configuration to follow.
However, this might introduce a single point of
failure. On the other hand, MaTaDoR could be
multiplexed without burden.
Luo, et al. proposed Random Port and Address
Hopping (RPAH) scheme that predefines the active
ports (Luo et al., 2017). A gateway handles the gen-
eration and management of virtual IP addresses. Luo
et al.s NAT implementation is similar to the proposed
mechanism. However, the transparent entry point in
the proposed mechanism acts as a router and does not
form a single point of failure.
3 PROPOSED MECHANISM:
MaTaDoR
The users of a network expect availability. A network
is constantly subject to attacks that jeopardize avail-
ability. Therefore, networks should distinguish adver-
saries from legitimate users. In the proposed mecha-
nism adversarial packets are routed to a decoy net-
work right after they are distinguished. Thus, even if
an attack is going on, legitimate users are served. Per-
ceiving a decoy but equivalent service is confusing for
adversary.
Considering these, MaTaDoR is designed as a
SECRYPT 2022 - 19th International Conference on Security and Cryptography
650
transparent router. The decoy network in this study is
presented as a placeholder and is not detailed. Else, it
is possible to use this network to redirect adversaries
or observe their actions further on a honeynet.
Slowloris and similar DoS attacks harm the avail-
ability of a web server even before connections are es-
tablished. Since the attack traffic will not include the
correct message authentication keys, MaTaDoR for-
wards this traffic into a decoy server. Even if the de-
coy server rendered unavailable, the legitimate users
of the genuine server are not affected. However, from
the adversary’s point of view, disrupting the service of
the website make them think their attack is success-
ful at first sight. Such additional confusion is another
layer of ambiguity for the adversary.
The relaying nature of MaTaDoR is transparent as
seen in Figure 1. The sole functionality of MaTaDoR
is to verify the message authentication header and de-
termine the future flow of the traffic. A realistic sce-
nario is to place MaTaDoR in between two organiza-
tions that share non-public network services without a
dedicated VPN (Virtual Private Network) or a leased-
line. Message authentication key is assumed to be
set up during initialization. Further key management
issues are deliberately keep out of the scope of the
discussion. A second, decoy server is also placed in
Figure 1 to handle attacker attention whenever MaTa-
DoR fails to authenticate the message. Moreover, the
decoy server could be replaced with an honeynet that
logs attacker behavior or non-verified segments could
simply be dropped.
The network topology presented in Figure 1 rep-
resents the aforementioned scenario with two clients,
two web servers and MaTaDoR. One of the clients
(marked red in Figure 1) represent the malicious IP
space where DoS attack is originating from. The
packets from the malicious IP space have missing or
incorrect MAC. The other client (marked blue in Fig-
ure 1) represent the benign users who want to use
the web server. The genuine web server represents
the protected resource. Finally, the decoy web server
is equivalent to the genuine resource, but consists of
mechanisms to distract adversarial efforts. As in ev-
ery other MTD solution, the adversary is hopefully
tricked to think that no defense mechanism exists to
protect the asset.
The flow is shown in Figure 2a when a segment
is authenticated. MaTaDoR verifies the authentica-
tion header, then directs the traffic to the web server
if the MAC verification is successful. The genuine
web server gets the segments and continues by send-
ing an acknowledge message to establish TCP hand-
shake with the authenticated client. The genuine web
server can directly see the initiator IP address of the
client, as MaTaDoR is transparent. Thus, the server
can directly send the response to that client. There-
fore MaTaDoR allows to establish the communication
of legitimate users and the genuine web server.
Whenever a packet authentication fails, MaTaDoR
redirects the packet to the decoy server. The flow of
unauthenticated segments is shown in Figure 2b. De-
coy server can also see the IP address of the initiat-
ing client. Therefore, the adversary might believe as
if they are directly connected to the server, without
any interceptors. Thus, unauthenticated or unwanted
messages are directed to the decoy server and genuine
server can continue its normal operation.
A DoS attack example in which the adversary
never finishes TCP handshake to waste the resources
of benign servers is shown in Figure 2c. In this sce-
nario, like in Slowloris attacks, the adversary sends
only “SYN” packets of the handshake and never re-
sponds to incoming “SYN/ACK” packets. This be-
havior makes serving computers to wait for an incom-
ing connection, by allocating some of its resources
to keep state. The increase of such requests makes a
web server to allocate more resources and eventually
the server becomes unavailable. Since MaTaDoR for-
wards the requests to an actual, yet decoy web server,
the decoy server becomes unavailable. Thus, MaTa-
DoR serves its purpose by confusing the adversary ei-
ther the web server is on or off. Genuine server oper-
ates normally in the meantime.
Note that, both benign and malicious traffic is han-
dled the same from the external network’s point of
view. Therefore, it is not possible to detect either the
genuine or decoy server is active. This attribute al-
lows users to be unaware of that there is a protection
mechanism in place.
4 METHODOLOGY
MaTaDoR’s feasibility is validated by a simulation
setup similar to Figure 1. A real network dump is re-
generated with Scapy (Rohith et al., 2018) to build
experimental traffic. Scapy is a packet manipula-
tion program that enables real time header processing.
Since popular operating systems like GNU/Linux, OS
X and Windows do not support TCP-AO message au-
thentication header is added at the client side with
Scapy. Then, MaTaDoR handles these headers via
custom iptables rules.
The original traffic data is named “SUEE1”
1
, in-
cludes both wireless and wired interfaces, and col-
lected by Lukaseder et al. (Lukaseder et al., 2018).
1
https://github.com/vs-uulm/2017-SUEE-data-set
Moving Target Defense Router: MaTaDoR
651
SYN + Correct
MAC
SYN
LEGITIMATE
CLIENT
MaTaDoR
GENUINE
SERVER
Time
SYN/ACK
SYN/ACK
ACK
ACK
DATA
DATA
1 - MaTaDoR verified the segment and forwarded
the client to the genuine server
2- Handshake with the genuine server is
completed
1
2
(a) Communication flow of
authenticated messages that are
destined to genuine server.
SYN +
Incorrect/Nonexistent
MAC
SYN
ADVERSARY MaTaDoR
DECOY
SERVER
Time
SYN/ACK
SYN/ACK
ACK
ACK
DATA
DATA
1 -MaTaDoR did not authenticatethe message and
forwardedthe client to the decoy server
2-Handshake with thedecoy server is completed
1
2
(b) Communication flow of
unauthenticated messages that are
destined to genuine server.
SYN +
Incorrect/Nonexistent
MAC
SYN
ADVERSARY MaTaDoR
DECOY
SERVER
Time
SYN/ACK
SYN/ACK
1 -MaTaDoR cannot authenticatethe segment and
forwardedthe adversary to the decoy server
2- Decoy Server responds to SYN packets
3- Decoy Server becomes unavailable so that the
adversary believes the attack has succeded. In
the meantime, MaTaDoR is still available as it is
stateless. Legitimate users are able to use the
genuine server.
SYN
SYN +
Incorrect/Nonexistent
MAC
SYN/ACK
2
1
3
(c) Communication flow of a DoS
Attack.
Figure 2: Characteristic packet flows of MaTaDoR.
The traffic includes both incoming and outgoing
HTTP/S packets. The malicious packets are sepa-
rately generated by running slowloris and slowhttptest
DoS attacks. The malicious source addresses are cho-
sen from a pre-determined IP space during the experi-
ments so that traffic statistics could be compiled easily
at the target.
The experimental traffic uses exact packet timing,
order and size of the original. Besides, the packet con-
tents are masked and the traffic is directed to the target
host either with enabled authentication or not. The
included malicious traffic is also an exact copy of a
pre-recorded DoS attack.
The overall traffic data includes a total of
2 089 436 packets, where 1 962 955 of them are be-
nign requests and 126 481 of them are DoS attempts.
When the DoS attack is conducted for any of the
web servers, it takes about 15 seconds to disrupt their
services. It is possible to observe this by not receiv-
ing any acknowledgment messages from the servers.
MaTaDoR does not alter the original packet, it only
marks these packets based on the hash-based mes-
sage authentication codes (HMACs) (Krawczyk et al.,
1997). Then the marked packets are forwarded to
genuine or decoy server depending on its authentic-
ity. The proposed approach is suitable and compatible
to be used by utilizing the optional TCP-AO header.
However, it is omitted for two resasons. First, it is
known that BGP frequently uses this option in be-
tween autonomous systems (ASs) and this may cause
that the proposed message authenticated packets be
filtered via BGP routers. This is the main reason
that the provided implementation allocates another
optional header area. Still, MaTaDoR could safely
be used via TCP-AO header within the same AS. Sec-
ond, end user devices do not support TCP-AO by de-
fault. It is more practical to utilize optional additional
header.
In the final design, benign and malicious traffic
are processed together, both with message authenti-
cation header and not, to measure the overhead gener-
ated by the proposed solution. In order to observe the
efficiency of the mechanism and observe the effects
of the attacks in the experiments, the server software
chosen to be vulnerable against DoS attacks. This is
a natural requirement that should be satisfied to see
that unwanted or unauthenticated packets never reach
the genuine server. In this expreriment, unwanted
traffic is the slowhttptest, slowloris and slowloris-ng
packets from the SUEE1 dataset. The DoS vulnera-
ble server software is chosen as the built-in Python2
module called SimpleHTTPServer. As this software
has limited connection capabilities, it is easy to know
that if it is disrupted by simply checking its status or
by observing whether it returns an acknowledgment
packet.
MaTaDoR’s source code is available on Github.
2
5 RESULTS
The overhead of using authentication header and the
additional redirection is measured to present the per-
formance of the proposed solution. Built-in time com-
mand is utilized to measure the execution time of
authentication verification. Moreover, the additional
2
https://github.com/BerkanUfuk/MaTaDoR
SECRYPT 2022 - 19th International Conference on Security and Cryptography
652
load of the verification to the CPU is collected.
The experiment is conducted on a Raspberry Pi
3 Model B Plus Rev 1.3 and two regular personal
computers with Intel Core i7-7700HQ CPU and 8 GB
RAM. One of the computers have the web servers and
named as server computer, and the other computer
has client software and named as client computer.
Raspberry Pi resides as the MaTaDoR in the experi-
ments. The server computer has 2 virtual machines
which have identical web servers constructed with
python’s http.server module. Also, these two virtual
machines have the same operating system and config-
uration. The only difference is that they are on differ-
ent interfaces and on different physical sub-nets that
have the same network configuration. client comput-
ers also have the same operating systems and placed
at the other end of MaTaDoR. The clients in the exper-
iments are GNU/Linux virtual machines with Scapy
to send packets.
Two scenarios are tested to point out the perfor-
mance difference of using authentication or not. The
experiment is repeated five times and results are pro-
vided as the arithmetic mean of each. Table 1 sum-
marizes the perspective of the client computers. Ac-
cording to the table, the proposed solution has limited
effect on the CPU load. The overall delay is notice-
able but the slight increase seems acceptable.
The columns shown on the Table 1 are as follows:
The Real column indicates the total amount of time
to process the traffic during the experiment. Simi-
larly, User and Kernel indicate the amount of time
consumed in user and kernel space. CPU indicates
the mean of additional CPU load throughout the ex-
periment. Finally, RTT is the round trip time to send
request from the client to the either of the servers, and
receive a response. Recall that, there are 1 962 955
benign and 126 481 malicious packets exist in the re-
generated pre-recorded real world traffic.
The total time spent for each benign packet is
around 25 ms. This is around 7 ms for each mali-
cious packet. This difference is due to the additional
response originated from the application layer pro-
cessed for benign packets. Therefore, this difference
cannot be considered as an additional cost of MaTa-
DoR, but as the natural outcome of using the network
resources.
The overhead of additional message authentica-
tion header for benign traffic is approximately 8% and
the difference in the CPU load is negligible. The over-
head is approximately 2% for malicious traffic. These
results indicate that MaTaDoR effectively and effi-
ciently filters out the malicious or benign traffic and
the overhead is depend mostly on the packet size. The
accuracy of this filtering is 100% as MaTaDoR relies
on cryptography unless a noise disrupts a packet.
Table 2 represents the average CPU load and addi-
tional CPU loads as observed from the MaTaDoR. In
order to measure the CPU load, the built-in top com-
mand is used in the MaTaDoR. As it can be seen,
MaTaDoR brings up an extra 4.3% CPU cost. This
is due to IPTables rules that MaTaDoR builds.
MaTaDoR also has an insignificant effect in
packet delays. The average time of packet delays with
MaTaDoR is 597 microseconds, while the average
packet delay is 487 microseconds without MaTaDoR.
The RTT delay is measured with Scapy by subtracting
acknowledgment response time from packet sending
time. Additionally, 90 microseconds of RTT differ-
ence seems unnoticeable by the users of a network. It
could be speculated that this difference mainly roots
from the additional router hop. This slight differ-
ence better demonstrates the undetectability of the
proposed solution. It could be seen that the transpar-
ently MaTaDoR barely effects the network behavior.
Therefore, it is very hard to notice if an additional de-
fense mechanism is in use in the network. It could
also be seen that, MaTaDoR has a very limited CPU
usage
Table 3 presents a comparison of the pro-
posed scheme with GhostMTD (Park et al., 2020).
GhostMTD is chosen as its quantitative experimental
results are available. Note that, GhostMTD is similar
to MaTaDoR as both relies on message authenticat-
ing bit streams at the network border and they both
forward invalid packets to decoy servers. However,
GhostMTD requires a connection to be established to
analyze incoming packets and keeps local state, so it
is not scalable.
The results presented in the table are normal-
ized as experimental setups are different. MaTaDoR
introduces around 2.86 % throughput performance
loss while GhostMTD decreases the performance by
3.84 %. MaTaDoR’s efficiency is likely to be the re-
sult of its use of authentication via TCP header early
before establishing the connections. Thus, packets are
handled early without further inspection in the pay-
load.
6 CONCLUSION
In addition to conventional network security prac-
tices, availability is a critical and expected fea-
ture. MaTaDoR chaffs away unauthenticated traf-
fic smoothly by introducing message authentication.
Therefore, it provides strict access control and avail-
ability at the same time. MaTaDoR acts as a trans-
parent router with authentication key based filtering
Moving Target Defense Router: MaTaDoR
653
Table 1: End to end performance metrics of the client computers.
Traffic Real [s] User [s] Kernel [s] CPU [%] RTT [µs]
Benign w/o MaTaDoR 49159 827 4803 13 512
Benign w/ MaTaDoR 52647 1534 5132 14 598
Malicious w/o MaTaDoR 1332 12 186 16 462
Malicious w/ MaTaDoR 1467 32 365 17 631
Table 2: CPU load comparison.
Traffic CPU [%] CPU Difference [%]
w/ MaTaDoR 0.7 4.25
w/o MaTaDoR 0.7 none
Table 3: Throughput comparison of GhostMTD vs. MaTa-
DoR.
GhostMTD MaTaDoR
Loss [%] 3.84 2.86
capabilities. As a result, it is stateless and easily scal-
able. This feature could be helpful to interconnect two
semi-trusted organizations without the necessity of a
leased-line or a VPN. Moreover, such a mechanism is
not directly detectable. This is useful to lure adver-
saries to honeynets, which is not discussed in depth in
the paper and left as a future work.
It is obvious that a practical authentication key ex-
change will be necessary to increase the usability of
the proposed mechanism. The authors aim to conduct
a complexity study after further refining the provided
code.
REFERENCES
Bogosyan, S., Akgul, T., and Gokasan, M. (2020). Mtd
based novel scheme for bms security against can
bus attacks during bev charging. In 2020 9th
Mediterranean Conference on Embedded Computing
(MECO), pages 1–7. IEEE.
Hong, J. B. and Kim, D. S. (2015). Assessing the effective-
ness of moving target defenses using security models.
IEEE Transactions on Dependable and Secure Com-
puting, 13(2):163–177.
Jafarian, J. H., Al-Shaer, E., and Duan, Q. (2012). Openflow
random host mutation: Transparent moving target de-
fense using software defined networking. In Proceed-
ings of the First Workshop on Hot Topics in Software
Defined Networks, HotSDN ’12, pages 127–132, New
York, NY, USA. Association for Computing Machin-
ery.
Jenkins, C., Vugrin, E., Manickam, I., Troutman, N., Hazel-
baker, J., Krakowiak, S., Maxwell, J., and Brown,
R. (2021). Moving target defense for space systems.
In 2021 IEEE Space Computing Conference (SCC),
pages 60–71. IEEE.
Kampanakis, P., Perros, H., and Beyene, T. (2014). Sdn-
based solutions for moving target defense network
protection. In Proceeding of IEEE International Sym-
posium on a World of Wireless, Mobile and Multime-
dia Networks 2014, pages 1–6.
Krawczyk, H., Bellare, M., and Canetti, R. (1997). HMAC:
Keyed-Hashing for Message Authentication. Request
for Comments 2104, Fremont, CA, USA: Internet En-
gineering Task Force.
Lukaseder, T., Maile, L., Erb, B., and Kargl, F. (2018).
Sdn-assisted network-based mitigation of slow ddos
attacks. In International Conference on Security and
Privacy in Communication Systems, pages 102–121.
Springer.
Luo, Y.-B., Wang, B.-S., Wang, X.-F., Zhang, B.-F., and Hu,
W. (2017). Rpah: A moving target network defense
mechanism naturally resists reconnaissances and at-
tacks. IEICE Transactions on Information and Sys-
tems, E100.D(3):496–510.
MacFarland, D. C. and Shue, C. A. (2015). The sdn shuffle:
Creating a moving-target defense using host-based
software-defined networking. In Proceedings of the
Second ACM Workshop on Moving Target Defense,
MTD ’15, pages 37–41, New York, NY, USA. Asso-
ciation for Computing Machinery.
Park, J.-G., Lee, Y., Kang, K.-W., Lee, S.-H., and Park,
K.-W. (2020). Ghost-mtd: Moving target defense via
protocol mutation for mission-critical cloud systems.
Energies, 13(8).
Rivest, R. L. et al. (1998). Chaffing and winnowing: Confi-
dentiality without encryption. CryptoBytes (RSA lab-
oratories), 4(1):12–17.
Rohith, R., Moharir, M., Shobha, G., et al. (2018). Scapy-
a powerful interactive packet manipulation program.
In 2018 international conference on networking, em-
bedded and wireless systems (ICNEWS), pages 1–5.
IEEE.
Touch, J., Mankin, A., and Bonica, R. P. (2010). The tcp
authentication option. Request for Comments 5925,
Fremont, CA, USA: Internet Engineering Task Force.
SECRYPT 2022 - 19th International Conference on Security and Cryptography
654