IMPLICATIONS OF MISBEHAVING ATTACKS ON PROBABILISTIC
QUORUM SYSTEM FOR MANETs
Elisa Mannes, Eduardo Silva, Michele Nogueira and Aldri Santos
Department of Informatics, Federal University of Paran´a, Curitiba, Brazil
Keywords:
Quorum system, Wireless self-organized networks, Security, Attack tolerance.
Abstract:
Reliable storage supports different data sharing services, such as mobility and cryptographic key management
and distributed naming. Mobile Ad hoc NETworks (MANETs) present issues in guaranteeing the consistency
of data on concurrent read and write due to dynamism of nodes, the inexistence of a central control entity
and support infrastructure. Probabilistic quorum systems, as PAN (Probabilistic Ad Hoc Quorum System),
were designed for MANETs to improve the efficiency of data replication by relaxing consistency constraints,
comprising a set of quorums with relaxed intersections among themselves. PAN ensures high probability of
consistency between replicated data by an asymmetric quorum construction and by a gossip-based multicast
protocol. However, it does not consider the presence of malicious or selfish nodes in its operations. This work
assesses the impact of lack of cooperation, timeout and data manipulation attacks against PAN. Simulation
results show that PAN is vulnerable to these attacks, particularly, the data manipulation attack.
1 INTRODUCTION
Mobile Ad hoc NETworks (MANETs) comprise mo-
bile devices (nodes) communicating among them-
selves by wireless channels. These networks are
dynamic, infrastructureless, and nodes individually
maintain network operation in a self-organized and
distributed way. Each node communicates directly
with others in its range or by a multihop routing (Luo
et al., 2003). Thus, network services depend on the
cooperation of nodes, being necessary tolerant opera-
tions in face to malicious actions.
MANETs present issues in guaranteeing reliable
storage of data with concurrent read/write accesses.
Reliable storage with reasonable cost and high avail-
ability enables the support to different data sharing
services, such as cryptographic key management and
distributed naming (Tulone, 2007). However, their
characteristics make difficult to provide the synchro-
nism for ensuring a high level of consistency.
To relax the consistency constraints of MANETs
special mechanisms are employed (Luo et al., 2003;
Gramoli and Raynal, 2007; Tulone, 2007). Proba-
bilistic quorumsystems are an example of them. They
improve the efficiency of data replication in MANETs
by better balancing the load between nodes. A quo-
Supported by CNPq/Brazil, grant 303770/2008-2
rum system consists of a subset of nodes with data
replicas, called quorums. The construction of inter-
sections on quorums ensure properties to guarantee
that nodes obtain the most recent value even if queries
or updates are performed on individual quorums.
PAN (Probabilistic Ad Hoc Network Quorum Sys-
tems) (Luo et al., 2003) is a quorum system more ap-
propriated to highly dynamic environments due to the
use of less strict rules for creating intersections among
quorums (Gramoli and Raynal, 2007; Tulone, 2007),
and guarantees a high probability of consistency on
replicated data by a gossip-based multicast protocol.
Albeit PAN’s advantages, it does not consider the
existence of misbehaving nodes in the network. Ex-
amples of misbehavior are the lack of cooperation,
timeout and data manipulation. Misbehaving attacks
can compromise effectiveness and reliability of quo-
rums system, affecting the performance of network
services. Hence, this paper quantifies the impact of
such attacks on the reliability of PAN, highlighting
which vulnerabilities need to be addressed.
We evaluate the implications of attacks on PAN’s
operations by simulations. We use the reliability de-
gree and number of intermediate nodes as metrics to
quantify the effects of lack of cooperation, timeout
and data manipulation attacks on writing and reading
operations. Results under different network mobility
show that PAN is vulnerable to these attacks, and its
189
Mannes E., Silva E., Nogueira M. and Santos A. (2010).
IMPLICATIONS OF MISBEHAVING ATTACKS ON PROBABILISTIC QUORUM SYSTEM FOR MANETs.
In Proceedings of the International Conference on Security and Cryptography, pages 189-195
DOI: 10.5220/0002995701890195
Copyright
c
SciTePress
reliability decreases with higher percentage of attack-
ers. Attacks result in a low reliability of the system,
having its worst performance under the data manipu-
lation attack. Node mobility also affects the reliabil-
ity, and write operations are more compromised by all
attacks than read operations.
This paper proceeds as follows. Section 2 presents
related works. Section 3 describes PAN’s characteris-
tics. Section 4 describes PAN’s vulnerabilities and at-
tacks. Section 5 depicts simulation scenarios and the
metrics used for analysis. Section 6 presents results.
Section 7 presents the conclusions and future works.
2 RELATED WORK
Quorums systems were first introduced as a voting
scheme to ensure data consistency (Gifford, 1979).
Generally they own strict rules for intersection con-
structions and must ensure that at least one quorum is
free of faulty nodes. Recently, authors have proposed
improvements in quorum systems to make them more
flexible and reliable (Herlihy, 1987; Naor and Wieder,
2003). In (Herlihy, 1987), they designed a system
of dynamic quorums, whereas in (Naor and Wieder,
2003), authors created a reconfigurable quorum sys-
tem. However, Friedman et al. (Friedman et al., 2008)
have proved that such approach is not applicable to
MANETs due to the overhead generated by the mes-
sages for reconfigurations. They conclude that the
ideal strategy for MANETs lies on probabilistic quo-
rums because of their capability to relax restrictions
of intersection (Malkhi et al., 2001).
Among the quorums systems proposed for
MANETS, we highlight three of them. The PAN
(Luo et al., 2003) that employs propagation mech-
anisms based in gossip. The timed quorum sys-
tem (Gramoli and Raynal, 2007) that aims to en-
sure intersections during an amount of time, and the
mobile dissemination quorum system (Tulone, 2007)
that uses the geographic location of nodes to con-
struct quorums. Strategies for probabilistic quorums
on MANETs are listed in (Friedman et al., 2008).
Despite those quorums systems consider characteris-
tics as dynamic topology and collaboration, they do
not consider misbehaving nodes, being susceptible to
them.
Misbehaving nodes in MANETs have been ad-
dressed in several works, such as (Hu and Burmester,
2009; Marti et al., 2000), but most of them consider
attacks in routing and application levels. Quorums
system have been evaluated with focus on the Inter-
net (Amir and Wool, 1996; Owen and Adda, 2006),
and for the best of our knowledge, there are no stud-
ies that analyze the behavior of quorum systems for
MANETs facing attacks. Hence, this work quantifies
the impact of misbehaving attacks in PAN in order to
provide directions for improvements on the design of
quorum systems to achieve reliability and robustness.
3 PAN: PROBABILISTIC
QUORUM SYSTEM FOR AD
HOC NETWORKS
A typical quorum system Q comprises a set of quo-
rums that has two distinct properties, consistency and
availability (Malkhi and Reiter, 1997). The consis-
tency property defines that any two quorums must in-
tersect - Q
1
, Q
2
Q , Q
1
Q
2
6=
/
0, whereas the avail-
ability property means that there is at least one opera-
tional and reliable quorum.
In probabilistic quorum systems, quorums
are chosen probabilistically within each inter-
action (Malkhi et al., 2001). This kind of sys-
tems tries to ensure that in a quorum system Q ,
two quorums Q
1
, Q
2
and an access strategy w,
then Q
1
, Q
2
Q , P (Q
1
Q
2
6=
/
0) 1 ε, being
ε a very small value. The access strategy w defines
the way that read and write quorums are constructed.
PAN is an example of probabilistic quorum sys-
tem, ensuring this probability P by gossiping to
create write quorums (Q
w
), and unicast messages to
construct read quorums (Q
r
).
Figure 1: Entities in PAN quorum system.
PAN elects some nodes to form a storage set (StS),
as shown in Figure 1. These nodes are the servers,
whereas the remainder are clients. When servers are
mediating read and write requests, they are the agent
of that specific operation. In both operations, clients
send requests to servers randomly chosen. Servers on
SECRYPT 2010 - International Conference on Security and Cryptography
190
read requests compose read quorums and the servers
on writing operations compose write quorums.
3.1 PAN Operations
PAN provides to clients reading and writing opera-
tions. Servers, acting as agents, are responsible for
receiving and replying requests from clients. In writ-
ing operations, agents deliver updates received from
clients with the collaboration of nodes in the StS. In
reading operations, agents consult a read quorum be-
fore respond to the client. These operations are illus-
trated in Figure 2 and occur as described:
WRITE - as part of the mechanism of writing, all
servers have a buffer that stores the latest update of
each data. Upon receiving a write request, the agent
updates its data and adds it to its buffer. Periodically,
nodes propagate data from the buffer to other nodes.
After some time, all servers have the updated value.
READ - the agent chooses nodes to compose the read
quorum based on predefined read quorum size. Then,
it forwards the request along with its local copy of
the searched data. Thus, members of the read quo-
rum reply to the agent if their data is more updated
than that from the agent. In this case, the agent waits
for replies, and in the lack of them, it responds to the
client with its own data as conclusion that the data it
holds is the most updated. Further, if servers in the
quorum receive a data more updated than theirs, they
revise their local copy and add it to the buffer in order
to be disseminated in the next round of propagation.
Figure 2: Read and write operations in PAN.
To explain PAN’s operation, we assume a net-
work of 20 nodes, in which ten are part of the StS
as illustrated in Figure 2. Node n
4
sends a write re-
quest to node n
8
. Node n
8
, the agent, propagates this
write to other nodes following the gossiping protocol.
Write operations are propagated to a certain number
of servers, being this number called fanout (F). It is
previously configured, and servers follow the same
value.
In Figure 2, F equals two servers and the size of
the read quorum four. Client n
17
issues a read oper-
ation. Then, agent n
18
consults the read quorum in
order to collect the updated value. Nodes n
11
and n
12
participate in both operations, being the intersection
between read and write quorums. Intersection nodes
have the most updated value to client n
17
.
4 ATTACKS ON PAN
The success of writing operations is related to the col-
laboration of all servers involved in the dissemination.
Thus, members of the StS may hinder the progress of
updates in several ways, such as refusing to propa-
gate the updates or making updates to progress slowly.
This can occur if nodes are selfish and want to save
their resources, or if nodes are malicious and want to
degrade the system. Further, nodes can manipulate
values that are being updated, causing inconsistencies
in the system.
Servers consulted by agents in reading operations
may deliberately not respond to requests, forcing the
agent to send its own data to client. They can also
modify the value of data and send this wrong value.
In the second case, generated by active attackers, the
problem is critical because in addition to providing
clients erroneousinformation, servers can trust in data
sent by agents and also update their data.
This work examines particularly three attacks:
lack of cooperation and timeout and data manipula-
tion. Considering the characteristics of PAN, these
types of attacks are more harmful than others for op-
erations in quorum systems. The operation of such
attacks are described below.
4.1 Lack of Cooperation
Lack of cooperation attacks occurs on both opera-
tions. Figure 3(a) illustrates the behavior of a com-
promised agent in a read operation. A client issues a
read request of a data v to server s
0
, which is a selfish
node. As a misbehaving agent, s
0
waits until the op-
eration timeout expires and replies the client with its
own data, that might be outdated.
When the selfish node is an intermediate node, it
simply stops responding to the agent. In Figure 3(b),
the client requests a data v to the server s
0
, that con-
sults the read quorum composed by s
1
, s
2
and s
3
.
Server s
2
is a selfish node and does not respond to
the agent. Under this situation, a reading operation
IMPLICATIONS OF MISBEHAVING ATTACKS ON PROBABILISTIC QUORUM SYSTEM FOR MANETs
191
can still complete correctly if at least one server in
the read quorum responds to the agent or if the agent
itself has the updated data.
Figure 3: Lack of cooperation in read operations.
In writing operations a selfish agent does not up-
date its data or disseminate data provided by the
client. In Figure 4(a), the client issues an update of
data v to s
0
, a selfish node. In this condition, s
0
does
not propagate the update and the write operation does
not complete. As an intermediate server receiving an
update via gossip, the selfish node does not update
neither its local data nor the buffer. As a result, the
propagation of the write operation does not progress,
as shown in Figure 4(b). The client sends a write re-
quest to server s
1
, that propagatesit via gossip. Server
s
2
is a selfish node and does not propagate the data.
Figure 4: Lack of cooperation in write operations.
4.2 Timeout Manipulation
The time interval between the gossip is an important
parameter in PAN. It determines write dissemination
speed, and as faster the propagation is, faster the for-
mation of the write quorum is. In timeout manipula-
tion attack, illustrated in Figure 5, servers in the StS
delay the propagation of updates. In this example, the
client issues a write request to the server s
0
, which
propagates via gossip this update. Server s
2
is a ma-
licious nodes, and arbitrarily delays the propagation
in one round. This behavior makes servers of the StS
take more time to update the data.
4.3 Data Manipulation
Data manipulation attacks consist in nodes that re-
ceive a data and deliberately change its value, in both
read and write operations. As a result, nodes can ac-
cept a manipulated data. Figure 6(a) shows the be-
Figure 5: Timeout manipulation attack.
havior of a misbehaving agent in the read operation.
The agent is the only contact between the client and
the StS, being a vulnerable point.
In the figure, the client requests a data v to the
server s
0
, which is a misbehaving node. Server s
0
changes the data value and timestamp. Then it con-
sults the read quorum, s
1
, s
2
and s
3
. As these servers
have an outdated data compared to the one received
by the agent, they update their values, believing the
data received is the most updated. Servers also store
this value in their buffer, and propagates it via gos-
sip. The read quorum does not respond to the agent
request, since they believe the agent has the most up-
dated data. As the timeout expires, the agent replies
the client with its manipulated data.
Figure 6: Data manipulation in read operations.
If the misbehaving node compose the read quo-
rum, the impact is minimized, as shown in Fig-
ure 6(b). In this example, the client issues a read re-
quest for data v to server s
0
that consults the read quo-
rum. Server s
2
is a misbehaving node and changes its
data and timestamp. When agent s
0
receives answers
from the read quorum, the data received from server
s
2
is the most updated, then agent s
0
updates its own
data, keeps this entry in its buffer and waits to propa-
gate this data. It also responds to the client with this
manipulated data.
In writing operations, malicious nodes can widely
compromise the system. Figure 7(a) illustrates the
case when the agent is a misbehaving node, and can
commits the system faster since any write will result
in an incorrect data. In this example, when client
sends a write request to server s
0
, it changes the value
SECRYPT 2010 - International Conference on Security and Cryptography
192
and the timestamp of the data to be updated, and then
propagates it to other nodes.
Figure 7: Data manipulation in write operations.
A case where the node is an intermediate node is
shown in Figure 7(b). The client issues a write re-
quest to server s
0
, that propagates it to servers. Server
s
2
is a misbehaving node, thus it changes the data and
propagates it to other two nodes. Each server that re-
ceives this data will propagate it and the StS will be
compromised by this manipulated data.
5 EVALUATION MODEL
PAN was implemented on the Network Simulator
(NS-2) version 2.33, and simulations parameters are
based in (Luo et al., 2003). Nodes communicate us-
ing a wireless channel with the TwoRayGround prop-
agation model. The network consists of 50 nodes in
which 25 are servers in StS. Nodes move according
to the Ramdom Waypoint model in an area of 1000m
by 1000m, and have maximum speeds of 2, 5, 10 and
20m/s, with pause time of 10, 20, 40 and 80s, respec-
tively. All nodes own a transmission range of 250m
and use AODV as routing protocol. Updates from the
buffer are propagated every 200ms, and F is set to
two servers. Read quorums are set to four servers.
The percentage of attackers varies in 20%, 28% and
36% of the servers in StS. A larger number of attack-
ers leads the system to a state where quorums are fully
compromised.
As in (Luo et al., 2003), readings and writings
intervals are determined by a Poisson distribution.
Writes have an average interval of 6s while readings
happens in an average of 4s. Results are the average
of 35 simulations with a 95% of confidence interval,
and the lifetime of the network is 1500s.
Two metrics are employed: reliability degree (R
d
)
(Luo et al., 2003) and the quantity of malicious nodes
in read operations (Q
m
). R
d
quantifies the probabil-
ity of intersection between read and write quorums.
R
d
also represents the amount of correct readings ob-
tained by clients. We consider correct reads the ones
retrieved from previously written values, in addition
to the last write. This is because the last value written
may be still in progress inside the StS. Therefore, the
previously written value is also considered correct. R
d
is defined in Eq. (1), where C
r
denotes correctly fin-
ished reads and R the amount of read requests issued
by clients.
R
d
=
C
r
|R|
(1)
Q
m
denotes the number of times that malicious or
selfish nodes participated in reading operations. The
timeout manipulation is excluded because it is carried
out only in write operations. This metric is expressed
by Eq. (2), where M
r
means a read operation in which
a member of the read quorum is malicious.
Q
m
=
Mr
i
|R|
i R, where Mr
i
=
1 if m Q
r
(R
i
)
0 if not
(2)
6 RESULTS
6.1 Lack of Cooperation
Figure 8 shows results obtained when PAN faces
the lack of cooperation attack. Each cluster of bars
presents results for the same value of node speed com-
paring variations on the type of operation and the per-
centage of selfish nodes. We observe that in both
operations, R
d
decreases with the increase of selfish
nodes. For all clusters, decreases in R
d
are more sig-
nificant on write operations than in read operations.
Comparing clusters, we observe that higher node
speed results in lower R
d
. We also observe that R
d
owns the smallest value for writing operations in sce-
narios with 36% of selfish nodes and when nodes have
a maximal speed of 20m/s. This occurs because al-
most half of the StS is compromised.
6.2 Timeout Manipulation
Figure 9 presents results for R
d
when PAN is under
timeout manipulation attacks. Malicious nodes prop-
agate updates every 400ms, 800ms and 3000ms, in-
stead of the normal 200ms. Results are grouped by
the nodes speed, and for each speed, R
d
values under
variations on the percentage of attackers and T are
compared. We observe that R
d
decreases proportion-
ally as the update propagation delay increases. For all
speeds, this behavior is observed.
IMPLICATIONS OF MISBEHAVING ATTACKS ON PROBABILISTIC QUORUM SYSTEM FOR MANETs
193
80
85
90
95
100
20 28 36 20 28 36 20 28 36 20 28 36
Reliability Degree (%)
2m/s 5m/s 10m/s 20m/s
Read
Write
Figure 8: Lack of cooperation.
94
95
96
97
98
99
100
20 28 36 20 28 36 20 28 36 20 28 36
Reliability Degree (%)
2m/s 5m/s 10m/s 20m/s
T=400
T=800
T=3000
Figure 9: Timeout manipulation.
0
5
10
15
20
25
30
35
40
20 28 36 20 28 36 20 28 36 20 28 36
Reliability Degree (%)
2m/s 5m/s 10m/s 20m/s
Read
Write
Figure 10: Data manipulation.
6.3 Data Manipulation
The data manipulation attack also affects PAN. Fig-
ure 10 shows results obtained from the simulation of
this attack. Results are represented by bars, grouped
by the speed of nodes. For each speed, the percentage
of malicious nodes is varied on both operations.
We observe that for all scenarios, R
d
is lower than
50%, and as the percentage of attackers grows, the
reliability decreases. When the nodes move faster,
the R
d
decreases. We notice that these attacks af-
fect significantly write operations in relation to results
achieved under the other two attacks. In this case,
R
d
is lower than 10% for all scenarios. Higher speed
and number of attackers issue the worst values to R
d
.
This attack has an uncommon behavior: as the speed
of nodes increases, the R
d
increases too. This is be-
cause MANET’s characteristic, in which higher node
speeds result in higher routes dinamicity. Such behav-
ior decreases delivered updates and the effectiveness
of a malicious node to propagateits manipulated data.
6.4 Malicious Nodes Participation
It was recorded the number of times that misbehav-
ing nodes participated in read quorums, in read oper-
ations. Figure 11 shows results obtained under lack
of cooperation attack. Increasing the speed of nodes
has the same effect observed in the data manipulation
attack, i.e. as the speed increases, the number of quo-
rums affected by misbehaved nodes decreases.
0
10
20
30
40
50
60
70
80
90
100
2m/s 5m/s 10m/s 20m/s
Misbehavior nodes (%)
f = 20%
f = 28%
f = 36%
Figure 11: Compromised reads in lack of cooperation.
Figure 12 shows the results under the data manip-
ulation attack. The number of readings affected by
misbehaving nodes also increases as the number of
attackers increases. It is observed that nodes speed
interferes in the number of compromised nodes. As
the speed grows, the number of misbehaving nodes
decreases. This occurs because increasing the speed,
less packets are received by nodes and less nodes are
consulted and can deliver manipulated data.
30
40
50
60
70
80
90
100
110
120
2m/s 5m/s 10m/s 20m/s
Misbehavior nodes (%)
f = 20%
f = 28%
f = 36%
Figure 12: Compromised reads in data manipulation.
7 CONCLUSIONS
This paper assessed the impact of misbehaving at-
tacks that affect data availability and consistency on
the PAN system, a probabilistic quorum system for
MANETs. Our analyses examined the implications
of lack of cooperation, timeout and data manipulation
attacks on the read and write operations. Albeit PAN’s
advantages, simulation results reinforced PAN vul-
nerability to attacks resulted from its MANET’s char-
acteristics, PAN’s design and operations over quo-
rums.
We observed that the percentage of malicious or
selfish nodes and node speed affects the reliability of
the system. All attacks influenced the read or write
operations, being write operations the most damaged.
As future work, we plan to propose a model to im-
prove the tolerance of probabilistic quorum systems
to attacks and intend to employ it in the context of the
key management for MANETs.
REFERENCES
Amir, Y. and Wool, A. (1996). Evaluating quorum systems
over the internet. In FTCS, pages 26–35, Washington,
DC, USA. IEEE Computer Society.
SECRYPT 2010 - International Conference on Security and Cryptography
194
Friedman, R., Kliot, G., and Avin, C. (2008). Probabilis-
tic quorum systems in wireless ad hoc networks. In
IEEE/IFIP DSN, pages 277–286, Washington, DC,
USA. IEEE Computer Society.
Gifford, D. K. (1979). Weighted voting for replicated data.
In ACM SOSP, pages 150–162, NY, USA. ACM Press.
Gramoli, V. and Raynal, M. (2007). Principles of dis-
tributed systems, chapter Timed Quorum Systems for
Large-Scale and Dynamic Environments, pages 429 –
442. Springer Berlin.
Herlihy, M. (1987). Dynamic quorum adjustment for parti-
tioned data. ACM TODS, 12(2):170–194.
Hu, J. and Burmester, M. (2009). Cooperation in Mobile
Ad Hoc Networks, chapter 3, pages 1–15. Springer
London.
Luo, J., Hubaux, J.-P., and Eugster, P. T. (2003). PAN:
providing reliable storage in mobile ad hoc networks
with probabilistic quorum systems. In ACM MobiHoc,
pages 1–12, New York, NY, USA. ACM.
Malkhi, D. and Reiter, M. (1997). Byzantine quorum sys-
tems. In ACM STOC, pages 569–578, NY, USA.
ACM.
Malkhi, D., Reiter, M. K., Wool, A., and Wright, R. N.
(2001). Probabilistic quorum systems. The Informa-
tion and Computation Journal, 170(2):184–206.
Marti, S., Giuli, T. J., Lai, K., and Baker, M. (2000). Miti-
gating routing misbehavior in mobile ad hoc networks.
In ACM MobiCom, pages 255–265, NY, USA. ACM.
Naor, M. and Wieder, U. (2003). Scalable and dynamic
quorum systems. In ACM PODC, pages 114–122, NY,
USA. ACM.
Owen, G. and Adda, M. (2006). Simulation of quorum sys-
tems in ad hoc networks. In 3rd International Con-
ference on Artificial Intelligence in Engineering and
Technology.
Tulone, D. (2007). Ensuring strong data guarantees in
highly mobile ad hoc networks via quorum systems.
Ad Hoc Networks, 5(8):1251–1271.
IMPLICATIONS OF MISBEHAVING ATTACKS ON PROBABILISTIC QUORUM SYSTEM FOR MANETs
195