Analysis and Enhancements of Leader Elections algorithms in Mobile Ad Hoc Networks

ACEEE Int. J. on Network Security , Vol. 02, No. 04, Oct 2011

Analysis and Enhancements of Leader Elections
algorithms in Mobile Ad Hoc Networks
Mohammad H. Al Shayeji1, AbdulRahman R. Al-Azmi2, AbdulAziz R. Al-Azmi3 and M.D. Samrajesh4
Computer Engineering Department, Kuwait University, Kuwait
1
alshayej@eng.kuniv.edu.kw, 2raphthorne@yahoo.com, 3fortinbras222@hotmail.com, 4sam@differentmedia-kw.com

Abstract - Mobile Ad Hoc networks (MANET), distinct from the related work, and section III provides background
traditional distributed systems, are dynamic and self- information on leader election algorithms. The factors for our
organizing networks. MANET requires a leader to coordinate analysis are described in section IV, section V presents the
and organize tasks. The challenge is to have the right election
algorithm that chooses the right leader based on various
factors in MANET. In this paper, we analyze four leader
election algorithms used in mobile Ad Hoc Networks. Factors
considered in our analysis are time complexity, message
complexity, assumptions considered, fault tolerance and
timing model. Our proposed enhancements include recovered
nodes inquiring about the current leader and the use of
candidates during election to reduce the overhead of starting
a new election session. In addition, better election criteria
specific to MANET, such as battery life and signal strength,
Figure 1. Typical leader coordination
are proposed. Our evaluation and discussion shows that the
proposed enhancements are effective. The analysis can be analysis and the proposed enhancements, and evaluation and
used as a reference for system designers in choosing the right discussion on the analysis is presented in section VI. Finally, in
election algorithm for MANET. section VII, we present our concluding remarks and suggestions
for future work.
Index Terms—Distributed Algorithms, Fault Tolerance,
Leader Election, Mobile Ad Hoc Networks II. RELATED WORK
I. INTRODUCTION Despite leader election being a classical problem, few researchers
have given importance to the leader election process in context
Distributed systems are the backbone of modern day of MANET. Leader election algorithms have different
computing services. An election algorithm elects a leader to assumptions as some algorithms use synchronous message
coordinate and organize tasks in distributed systems that transfer [8], and others use asynchronous message transfer[7][8].
includes Mobile Ad Hoc networks (MANET). In contrast to The leader election process has been studied and compared
traditional distributed systems, nodes arrive and leave MANET based on its complexities in [9], and the paper shows the
more frequently. In the case of a leader node departure or failure, relationship between the number of messages and the time taken
nodes detecting the non-availability of the leader initiate a leader for electing a leader additionally, it addresses the question of
election process to select a new leader. This election process the trade-off between the time and message complexities.
should be completed in a finite number of steps with a consensus However, the algorithms are limited to distributed systems. A
among the nodes on the new leader [24]. Leader election is a survey on Service Discovery Protocols (SDP) in MANET is
primary control problem in both wireless and wired systems [1]. presented in [10]. The survey discusses the features of several
In wireless networks, leader has a variety of applications SDPs including service discovery architecture, search methods,
including key distribution [2], routing coordination [3], sensor mobility support and service description methods; however,
coordination [4], and general control [5][6]. In this paper, we the study is not specific to leader election algorithms. MANET’s
analyze four leader election algorithms applied in MANET using ability to continue its task uninterrupted in spite of attacks or
common factors such as time complexity, message complexity, intrusions is presented in [11]. The paper identifies the
assumptions considered, fault tolerance and timing model. Our survivability properties and categorizes it into three groups:
proposed enhancements include waking up nodes that send an route discovery, data transmission and key management. Finally,
inquiry message to their functioning neighbors to indentify the the paper recommends that a survivable MANET needs to
current leader instead of initiating a costly new election. In consider a multi-layer and multi-attack solution. However, the
addition, we propose utilizing candidates during election to paper deals only with the security aspect of MANET. A survey
reduce the overhead, and use better election criteria specific to on use of Genetic Algorithms (GA) for QoS routing in MANET
MANET such as battery life and mobility. Moreover, we propose is discussed in [12]. The paper does a comparative study
to use the CSMA/CD algorithm to handle multiple diffusion between the two GA-based algorithms GAMAN and GLBR.
computations since it requires a fewer number of messages. The The performance evaluation that was done using simulations
organization of the paper is as follows: section II describes shows that the GAMAN algorithm is a better and more
© 2011 ACEEE 19
DOI: 01.IJNS.02.04.37


promising algorithm for QoS routing in Ad-Hoc networks. participating nodes.
However, the survey concentrated only on the QoS routing  Time Complexity: The number of steps required to reach
in Ad-Hoc Networks. Mobile nodes that elect a leader within consensus on the newly elected leader measures time complexity.
a bounded time is presented in [27]. The algorithm does not
rely on the knowledge of the number of nodes in the system  Message Complexity: The number of messages required
nor on a common start-up time. It elects a leader within a by nodes to elect a new leader measures message complexity.
bounded time upon mobile nodes entering a specific region.
B. Assumptions
However, this approach is restricted to a region and has a
bounded communication diameter. An analysis of leader Assumptions include all the pre-conditions required for the
election algorithms in MANET is presented in [28]. The paper algorithm to work. Some algorithms assume reliable channels
attempts to reduce the number of leader election processes [18], FIFO (First in First Out) channels, or nodes that know
and make the system more energy efficient. The algorithm information about their neighbors.
proposes that each node should maintain a list of candidates C. Fault Tolerance
to minimize the total number of leader elections. However,
Discussions on fault tolerance techniques are presented in
energy consumption aspect is only considered in the
[19][20][6]. Most MANET algorithms must deal with network
analysis.Most of the above work focuses on the general
partitioning and merging.
aspects of MANET while our work focuses on the specific
leader election problem in MANET. Additionally, our analysis D. Timing Models
and enhancement provide a reference point for system Timing in distributed systems is not global as there is no
designers to choose a right election algorithm in MANET. global clock; distributed algorithms use synchronous
assumptions to achieve synchrony [18][3][21].
III. BACKGROUND
A. General Leader election V. ANALYSIS AND ENHANCEMENT
Distributed systems require a coordinator/leader to
A. Research of Asynchronous Leader Election Algorithm
coordinate and organize tasks such as implementing mutual
on Hierarchy Ad Hoc Network[22]
exclusion. Leader election is required when the system is
initialized, or the leader crashes. A leader election algorithm 1) Analysis
elects a leader from among the nodes considering application- Initially, the network establishes a hierarchy within election
specific conditions. The key to the leader election problem is groups as shown in Fig. 2. The algorithm has two phases: the
attributed to G. LeLann [13]. He formalized a method to create election phase and leader announcement phase. a) Election
a new token when the token is lost in a ring network. The phase: When a node or several nodes detect the loss of a
token holder can transmit on the medium, and this token leader, each wait for a given time-out. After the time-out
holder represents leader that coordinates tasks in distributed expires, the node sends an election message. The other node,
systems. Primary leader election algorithms are the Bully upon the receipt of the message, forwards it to its unvisited
Algorithm and Invitation Algorithm [8]. neighbors with highest priority. Nodes that have received
the election message are flagged to avoid duplicate messages.
B. Mobile Ad Hoc Network When a node has no neighbors, it will resend the election
MANET is an Ad Hoc network, Ad hoc is a Latin phrase message back to sender. b) Announcement phase: Similar to
meaning “for this” [14] and no topology exists among the the election phase, but in reverse order, the last unvisited
communicating nodes that are connected [15]. MANET is highly node transfers the leader message, containing leader identity,
decentralized and need a coordinator to organize their work. to all the other nodes. Message handling is the same as in
The coordinators are leaders of the sub-network that control the election phase, but here the node forwards the leader
routing, distributing the workload, and synchronizing the other message to all of its neighbors. This phase terminates when
nodes. These aspect help MANETs deal with dynamic changes all the nodes have been visited by the leader message.
simultaneously and thus MANETs are one of the best flexible
networks available. Applications of MANET include wireless
sensor networks, military and industrial unmanned robots, and
more.

IV. FACTORS FORANALYSIS
We establish common factors for the analysis of the four
algorithms. These factors reflect major design aspects that
specify the advantages for each algorithm [16][17].
A. Complexities
Time and message complexities depend on the number of
Figure 2: Hierarchy Ad Hoc Network
© 2011 ACEEE 20
DOI: 01.IJNS.02.04.37


When a node is separated from the network during the election,
it is not allowed to join the election process. However, if the
node sends a ‘hello’ message, receiving nodes compare it with
its Cluster ID and will accept or deny. The paper uses extrema
finding diffusion computation and leaders are chosen
considering the system’s most beneficial choice. Moreover, this
approach does not consider the case, in which leader node is
separated during the election process. The response is to have
another election.
2) Enhancements
a) Nodes create clusters based on their locality and closeness
in vicinity. Elections are conducted in tiny clusters, and then
cluster’s newly elected leaders broadcast their identity to
other clusters similar to the approach in Garcia’s Invitation
Algorithm [8]. The performance is better since the initial leader
election is in small groups.
b) Un-visited nodes should not be allowed to join the election Figure 4: Spanning tree shrinkage
process until it is completed since this increases the overhead their children and declare the sender as their parent. When
on election. the source receives an acknowledgement, it diffuses a ‘leader’
message to the entire newly constructed spanning tree with
B. Design and Analysis of a Leader Election Algorithm
the information of the current new leader.
for Mobile Ad Hoc Networks[1] As the number of nodes increases in the system, the ‘fraction
1) Analysis of time without a leader’ increases. In addition, when
The algorithm is based on diffusion computation and considering nodes’ mobility, the ‘mean election rate’
extrema finding, and it handles topological changes increases with the increase in their mobility.
dynamically. When election is triggered in a node, the node 2) Enhancement
starts a diffusing computation to find out its new leader. a) Algorithm assumes when a node wakes up from a crash, it
Several nodes can start diffusing computations in reply to should start an election to identify its new leader. This is
the exit of a leader, and hence, several diffusing computations costly. In the proposed enhancement, waking nodes send an
can be in progress at the same time; however, a node inquiry message to their functioning neighbors and identitfy
participates in only one diffusing computation at a time. the current leader instead of starting a new election.
Diffusion computation will search for a node with highest b) The algorithm uses extrema finding using the nodes UIDs
value, which is determined by the node’s Unique Identity in which a node aborts its current diffusion and joins the
(UID). Fig 3, 4 shows a typical diffusion and shrinkage. The higher order UID. A better method is to resolve the comparison
algorithm works in an asynchronous environment with based on the diffusion starter’s quality as this resolves the
channels that are not FIFO. When any node detects the loss process more rapidly.
of its leader, the algorithm starts. The initiator sends an
‘election’ message that spreads to all its neighbors thus C. A Leader Election Algortihm within Candidates on Ad
spanning out as a tree growing. The other nodes, upon the Hoc Mobile Networks[23]
arrival of the ‘election’ message, forward it to 1) Analysis
The paper is an extension of [8] and [20]. The assump-
tions include nodes with UIDs, asynchronous FIFO bidirec-
tional channels, and sufficiently large node buffers. The al-
gorithm also deals with networks partitions and mergers that
can occur during the election process. The addition to the
original algorithms [8][20] is that it uses candidate lists. Can-
didate lists contain the UIDs of current leader and four other
main candidates. When nodes confirm that the leader’s heart-
beats have stopped, the nodes initiate election messages.
Multiple messages are handled by checking the message with
the highest UID attached to it. When this election message
reaches the leaf nodes in the diffusion computation-span-
ning tree, they send acknowledgement messages to their
parents. These acknowledgment messages contain the can-
didate lists. On the way back to the initiator node, the lists
are updated. Update is done to obtain the highest known
Figure 3. The diffusion of tree UID of the nodes. Finally, the initiator combines the candi
© 2011 ACEEE 21
DOI: 01.IJNS.02.04. 37


date lists and the highest UID is elected as the new leader. establish its leader, then if its role is a master node then it
increases the counter by 1 or takes no action if it is a pure
slave node.

2) Enhancements
a) In the algorithm, during election only the master can
communicate with the rest of the nodes. In the proposed
enhancement, we recommend slave nodes be allowed to
communicate during election. This reduces the election time.
b) If a node wishes to start an election process, it sends a
Figure 5. Participant in Multiple Piconets (PMP) message to its master or other active masters if in a PMP. If
The paper derives an efficient leader election algorithm that no master is currently active for the node, it broadcasts the
ensures that every node has a leader in every situation and need to its neighbors to start the election process, and nodes
saves energy. The focus for saving energy is to reduce the wishing to become leaders will make themselves candidates.
number of leader election processes. Since leader election This creates more chances for nodes to become leaders.
requires three phases of transmissions and receptions of
messages both using a great deal of energy. The paper shows VI. EVALUATION AND DISCUSSION
that the candidate-based algorithm saves considerable energy.
The summary of our analysis based on Time and Message
2) Enhancement
Complexity, Assumptions considered, Fault Tolerance, Timing
a) We recommend that election of leaders must use the
Model, Proposed Enhancements and the Justification for
highest remaining battery life and the lowest mobility range
Enhancements is presented in Table-I. All the algorithms
in the election process. Since nodes with low mobility have
consider fault tolerance, most MANET use asynchronous
least chances of merging and partitioning from their current
timing model.
clusters.
b) The algorithm uses the UID to handle the multiple diffusion A. Comparative Evaluation
computations to elect a new leader. We recommend using a) In [1], when crashed nodes wake up, they by default
CSMA/CD algorithm to handle such multiple diffusion start an election. Accordingly, the number of elections
computations since it requires less number of messages. conducted is directly proportional to the number of nodes
D. Cooperative Leader Election Algorithm for Master/ waking up from crash. In the proposed enhancement, nodes
Slave Mobile Ad Hoc Network[26] send inquiry messages to their neighbors to inquire about the
current leader. This alternative avoids full-scale elections. For
1) Analysis
example, if we consider 5% of nodes failed, we further assume
An election algorithm for master/slave networks is
that 25% of the failed nodes contain the leader. As shown in
proposed in [26]. The master is responsible for assigning
Fig 6, the number of elections conducted is considerably
and coordinating tasks and roles including scheduling mutual
less and the cost of election is minimum. In [23] ,when CSMA/
exclusive access. The slaves follow the master’s orders; the
CD is used before sending any message, the nodes wait for
algorithm assumptions are that by utilizing the algorithm
a certain timeout. When the ID is higher, the lower the timeout.
iteratively, the MANET will end up with one unique leader.
Fig. 7 shows an scenario where node-1 fails and the message
The system is synchronous, and its channels are reliable.
transfers. The dotted lines represent the case of using CSMA/
Nodes have unique IDs. In the election algorithm, each node
CD and the number of messages are few.
maintains a tuple of two values, the ID of the current leader,
and the current leader’s master value. Nodes that wish to be
the new master, replace their tuple with their own values, ID
and criterion value. The master exchanges its tuple with each
of its slaves. If it finds a node with higher criterion value, it
will replace its own and broadcast it to the entire network of
its slaves. In the network, there can be nodes that are
Participant in Multiple Piconets (PMP) - a piconet is a network
with at least two nodes and, at most, one master. Fig.5 shows
a PMP layout. These nodes participate in multiple networks,
but during each time slot, they obey only one master. When
a new node joins the network and if its value is lower than the
current leader of the network it just accepts the current leader.
However, if the new node’s value is greater than that of the
current leader, the new node is the leader and the leader
information is broadcast to the entire network. In case when
the new node joins the network that has no leader or is yet to
© 2011 ACEEE 22
DOI: 01.IJNS.02.04.37


TABLE I.
ANALYSIS SUMMARY

* ANoN is the Average Number of Neighbors in the participating nodes in the elections. PN is the Participating Nodes number.
** PN is the Participating Nodes number during the elections. SY-Synchronous AS -Asynchronous
*** P is the number of PMP nodes in the network G, P(G) is the set of PMP nodes in network G, and d(i) is degree of the PMP node i.

3) Comparisons on[1] and [22] shows, the former uses diffu- [26], is that it is a ynchronous algorithm while the rest are
sion computation. This is better since the system can handle asynchronous.
floods of messages. Moreover, when the system has more
mobility, that leads nodes to leave the regions of connection B. Evaluation based on Factors
and the locality of the leader. It would be better to choose [1] 1) Complexities
over [22] since the former allows multiple diffusions that may The time and message complexities of all the algorithms
compensate the nodes’ departures. presented in Table-I show they are linear. All algorithms are
efficient; they only differ in the number of participating nodes or
delay.
2) Fault Tolerance
Most algorithms handle partitions and merges. The use of
FIFO channels is also present. In [23], candidates are used to
avoid consecutive elections. We recommend using candidates
in algorithms, since this reduces the overhead of starting a new
election session.
3) Assumptions
Assumptions limit the algorithm from its wide applications,
Figure 7. CSMA/CD in Leader election
hence minimum the assumptions lead to the greater adaption of
4) Comparisons on [25] and [26] show that both use extrema the algorithm in real applications.
finding, yet the latter uses a valued node property in order to 4) Timing Models
evaluate candidate nodes for leadership. Also, in the latter Most MANET use asynchronous timing model. Nodes have
the master can only exchange messages with the slave nodes no global time coordination among them.
while the former allows nodes to flood messages forming
trees rooted at the initiator node. Finally, a major difference in
© 2011 ACEEE 23
DOI: 01.IJNS.02.04. 37


VII. CONCLUSIONAND FUTURE WORK [11] M. Lima, A. dos Santos, and G. Pujolle, “A Survey of
Survivability in Mobile Ad Hoc Networks,” IEEE Communications
We presented an analysis on leader election algorithm in Surveys & Tutorials, 11 (1): 66-75, First Quarter 2009.
MANET. Factors considered in our analysis are time com- [12] B. Kannhavong, H. Nakayama, Y. Nemoto, and N. Kato,
plexity, message complexity, assumptions considered, fault “Application of Genetic Algorithms for QoS Routing in Mobile Ad
tolerance and timing model. The analysis results showed all Hoc Networks: A Survey,” in the Proceeding of 2010 International
algorithms analyzed considered fault tolerance in-depth,. Our Conference on Broadband, Wireless Computing, Communication
proposed enhancements include on recovery, nodes inquire and Applications BWCCA, pp. 250-256, November 2010.
[13] G. Le Lann. “Distributed systems: Towards a formal approach”,
about the current leader instead of initiating a new election
Information Processing 77, Proc. of the IFIP Congress, pp. 155-
process. In addition, we propose utilizing candidates during 160, 1977.
election to reduce the overhead, and use better election crite- [14] Humayun Bakht, “History of Mobile Ad Hoc Networks”,
ria specific to MANET such as battery life and mobility, using SCMS, Liverpool John Moores University, available:
an asynchronous timing model is a better choice as most ww.oocities.org/humayunbakht/ HMANET.pdf
MANET are asynchronous. Moreover, we propose to use the [15] Advanced Network Technologies Division, “Wireless Ad Hoc
CSMA/CD algorithm to handle multiple diffusion computa- Networks” National Institiue of Standards and Technology NIST,
tions since it requires fewer number of messages. As a part of USA, available: www.antd.nist.gov
future work, we intend to include more recent leader election [16] Ali Ghodsi, “Distributed Algorithms 2g1513,” available:
www.sics.se/~ali/teaching/dalg/l06.ppt
algorithms, additional factors for our further analysis, and
[17] Mitsoa Valia, “Distributed Leader Election Algorithms in
empirical studies on proposed enhancements in the paper. Synchronous Neworks,” available: www.web.cs.gc.cuny.edu/
~vmitsou/presentation.pdf
REFERENCES [18] Sukumar Ghosh, Distributed Systems: An Algorithmic
[1] S. Vasudevan, J. Kurose, and D. Towsley, “Design and Analysis Approach, Chapman & Hall/CRC, Tylor & Francis Group, LLC
of a Leader Election Algorithm for Mobile Ad Hoc Networks,” ,2007.
ICNP’04, 2004. [19] M. Castro and B. Liskov, “Practical Byzantine Fault Tolerance
[2] V. Park, and S. Corson, “A Highly Adaptive DistributedRouting and Proactive Recovery, ACM Transactions on Computer
Algorithm for Mobile Wireless Networks,”Proceedings to Systems,” v. 20 n. 4, pp. 398-461, 2002.
INFOCOM 97, pp. 1405–1413, 1997. [20] Leslie Lamport, “Time, Clocks, and the Ordering of Events in
[3] N. Mohammed, H. Otrok, W. Lingyu, M. Debbabi, and P. a Distributed System,” Communications of ACM, vol. 21, no. 7,
Bhattacharya, “Mechanism Design-Based Secure Leader Election July 1978.
Model for Intrusion Detection in MANET,” IEEE Transactions on [21] S. Stoller, “Leader Election in Asynchronous Distributed
Dependable and Secure Computing, vol.8, no.1, pp.89-103, 2011 Systems”, IEEE Transactions on Computers, vol. 49, no. 3, March
[4]Allen Clement, Reaserch Statement, available: 2000
www.cs.utexas.edu/users/aclement/Allen_Research.pdf [22] Gang Zhang, Jing Chen, Yu Zhang, and Chungui Liu, “Research
[5] K. Hatzis, G. Pentaris, P. Spirakis, V. Tampakas and R. Tan. of Asynchronous Leader Election Algorithm on Hierarchy Ad Hoc
Fundamental Control Algorithms in Mobile Networks. In Proc. of Network,” WiCom ’09. 2009.
11th ACM SPAA, pages 251-260, March 1999. [23] S. Lee, M. Rahman, and C. Kim, “A Leader Election Algorithm
[6] G. Singh, “Leader Election in the Presence of Link Failures,” Within Candidates on Ad Hoc Mobile Networks,” Embedded
IEEE Transactions on Parallel and Distributed Systems, vol. 7, no. Software and Systems, Lecture Notes in Computer Science, Vol.
3, March 1996. 4523, pp: 728-738, Springer Berlin / Heidelberg 2007.
[7] Amin Ansari, “Verification of Peterson’s Algorithm for Leader [24] M. Pease, R. Shostak, and L.Lamport, “Reaching Agreement
Election in a Unidirectional Asynchronous Ring Using NuSMV,” in the Presence of Faults,” Journal of the Association for Computing
Cornell University Library, August, 2008, available: Machinery, Vol 27, No 2, April 1980, pp 228-234
arXiv:0808.0962v1 [25] N. Malpani, J. L. Welch, and N. Vaidya, “Leader Election
[8] G. H. Molina, “Elections in a Distributed Computing System.” Algorithms for Mobile Ad Hoc Networks,” In the Proceedings of
IEEE Trans. Comp, vol.31, no. 1, pp.48-59, 1982. DIALM ’00, NY, USA, 2000, ACM
[9] H. Abu-Amara, and A. Kanevsky, “On the Complexities of [26] R. Ali, S. Lor, R. Benouaer, M. Rio, “Cooperative Leader
Leader Election Algorithms,” Proceedings Fifth International Election Algorithm for Master/Slave Mobile Ad Hoc Network,”
Conference onComputing and Information ICCI, Page(s): 202, 29 2nd IFIP Wireless Days (WD), Paris, 15-17 December 2009, pp.
May,1993 1-5.
[10] A. Mian, R. Baldoni, and R. Beraldi, “A Survey of Service [27] M. Rahman, M. Abdullah-Al-Wadud, and O. Chae. Performance
Discovery Protocols in Multihop Mobile Ad Hoc Networks,” IEEE analysis of leader election algorithms inmobile ad hoc networks.
Pervasive Computing, 8 (1): 66-74, January-March 2009. Int’l J. of Computer Scienceand Network Security, 8(2):257–263,
2008.
[28] SungSoo Lee, Rahman M. Muhammad, and ChongGun Kim, A
Leader Election Algorithm Within Candidates on Ad Hoc Mobile
Networks, ICESS 2007,Springer Berlin / Heidelberg, 2007

© 2011 ACEEE 24
DOI: 01.IJNS.02.04.37

Analysis and Enhancements of Leader Elections algorithms in Mobile Ad Hoc Networks

More Related Content

What's hot (17)

Viewers also liked (8)

Similar to Analysis and Enhancements of Leader Elections algorithms in Mobile Ad Hoc Networks (20)

More from IDES Editor (20)

Recently uploaded (20)

Analysis and Enhancements of Leader Elections algorithms in Mobile Ad Hoc Networks