SEARCHING FOR RESOURCES IN MANETS:
A cluster based flooding approach
Rodolfo Oliveira, Luis Bernardo, Paulo Pinto
Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, P-2829-516 Caparica, Portugal
Keywords: Performance analysis of wireless ad hoc networks, searching service, clustering protocol.
Abstract: In this paper, we propose a searching service optimized for highly dynamic mobile ad-hoc networks based
on a flooding approach. MANETs unreliability and routing costs prevent the use of central servers or global
infra-structured services on top of a priori defined virtual overlay networks. A flooding approach over a
virtual overlay network created on-demand performs better. Flooding is supported by a light-weight
clustering algorithm. The paper compares the relative efficiency of two clustering approaches using 1.5-hop
and 2.5-hop neighborhood information, and of a non-clustered approach. It presents a set of simulation
results on the clustering efficiency and on searching efficiency for low and high mobility patterns, showing
that the 1.5-hop algorithm is more resilient to load and to node movement than the 2.5-hop algorithm.
1 INTRODUCTION
The problem of looking for resources on 802.11
Mobile Ad hoc NETworks (MANETs) is complex
due to the networks unstable nature. Nodes move
around independently creating a very dynamic
network topology. It is assumed that no geographic
position information is available, which is most of
the time true, mainly in indoor scenarios. MANET
routing protocols can be seen as resource lookup
service that look for IP addresses.
Experience with fast moving nodes (Tsumochi,
03) showed that standard proactive, table-driven,
routing protocols perform worst than on-demand
routing protocols, which flood the network looking
for an address only when it is needed. It also shows
that both approaches fail to handle extreme mobility
conditions. The problem is that routing information
becomes outdated too fast, especially for lengthy
paths. Due to bandwidth restrictions, it is not
feasible to maintain proactively the tables always
updated. On-demand approaches fail due to packet
collisions and due to the breaking of the return path
in result of intermediate node movement. These
conclusions are extensible to generic searching
services implemented at application layers.
Structured peer-to-peer p2p (services) and directory
services have much higher update costs than
flooding based services (Bernardo, 04). A flooding
approach is more adapted to unstable MANETs due
to the null registration costs. All efforts are
concentrated during the search phase.
The searching protocol performance depends
strongly on the lower layers of the protocol stack,
responsible for routing IP packets, and for handling
the Medium Access Control (MAC). Traditional
flooding peer-to-peer (p2p) services create virtual
overlay networks. They are formed by several nodes
connected using static TCP links. Their performance
drops sharply on a MANET if the virtual overlay
topology is not similar to the network physical
topology, due to the routing protocol overhead.
Crossing a virtual link may lead to a route recovery
procedure (usually a network flood) if the MANET
topology changes.
The MANET routing protocol overhead can be
avoided if the searching protocol's query message is
broadcasted, hop by hop, during the searching flood
(e.g. ad hoc mode of the JXTA rendezvous protocol
(JXTA, 04)). However, two problems may occur:
the 802.11 MAC layer is more error prone for
multicast/broadcast packets than for unicast packets,
and dense networks may suffer from the broadcast
storm problem (Tseng, 02). This latter problem can
be minimized organizing nodes into clusters, and
reducing the number of nodes sending messages to
the network.
This paper presents a new searching algorithm,
optimized for very dynamic continuous MANETs. It
focuses mainly on the clustering algorithm. Section
two overviews existing cluster based searching
algorithms. The proposed clustering and searching
104
Oliveira R., Bernardo L. and Pinto P. (2005).
SEARCHING FOR RESOURCES IN MANETS - A cluster based flooding approach.
In Proceedings of the Second International Conference on e-Business and Telecommunication Networks, pages 105-111
DOI: 10.5220/0001417001050111
Copyright
c
SciTePress
protocols are presented on sections three and four.
Section five presents the ns-2 simulation setup, and
several performance measurements. Finally, section
six draws some conclusions and presents future
work directions.
2 CLUSTER BASED SEARCHING
Cluster based searching approaches group nodes into
broadcast groups (BGs), and using a set of heuristics
select BG leaders (BGLs), responsible for
forwarding packets for their BG. Nodes periodically
broadcast a beacon packet, that may carry (Wu, 03):
local information (1-hop); its BGL (1.5-hop); a
neighbor node (within radio range) list (2-hop); the
neighbor's list with the BGLs information (2.5-hop);
etc. Most MANET protocols adopt a 2-hop or above
approach (e.g. OLSR (Jacquet, 01)). 2-hop is the
minimum information required to define a set of
active nodes that cover all nodes, usually called a
Connected Dominant Set (CDS). Although, on an
unstable MANET, it is possible that 2-hop and
further distant neighbors information is out-of-date,
introducing errors on the CDS construction that
result in failure to cover all nodes. Additional errors
may result from the impossibility of revoking
explicit state configurations created using signaling
(e.g. OLSR BGL election).
Another approach is to adopt 1-hop strategies
(e.g. SBA (Peng, 00)), just for maintaining the list of
neighbors, and to use an external searching protocol
for restraining the number of active nodes. In SBA,
nodes delay the sending of query messages for a
random time waiting for its possible transmission by
other neighbors. Nodes include their neighbor list
(Nq) on the query message before sending it.
Receivers store the union of Nq lists received (Nu)
and compare it with their list of neighbors (Nr),
canceling the transmission when Nu is equal to Nr.
ABC-QS (Choi, 02) modifies the forwarding rule
proposed by SBA reducing searching delay: nodes
do not delay the query message sending if the
number of neighbors in Nr and not in Nq is above
the number of neighbors common to Nr and Nq.
Otherwise, they delay an average time proportional
to the number of nodes in Nr and not in Nq. ABC-
QS may fail for dense MANETs where overlapped
nodes may send packets without waiting.
MANETs are not homogeneous. Some nodes
stay together during a large period of time (e.g.
students on a bus tour) while others move
independently. Beacons can also be used to detect
relative stability relationships. Toh introduced the
concept in ABR (Toh, 97), measuring the number of
beacons received. ABC-QS extended the metric to
cope with asynchronous piggybacked beacons.
Other authors introduced link stability measurements
based on packet probability failure (McDonald, 99).
Nevertheless, most clustering approaches do not take
link stability into consideration (OLSR, etc.),
producing unstable clusters for unstable MANETs.
ABC-QS and (McDonald, 99) create proactive
routing information within islands of stable
connected nodes, to speed up searches. However,
they ignore the stability information for thorough
flooding network searches that cover several stable
islands. This paper proposes a new solution, which
improves flooding using stability information.
3 CLUSTERING ALGORITHM
The proposed clustering algorithm groups "stable"
nodes into 1-hop radius clusters. Each node selects a
BGL periodically using a local soft-state protocol.
The resulting network simplified view is used to
reduce the flooding search overhead.
Each node periodically broadcasts a beacon
message. All nodes that receive a beacon from a
node n
y
are defined as n
y
neighbor nodes. Nodes
keep a table of neighbors' link stability η (called the
beacon table). Following ABR, link stability for n
y
is
defined as the sum of consecutive beacons received
from n
y
. If more than one beacon is lost, then link
stability is set to null. The stability measurement
trades-off a faster link failure detection (compared to
packet loss rate measurements) for a higher
probability of false link loss detection due to two
successive beacon collisions. High stability values
represent low nodes relative mobility and vice-versa.
In each beacon message, a node sends its node
identification, its BGL node address, and the higher
link stability value contained in its beacon table,
which is represented by µ. The beacon table includes
the neighbor's address, their link stability (η); their
BGL address; and the µ value received in the last
beacon. Every beacon table entry is automatically
destroyed if a beacon is not received during two
beaconing time periods. Table 1 is a hypothetical
beacon table of node 3 illustrated in figure 1. Node 3
received 43 beacons from neighbor node 1.
A node is stable if there is at least one η value in
its beacon table that is higher than a defined
stability_threshold. BGL selection algorithm is
run on each node before sending a beacon. The
selection algorithm for node n
a
is summarized in
figure 2.
SEARCHING FOR RESOURCES IN MANETS - A cluster based flooding approach
105
Table 1: Beacon table of node 3 on figure 1
Neighb. Stability (η) BGL Neighb. Stability (µ)
1 43 1 43
2 8 6 64
4 2 5 33
Figure 1: Illustration of a MANET with 3 BGs. Nodes
1, 5 and 6 are BGLs.
1. (η
max
)=find_maximum_η_value_in_table()
2. last_addr = MAX_INT
3. pre_selected = -1
4. if is_stable(n
a
) // stable node
5. //insert all known BGL’s stable neighbor
//nodes in BGL_list
6. for each neighborhood_node n
x
7. insert_in_sort_list(BGL(n
x
),BGL_list)
8. if is_BGL(n
a
) // if this node is BGL
9. insert_in_sort_list(n
a
,BGL_list)
10. // Choose BGL based on stability and
// lowest address criteria
11. for each bgl
x
contained in BGL_list
12. for each neighborhood_node n
x
13. if ((n
x
=bgl
x
)and(is_stable(n
x
)))
14. pre_selected = n
x
15. if (pre_selected -1) break;
16. if (n
a
=bgl
x
) // self-selection
17. pre_selected = n
a
18. break
19. //select new BGL
20. if (pre_selected=-1)//BGL is not selected
21. for each neighborhood_node n
x
22. if (η
max
-η(n
x
)-transient_threshold 0)
(addr(n
x
)<last_addr)
23. last_addr = addr(n
x
)
24. pre_selected = n
x
25. BGL_SELECTED = pre_selected
Figure 2: BGL node selection algorithm applied in node
n
a
.
A stable node first computes a sort list of all
available neighbor's BGL (lines 5 to 7), that includes
the node in case of being BGL (lines 8 to 9). This
list is sorted from the smallest to the largest BGL
address. If there are BGLs selected in the
neighborhood, the node chooses the BGL that has
the lowest address (lines 11 to 15), which can be the
node itself if it was chosen as BGL by a neighbor
(lines 16 to 18). If there are no BGLs selected in its
neighborhood, a node simply selects as BGL its
neighbor with the highest η value (lines 20 to 24). If
there is more than one neighbor owning the
maximum η value then it is selected the node with
lowest address.
During system startup, transitory cluster overlap
may appear, because the initial criteria for selecting
BGL is a local measurement for link stability (lines
21 to 23), which may differ from node to node. The
transient_threshold was set to one, to compensate
different beacon delivery time drifts (jitter). The
initial BGL is the neighbor with the lowest address
that could get the maximum stability value during
the present beaconing period. Yet, when several
BGLs exist within radio range connected by stable
links, they are merged into a single cluster (lines 10
to 18) after one beacon period. Two nodes from
overlapped clusters sort neighbor's BGLs
independently into the same order (only node
address is considered) and converge to the same
BGL.
Cluster overlapping also occurs for continuous
groups of stable nodes wider than one hop. The
algorithm leads to the construction of multiple tree
structures of BGLs, called cluster-trees, centered on
BGLs with local minimum addresses. Each branch
has a sequence of BGLs (n
i
) with increasing
addresses, whose BGL is the branch predecessor
(BGL(n
i
) = n
i-1
). The exception is the periphery of
the cluster-tree, where nodes with lower addresses
can exist. Due to line 16, a node can only self-select
as BGL if another node previously selects him. This
avoids the existence of single node clusters on the
periphery of a cluster-tree.
Within a connected stable group (a group of
cluster-trees connected by stable links), the border
between cluster-trees' BGLs is composed by one or
two non-BGL nodes. It cannot be zero because lines
12-15 would merge the BGLs. Also, it cannot be
more than two because that would mean that a node
would not have a BGL in the neighborhood, and
lines 20-24 would select a new BGL.
Figure 1 presents a cluster-tree with a root BGL
(node 1) and two branch BGLs (nodes 5 and 6).
Node's 6 BGL is node 1, but node 6 is also a BGL
selected by node 2. Node 2 will only form an
independent cluster-tree if a new node creates a
stable link and selects him as BGL.
The clustering algorithms' performance depends
on the network stability. If a large percentage of the
nodes are stable, the algorithm is able to detect them,
and reduce their load by grouping them in clusters.
If all nodes are unstable, beaconing only introduces
overhead. A lower beacon period value tolerates
higher nodes velocity. However, it increases the
bandwidth overhead and the network collisions. It is
better to reduce the clustering overhead and increase
6
2
3
4
5
1
3
ICETE 2005 - WIRELESS COMMUNICATION SYSTEMS AND NETWORKS
106
the flooding algorithm redundancy, to tolerate
clustering inaccuracies. If conventional criteria were
used, the clustering algorithm would create highly
unstable clusters, which would include passing-by
moving nodes, and would route query packets based
on this error prone information.
4 SEARCHING ALGORITHMS
The searching algorithms were developed as an
evolution of the basic source routing flooding
algorithm (SR). In SR the lookup operation is started
with a query message originated by a source node,
which carries a unique identification (Q
id
), the
source node address (n
source
), a resource
identification pattern to locate (R
id
), and the path (P).
This message is successively resent by each node, as
long as it has not been received before and the hop
limit is not reached. Each sender appends its
identification to P. Nodes maintain a local table
indexed by source node id, with last query' ids
received. A hit message is sent to the source node
when any local information satisfies the query. Hits
are routed to the query's node source using the path
included in the query message.
This paper proposes 1.5-hop and 2.5-hop
algorithms that enhance SR flooding phase, reducing
its overhead, and the hit message routing, improving
its resilience to node movement and failure. SR is
modified in three ways:
(a) The number of active nodes is reduced using
the clustering node information. A node can be: a
BGL if it receives a beacon selecting it; a non-BGL
if it selects a BGL but is not selected a BGL; or
isolated if it does not select a BGL. An unstable
node with one or more stable nodes in its
neighborhood selects for BGL the node with the
highest µ value, strictly for flooding purposes. Two
approaches are presented above;
(b) Query message size is reduced by removing
all non-BGLs and isolated nodes' ids before the last
BGL from the path field (P). The partial path is
stored and pruned, each time the message passes on
a BGL. In case of node failure, the node can always
use the BGL list (stored in the query message) to
recover the route to the source node;
(c) When hit messages follow the query reverse
path, unicast is used and their sending is confirmed.
When a link fails, the node looks at its neighbor list,
and neighbor's BGL list, looking for any node on the
reverse path. As a last resort, when no information is
available, the node that detects the failure starts a hit
message flooding. The hit message is treated as a
special query packet, looking for a node id within
the remaining query path list, which does not receive
any reply. Hit flooding stops when the message
reaches a node whose neighbor's (or the node itself)
are part of the remaining path. Therefore, contrary to
SR, the proposed algorithm is able to survive to
extreme mobility, and is able to route hit messages
over failed or moving nodes.
A. 1.5-hop searching algorithm
BGL and isolated nodes always broadcast
queries one time (though isolated delay message
transmission). A non-BGL delays the query sending
for a fixed delay plus a jitter interval, and lists the
visited BGL on a local variable. While the timer is
active, the node continues to receive replicas of the
query message resent by neighbors. It just extracts
the query path list (P), and updates the visited BGL
list with the node's address and the nodes's BGL
address. When the timer goes off, the node checks to
see if all its neighbors' BGLs and his own BGL are
already listed. If they are not, then it resends the
message to cover the missing BGLs. Otherwise, it
drops the message.
Since BGLs do not delay the message and
isolated nodes do, search path goes preferentially
over BGL nodes. For cluster-tree borders defined by
non-BGLs, the timer's jitter limits the number of
retransmissions that occur on dense networks. The
faster non-BGL on an area transmits the query to the
destination BGL (or non-BGL for BGLs separated
by two non-BGLs), which retransmits it. The BGL is
added to the visited BGL list of other non-BGLs on
the same area suspending their transmissions.
The algorithm improves SBA (Peng, 00) and
ABC-QS (Choi, 02): It reduces the searching delay
while crossing a connected set of stable nodes
because BGLs never delay a query message; it
reduces the message size (the number of BGL is
lower than the number of nodes); it bases search
paths preferentially over stable nodes, less likely to
disappear; and it degrades more gracefully in the
presence of transmission errors. It handles
transmission errors similarly to SBA and ABC-QS:
nodes keep sending a query message as long as a
BGL does not appear on the path. Therefore, it only
fails to reduce the load if none of the neighbor
members of a BGL cluster retransmit the query
message. This behavior improves the algorithm
effectiveness for high network loads (due to the
higher collision rate) and for high mobility
conditions.
The algorithm does not guarantee total coverage
on unstable networks, because it does not take into
account unstable nodes in the neighborhood that did
not yet transmit a beacon.
B. 2.5-hop searching algorithm
SEARCHING FOR RESOURCES IN MANETS - A cluster based flooding approach
107
A second searching algorithm was developed as
an extension of the 2.5-hop algorithm proposed in
(Wu, 03).
A clustering algorithm modification is needed to
support 2.5-hop searching algorithms: the neighbor's
BGL list is added to the beacon message. The
original 2.5-hop clustering algorithms (Wu, 03) sent
the entire list of neighbors on the beacon producing
more overhead.
On this algorithm, a node has information about
all BGLs and isolated nodes within 2-hop distance.
In order to reduce bandwidth usage, each sending
node puts in the query message the list of non-BGL
nodes at 1-hop distance (v) that must resend the
message. The message is sent by the query starting
node; by each BGL visited and isolated nodes; and
by the non-BGL nodes that are in list v. List v is
constructed from the set of 1 hop neighbors, and
includes the non-BGLs required to cover all 2-hop
distance BGLs. The algorithm: 1) first adds the
neighbor nodes with unique paths to a BGL; 2) then,
adds the neighbors that cover the maximum number
of BGLs not yet in the list. A minimum node
identification criterion was used to select from nodes
with similar number of BGLs accessible.
The algorithm is more sensible to errors in the
clustering information than the 1.5-hop version,
since it uses topology information received one
beacon period ago to select on-demand the next hop
for the query message flooding. It also has less
redundancy to tolerate transmission and topology
errors, because it floods queries on a minimum CDS.
5 SIMULATIONS
The proposed algorithms and the source routing
algorithm were implemented on version 2.27 of ns-2
platform (ns-2). The presented simulations compare
the algorithms performance, using the same query
generation and node movement patterns. In each
simulation scenario 200 nodes are moving during
1000 seconds on a 1000m x 1000m area according
to the Generalized Random Waypoint mobility
model. Five different mobility scenarios were
defined to study the mobility behavior of each
flooding technique. Node’s average speeds of 0m/s,
1 m/s, 10m/s, 30 m/s and 40m/s were obtained using
constant pause times of 1000, 150, 10, 9 and 5
seconds, respectively. Each node has approximately
100 meters of communication range using IEEE
802.11b over the two-ray ground propagation model.
The beaconing frequency of each node is 1 Hz. The
clustering algorithm parameters
transient_threshold and stability_threshold
are one and five seconds, respectively.
Ten thousands of different resources are
randomly distributed on the network nodes. Three
different behavior patterns were defined using the
model presented in (Ge, 03). High, medium and low
network load correspond, to 10927, 1125 and 267
generated queries, respectively. Finally, all
broadcasts are sent with a jitter value of 100 ms, and
the 1.5-hop algorithm uses a delay of 700 ms for
non-BGLs and isolated nodes.
Figure 3 presents experimental results for the
average BGL selection time for the fifteen
combinations of speed and load, and for 1.5-hops
and 2.5-hops neighborhood information. The
selection time values not shown on the graph, for
low and medium load and mobility zero were
respectively 195 and 78 seconds for 1.5-hop and 118
and 61 seconds for 2.5-hop algorithm. These results
show that the BGL selection time is negatively
influenced by the beacon size and the load, but the
average speed is the dominant parameter.
For zero mobility, all BGL changes resulted from
having two successive beacon losses, producing a
significant churn on the BG composition for heavy
loads. Beacons are sent using multicast, and these
results show how sensible multicast traffic is to
collisions. The clustering algorithm stability could
be improved for low mobility scenarios by tolerating
more beacons losses. However, the algorithm's
performance would degrade significantly for high
average speed values. Notice that the algorithm also
degrades for the 2.5-hops algorithm, due to the
largest beacon length.
Node movement introduces extra BGL re-
selections due to topology changes, which become
the dominant factor for the two highest speeds. For
node average speeds of 30 and 40 m/s the BGL
persistent time converges for the minimum possible
value (5 minus the selection tolerance of one). The
percentage of nodes without a BGL also increased
significantly, which means that on these scenarios
the clustering is almost turned off.
ICETE 2005 - WIRELESS COMMUNICATION SYSTEMS AND NETWORKS
108
0
5
10
15
20
25
30
0 1 10 30 40
Node's average speed [m/s]
BGL average election time [s]
2.5hop-low 2.5hop-med 2.5hop-high
1.5hop-low 1.5hop-med 1.5hop-high
Figure 3: Average BGL selection time versus node
average speed for 1.5-hop and 2.5-hop algorithm.
Figure 4 presents the percentage of successful
queries for two extreme mobility scenarios of 1 m/s
and 40 m/s, where 1.5-hop, 2.5-hop and source
routing algorithms are compared using the medium
load. It shows that the 1.5-hop searching algorithm
outperforms the other two algorithms on both
scenarios. It also shows that the pure source routing
algorithm performance is poor for both scenarios.
The main factor that penalizes source routing
algorithm is the dependence on a single reverse path
to route the hit packet. Source routing performance
for 1 m/s is conditioned by the higher number of
nodes disseminating query messages and the longest
query message (it carries the complete path), which
lead to more packet collisions, destroying query
messages and hit messages. For the highest speed,
the success probability drops to 2% is result of a
high probability of return path failure.
0
10
20
30
40
50
60
70
80
Src. Rt 1m/s 2.5-hop 1m/s 1.5-hop 1m/s Src. Rt 40m/s 2.5-hop 40m/s 1.5-hop 40m/s
0.0
1.0
2.0
3.0
4.0
5.0
6.0
time [s]
Successful Queries[%] Total Load [bytes/(node*query)]
Beacon Load [bytes/(node*query)] End-to-End delay [s]
Figure 4: % Successful queries, load and end-to-end delay
versus algorithm and node speed for medium load.
2.5-algorithm has a higher beacon overhead,
which penalizes the bandwidth load. It is also
sensible to packet loss during the query message
dissemination due to using a minimum CDS. A node
movement or a packet loss may produce a query
coverage shedding, reducing the successes rate for
higher speeds.
The 1.5-hop algorithm has the lowest load levels
and is more tolerant to network changes, presenting
a low degradation on the successful query rate. On
the other hand, it increases the end-to-end search
delay. Notice that due to the clustering reduced
efficiency for high mobility, the 1.5-hop algorithm
load increases, tending for SBA model for extreme
mobility scenarios. This characteristic limits the
network scale and to the network load supported by
the algorithm for very high node average speeds.
6 CONCLUSIONS
The results presented in this paper show that the
proposed 1.5-hop searching protocol has a strong
resilience to network load and node movement,
constituting a good choice for extreme mobility
scenarios with low load levels. Its adaptability
results from an adaptive clustering protocol, based
on link stability, which adapts the controls the
clustering granularity based on the network
conditions. It reduces the cluster size and duration
for extreme mobility scenarios increasing searching
redundancy; it reduces redundancy for low mobility
nodes, reducing the searching overhead.
The obtained results show that for high mobility
scenarios, performance improves for the algorithms
that use the least possible network information (1.5-
hop). It is concluded that source routing approach
fails for high mobility scenarios. Since most
MANET routing protocols are based on source
routing, this can present an important problem for
common applications, not prepared to handle this
kind of instability.
This paper presents on-going work. Further
study is being made on beacon overhead reduction
and beacon self-stabilization algorithms, which
reduce beacon collision effects.
REFERENCES
Bernardo, L., Pinto, P., 2004. A Scalable Location Service
with Fast Update Responses. In Proc. ICETE'04,
Vol.1, pp.39-47.
Choi, Y., and Park, D., 2002. Associativity based
clustering and query stride for on-demand routing
protocol. Journal of Communications and Networks
4(1), pp. 4-13, Korean Institute of Communications
Sciences (KICS).
SEARCHING FOR RESOURCES IN MANETS - A cluster based flooding approach
109
JXTA Project, 2004. JXTA v2.0 Protocols Specification.
Retrieved September 2004 from
http://spec.jxta.org/nonav/
v1.0/docbook/JXTAProtocols.html
Ge, Z., Figueiredo, D., Jaiswal, S., Kurose, J., Towsley,
B., 2003. Modeling Peer-Peer File Systems. In IEEE
INFOCOM'2003.
Jacquet, P., Mülethaler, P., Clausen, T., Laouitit, A.,
Qayyum, A., Viennot, L., 2001. Optimized Link State
Routing Protocol for Ad Hoc Networks. In IEEE
INMIC'01, pp. 63-68. IEEE Press.
Klingber, T., Manfredi, R., 2002. RFC Draft of Gnutella
v0.6. Retreived from http://rfc-
gnutella.sourceforge.net/src/rfc-0_6-draft.html.
McDonald. A. B., Znati, T.F., 1999. A Mobility-Based
Framework for Adaptive Clustering in Wireless Ad
Hoc Networks. IEEE Journal on Selected Areas in
Communications 17(8) pp. 1466-1487
Network simulator - ns-2. Retrieved from
http://www.isi.edu/nsnam/ns/
Peng, W., Lu, X., 2000. On the Reduction of Broadcast
Redundancy in Mobile Ad Hoc Networks. In
MobiHoc'00, pp. 129-130, ACM Press.
Toh, C.-K., 1997. Associativity-Based Routing for Ad-hoc
Mobile Networks. Journal of Wireless Personal
Communications, vol. 4, pp. 103-139, Kluwer
Academic Publishers.
Tseng, Y.-C., Ni, S.-Y. , Chen, Y.-S., Sheu, J.-P., 2002.
The Broadcast Storm Problem in a Mobile Ad Hoc
Network. In Wireless Networks, vol. 8, nos. 2/3, pp.
153-167, Mar.-May 2002. Kluwer Academic
Publishers B.V.
Tsumochi, J., Masayama, K., Uehara, H., Yokoyama, M.,
2003. Impact of Mobility Metric on Routing Protocols
for Mobile Ad Hoc Networks, In IEEE Pacific Rim
Conf. on Communications, Computers and Signal
Processing (PACRIM03), pp.322-325. IEEE Press.
Wu, J., and Lou, W., 2003. Forward-node-set-based
broadcast in clustered mobile ad hoc networks.
Wireless Communications and Mobile Computing,
N.3, pp.155-173, John Wiley & Sons, Ltd.
ICETE 2005 - WIRELESS COMMUNICATION SYSTEMS AND NETWORKS
110