AN EXTENDED ANALYSIS OF AN INTEREST-BASED SERVICE
DISTRIBUTION PROTOCOL FOR MOBILE AD HOC NETWORKS
Mohamed Hamdy and Birgitta K¨onig-Ries
Institute of Computer Science, Friedrich-Schiller University, Jena, Germany
Keywords:
MANET, SOA, replication.
Abstract:
In principle, service orientation is a suitable paradigm to allow for effective resource sharing in wireless ad
hoc networks. However, special attention needs to be paid to ensure a high service availability since this is
the basis for reliable service execution. Unfortunately, typical characteristics of ad hoc networks like ever-
changing topology and limited resources have a dramatic negative effect on service availability. Therefore,
special measures are needed to cope with this problem. In general, replication for services as well as data
represents an efficient solution when the availability of some resource needs to be ensured. In this paper,
we extend the service replication and distribution protocol presented in our prior work which is based on the
interest of clients and providers of a specified service by taking into account not just one but all partitions of
the network. By elaborating an extensive detailed simulation, the efficiency of replication and the allocation
correctness are being examined. The results show that relying on our protocol is feasible. Since our protocol
utilizes high level (application level) information about the available services in the ad hoc network only and
does not rely on lower level information like network or protocol specifics, it is applicable in a wide range of
settings.
1 INTRODUCTION
In many applications, wireless ad hoc networks rep-
resent the only feasible solution to achieve commu-
nication for each of the network participants. The
mobile hosts are working in a collaborative way to
achieve the core network functions such as routing.
In addition, collaboration is needed to provide indi-
vidual devices with information and/or functionality
they do not possess themselves. Ad hoc characteris-
tics like resource-restricted, wireless-enabled partici-
pants’ devices, limitations of battery lifetimes, and the
ever-changing network topology pose very tight con-
straints on all of the applications to be deployed on
this type of network. In order to enable mobile hosts
to share functionality as is needed to guarantee the
functioning of the network, service orientation can be
used: Provided functionality is offered as a service,
required functionality is searched for via a service re-
quest.
However, provider mobile hosts can become tem-
porarily or permanently unavailable at any point in
time for some or all of the clients, the availability of
the services offered by them cannot be guaranteed.
Unavailability can be caused by service providers
leaving the network, but also by the development of
network partitions. Thus, mechanisms are needed to
ensure service availability in such a setting.
The idea to replicate some resource in order to
increase its availability is one of the classical con-
cepts of computer science, applied in many applica-
tions and systems (e.g., DDBMS, RAIDs, DNS, ...).
In fact, in many cases replication is the only feasible
solution to ensure availability. Unfortunately, direct
mapping of replication concepts into ad hoc networks
is not possible as will be explained below.
The ad hoc network topology consists of a vary-
ing number of separate network partitions. Most of
the related work is based on the prediction of the net-
work partitioning behavior, host disjoins, etc. It as-
sumes that all services offeredare vital to the network,
and thus aims at trying to deploy a copy (replica) of
each of these vital services into developing new net-
work partitions, and then finding solutions to manage
the service and its replica concurrently in the differ-
ent network partitions (Derhab et al., 2005; Derhab
and Badache, 2005; Dustdar and Juszczyk, 2007). As
mentioned in (Hamdy and k¨onig Ries, 2008) this type
203
Hamdy M. and König-Ries B. (2008).
AN EXTENDED ANALYSIS OF AN INTEREST-BASED SERVICE DISTRIBUTION PROTOCOL FOR MOBILE AD HOC NETWORKS.
In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 203-210
DOI: 10.5220/0002024202030210
Copyright
c
SciTePress
of scheme couples the replication decision tightly to
a specified lower layer component like the routing
component as in (Derhab and Badache, 2005; Dust-
dar and Juszczyk, 2007). This is a tremendous dis-
advantage. Instead , our work in (Hamdy and k¨onig
Ries, 2008) proposes a protocol for service distribu-
tion for ad hoc networks based on the providers’ and
clients’ interest in a specified service. This protocol
avoids dealing with the lower layer component and
can thus be flexibly used in wide range of settings.
Also, it takes into account that not all deployed ser-
vices will be equally important to all participants at
all times. Therefore, we introduce a time-varying im-
portance degree, the vitality of a service for a client.
Moreover, how interested a provider is in hosting a
service should be proportional to the overall interest
in it by the client group of this server. This means
that a service that clients and providers are inter-
ested in should be replicated, until interest decreases
in which case it should be hibernated. Our proto-
col achieves this behavior by using two measurements
(client-service-interest and provider-service-careless)
and employing two opposite mechanisms, namely (a)
replication mechanism: allows the service to be repli-
cated to a specified client based on a certain client-
service-interest level of it and (b) hibernation mech-
anism: allows a provider to hibernate a service after
gaining some certain provider-service-careless or the
gross client’s interest of that service becomes low.
By maintaining an appropriate number of replicas
for the current service interest, the following main ad-
vantages are achieved:(a) increasing the availability,
(b) avoiding the time and computation intensive op-
eration of network partition prediction and detection
and resolving the couplingto the lowernetwork layers
, (c) introducing the service interest as a realistic mea-
surement to be used in the replication process besides
the hibernation mechanism which represents a realis-
tic behavior of service providers in many cases, (d)
introducing the ability of tuning the degree of repli-
cation by utilizing the information of the application
layer.
The work of (Hamdy and k¨onig Ries, 2008) esti-
mated the performance of the service distribution pro-
tocol on just one partition of the network. In this
paper we are extending the protocol to be used on
the whole set of ad hoc network partitions. The con-
cepts to be evaluated by the current work are: Is the
generated number of replicas based on the interest
enough to satisfy the whole number of formed parti-
tions? How is the efficiency of the replica placement
based on the service interest? In order to evaluate our
concepts, we developed our own network model and
its related performance measurement. A detailed sim-
ulation and further analysis of the results show a good
and promising performance.
The rest of this paper is organized as follows: In
Section 3, the network model that we use to evaluate
our concepts and the proposed protocol is presented.
The proposed protocol is presented in Section 4 with
its two mechanisms of replication and hibernation. In
Section 5, a primary and extended results of an elab-
orated simulation for our proposed protocol are ana-
lyzed. Finally, the related work and conclusions are
discussed in Sections 2, and 6 respectively.
2 RELATED WORK
J. B¨ose et al. introduce in (B¨ose et al., 2005) an adap-
tive pull protocol for data dissemination over ad hoc
networks. That work estimates the data freshness con-
sidering the data load, by comparing their proposed
optimistic protocol to others techniques like flood-
ing and combinations of the proposed protocol with
flooding feature, they can nearly save 13% of the net-
work load and achieve high freshness rates. S. Mous-
saoui et al proposed in (Moussaoui et al., 2006) a
method for data replication in ad hoc networks with
after building the required replicas of available data,
starts a recovery stage to overcome the effects of
the mobility and ever-changing topology. The work
is based on the frequency of accessing based on a
moving averages equation (like in (Fu and Cheung,
1994)). Also, Work of T. Hara in (Hara, 2003) is
considering the data accessibility frequency to intro-
duce data replicas in many approaches of replicating
a specified data item on the whole mobile hosts.
In (H¨ahner, 2007) J. H¨ahner introduces a survey
over many data consistency models, then introducse
full and partial replication algorithms taking into con-
sideration the data consistency based on ordering the
observer’s graphs.
In (Derhab et al., 2005; Derhab and Badache,
2005) A. Derhab et al., by estimating the link qual-
ity and employing a partition prediction mechanisms
based on TORA (Park and Corson, 1997) supplies
two mechanisms for pull based replication; (a) repli-
cation (pre-partition formation) and (b) merging (af-
ter two partitions merged) mechanisms. In (Dustdar
and Juszczyk, 2007; Juszczyk, 2005) S.Dustdar et
al. introduce algorithms that take care of replication
and synchronization of services in ad hoc mobile net-
works. Based on a global view of all network nodes,
a replication mechanism (component) by the origi-
nal service node moderates the replication process
per predicted partition, replicas in the new formed
partitions are supposed to be hosted by a powerful
WINSYS 2008 - International Conference on Wireless Information Networks and Systems
204
elected node. The used service model assumes pres-
ence of master nodes in order to keep services syn-
chronized. (Hauspie et al., 2001) M.Hauspie et al.
(2001), other research like (Wang and Li, 2002) goes
also on the same fashion and concepts of link evalu-
ation and availability of global view about the ad hoc
network, (Derhab et al., 2005) presented a compari-
son between these approaches.
3 NETWORK MODEL
The goal of this research is to evaluate the per-
formance of the proposed replication/hibernation
mechanism across all network partitions. We use the
network model described in (Hamdy and k¨onig Ries,
2008), which models the network at a certain time
as an undirected, unweighted graph G(N,E) where
N represents the set of uniquely identified nodes,
and E is the set of edges representing network con-
nections between nodes. G
x
(N
x
, E
x
) represents one
of the network partitions, where: G(N, E) =
G
1
(N
1
, E
1
)· ··
S
G
x
(N
x
, E
x
)· ··
S
G
k
(N
k
, E
k
),
N = N
1
···
S
N
x
···
S
N
k
, and E = E
1
···
S
E
x
···
S
E
k
.
Each of the mobile hosts can cover a fixed range with
radius R, a connection is established between two
nodes if the distance between them is less than or
equal to R. All mobile hosts are placed in a square
area. The other components of the network model,
namely mobility , service, and calling models are
described in the following paragraphs.
Mobility Model. We use the random waypoint
mobility model (Lin et al., 2004), in which each
mobile node picks a random constant speed uni-
formly between some preset interval (in our model
[1..12]m/s), then generates a random destination
location to visit after waiting for a pause time
uniformly selected (in our model) between [0..30]
minutes. A slight modification was introduced.
By introducing the ”mobility index”, which is a
percentage [0..100]%, we can change the mobility
status of the network. The higher the mobility index,
the higher speed value selection is allowed, and lower
pause times are generated, and vice versa.
Service Model. To simplify the analysis, the
network is maintaining just one service. This service
is placed on the first created node in the network.
Three assumptions are made in the service model;
(a) all mobile nodes can participate in the replication
mechanism; (b) the original service is replicable;
(c) all participants do not mind to cache the replicas
in case of service hibernation. Each replica is
described by a requirement index which quantifies
the requirements needed to run this service. These
values are generated as a normal distribution of
about 20% of a general requirement index. The
requirement index is a mimic of the reality; normally
and even if two providers provide the same service,
requirements by each of them to use its service (or get
a replica) will differ. Clients are supposed to find the
minimum requirement index from the neighboring
services to communicate with. This varying of the
offered requirement index is one of the responsible
components of distributing the interest of the clients
among the offered services/replicas in the network.
By the work in this paper, since we are relying on
a optimistic replication model, the synchronization
management of the service/replicas is not considered.
Another current ongoing work is addressing the
concurrent service synchronization and states.
Calling Model. Initially, all of the created nodes
seek for the initial (original) service provider node;
only those nodes with at least one feasible path to
the provider node are supposed to start evaluating
the service calling and be involved in the related
replication/hibernation processes. After a while,
service/replicas prevalence through the network is
supposed to cover as much as possible of the ad
hoc formed partitions. Variant calling rates are
maintained by each node; the calling rate is generated
between [0..4] calls per minute, the calling rate
is supposed to be constant during a calling period
of {5,10,15} minutes, and after a pause time of
{0,5,10,15} minutes, the node is supposed to select
another calling rate and so on. Calling rate, calling
period, and pause period are uniformly randomly
generated.
4 THE APPLIED PROTOCOL
The main players in the protocol components are
the two measurements of the client-service-interest
and the provider-service-carelessness.For simplicity,
currently, both of these measurements are based
just on the client calling frequency and the service
requirement index. For a client, it will be considered
to be ”a replica-interested client” if it achieves a
certain number of calls within a specified time
interval; then the replication mechanism should start.
On the other hand, a provider will be considered as
a ”service-careless provider” if it receives no more
than one call in a specified time interval; then the
hibernation mechanism should start. Our motivation
for this research was more to investigate the concepts
AN EXTENDED ANALYSIS OF AN INTEREST-BASED SERVICE DISTRIBUTION PROTOCOL FOR MOBILE AD
HOC NETWORKS
205
of replicating the services based on the interest and
watching the service prevalence than on establishing
a sophisticated computation of the interest itself.
Finding more expressive definitions of interest is,
however, part of our ongoing work. As in (Hamdy
and k¨onig Ries, 2008), the core component actions
of the replication and hibernation mechanisms are
described below.
Replication Mechanism. The core actions of the
replication mechanism are as follows:
Restore from Cache. If a client is interested
enough to host a replica, it should search first if
it had a replica before, if yes it restores it.
Find Least Requirement Service. If a client is in-
terested enough to host a replica, then, it should
discover the replica with the least requirement in-
dex.
Pass a Replica. If a client is interested enough
to host a replica, it receives a replica from its
provider.
Switching to the Local Service. When a replica is
received by a client node (new provider), then the
node switches its calling to the local replica.
Publish. Allows publishing the new ser-
vice/replica status.
Check the Correctness. Enforces the interested
client to check if it can achieve a certain correct-
ness of replica placement if it receives its own
replica.
Hibernation Mechanism. The core actions of the hi-
bernation mechanism are as follows:
Shutdown. Hibernates a local replica.
Publish. Allows publishing the new replica status.
Find another Service. Finds another replica of the
called service, if that service is not found.
5 SIMULATION AND
DISCUSSIONS
A detailed simulation for the extended application
of the proposed protocol of (Hamdy and k¨onig
Ries, 2008) has been elaborated. The results are
divided mainly into two groups, the first group comes
from applying just the replication mechanism (R
group), and the other one comes from applying both
replication and hibernation mechanisms(R-H group).
In our performance analysis, four main performance
measurements have been introduced:
Service Availability is the ratio between the time
during which at least one replica was available in
any of the network partitions to the total running
time of the network.
Success Ratiois the ratio between the number of
successful service calls to the overall number of
calls in the entire network.
Service Prevalenceis the ratio between the num-
ber of mobile hosts that hosted a replica to the
whole number of network participants.
Residence Timeis the average time that the replica
remained running (not hibernated) on some mo-
bile host.
5.1 Configurations
The mobile hosts are placed in a 500 meter
2
area. The
transmission range of each node is fixed to 120 me-
ters. The network operation time has been set to be
2 hours per sample run; results are obtained from the
average of 20 runs. The replication threshold is set to
be 4 calls per minute; the hibernation threshold is set
to be 1 call in 5 minutes.
The network size is varying from 10 to 140 nodes. In
case of varying the network size, the mobility index is
fixed to be 50% and the maximum allowed prevalence
is set to be 100%. In case of observing the effects of
varying the mobility index, the network size is set to
be 50 nodes, and the maximum allowed prevalence is
set to be 100%. Finally, in case of varying the maxi-
mum allowed service prevalence, the network size is
fixed to be 50 nodes, and the mobility index is 50%.
5.2 Basic Performance Analysis
In both of the proposed groups of experiments (R and
R-H) the network size, mobility index, and maximum
allowed prevalence are varying and the service avail-
ability, success ratio, and residence time are observed.
Service Availability. In the (R) group of experi-
ments, by definition, the service is always available
in some partition, it might, however, be inaccessible
by nodes from other partitions as depicted in Figure
1(a,b,c). For the (R-H) group of experiments, allow-
ing the hibernationprocess enables the providernodes
to evaluate their carelessness about their offered ser-
vice, this leads to shutting down replicas by some
providers. Figure 1(a) shows, that the higher network
sizes the higher service availability. Starting from a
moderate network size of 30 nodes, about 71% of ser-
vice availability is observed.
The effect of the mobility is presented in Figure 1(b):
the higher the mobility index, the higher service avail-
ability, the reason is that higher speeds and lower
pause times result in the service host nodes to traverse
WINSYS 2008 - International Conference on Wireless Information Networks and Systems
206
Figure 1: Service Availability.
more of the network partitions in a shorter time; this
traversing enables more pervasive replication in the
whole network. The proposed protocol shows achiev-
ing high service availability for the lower mobility in-
dices as well, about 80% as a minimum service avail-
ability for the minimum mobility index 10%.
In Figure 1(c), The smaller values of the maximum
allowed service prevalence lead to achieve higher
and steady service availability, these small values
of the maximum allowed prevalence concentrate the
overall client interest to a few number of provider
nodes on the whole network, this interest concen-
tration pushes those provider to keep their services
one and of course minimize their provider-service-
careless measurements. The higher values of maxi-
mum allowed prevalence increase the number of the
service providers and of course distribute the gross in-
terest of the client among them, this leads to varying
achieved service availability, Despite of that, the min-
imum achieved service availability lies above 75%.
Success Ratio. The success ratio is much important
than the service availability because it indicates the
service accessibility form all of the network partitions
Figure 2: Success Ratio.
over the operation time. In Figure 2(a), even applying
just the replication mechanism -in (R) group- does not
insure 100% success ratio for moderate and low net-
work sizes, it achieves 68% for a very low network
density (10 nodes). The difference between success
ratios of (R) and (R-H) groups of results is due to the
reduced service availability in the (R-H) group. Start-
ing from a network size of 30 nodes, the success ratio
is above 70%, and the average difference between the
two groups is less than 15%. The effect of varying the
mobility index in Figure 2(b) is on the same fashion
of the service availability, for the (R) group, the suc-
cess ratio is constant about 100% because the network
size is sufficient to achieve enough interest by the dif-
ferent partitions to maintain at least one replica inside
each of them for all values of the mobility index. For
the (R-H) group, because the higher speeds and lower
pause times enable wider service dissemination on
more network partitions, the success ratio increases
as the mobility index increases. Figure 2(c) shows
the effects of varying the maximum allowed preva-
lence ratio, both curves are dramatically increasing
by slight increments of the maximum allowed preva-
lence, starting from 6% allowed prevalence the (R)
AN EXTENDED ANALYSIS OF AN INTEREST-BASED SERVICE DISTRIBUTION PROTOCOL FOR MOBILE AD
HOC NETWORKS
207
Figure 3: Service Prevalence.
curve is increasing very slowly to close to 100%, on
the other hand, the (R-H) curve is hesitating about av-
erage value about 83%. The valuable notice here is
that there is no need to have high prevalence ratios to
achieve high success ratios.
Service Prevalence. Figures 2(a,b) show the effects
of both the varying network size and mobility index
on the service prevalence. In the (R) group, in Fig-
ure 2(a) we can easily deduce that the prevalence ra-
tio is to be about half (55%) of the number of par-
ticipants which represents a high number of simulta-
neously running replicas over all network partitions.
Also, varying the mobility index (Figure 2(b)) has the
same effect, it produces about (55%) service preva-
lence on the network. On the other hand, in the (R-H)
group, by applying the hibernation mechanism, a sig-
nificant reduction of the service prevalenceratio could
be achieved. The reduction value is about(48%), 2(a),
in case of varying the network size and about(44%),
2(b), in the case of varying the mobility index. The
notable result here, is the effect of the hibernation
mechanism in enhancing the many criteria related
to the number of the running replicas like minimize
the required effort of service/replicas synchronization
and of course link’s utilization.
Residence Time. Figures 4(a,b) are showing that
applying just the replication mechanism of the (R)
curves makes the average residence time of the ser-
vice by the hosting node seem to be constant, the rea-
son is that, the gross interest of the network partici-
pants is divided on the same set of providers, since
Figure 4: Service Residence Time.
that gross interest comes from the uniform distribu-
tion of the calling rates, it is always supposed to have
a constant average. On the other hand, in case of ap-
plying replication/hibernation mechanisms together,
by the (R-H) curves, the sets of the hosting nodes are
supposed to be increased, not all of that participants
can receive the same client interest portion, so many
providers trigger the hibernation mechanism and shut
down their service, this behavior makes the average
residence time decrease affected by both increasing of
the mobility index and the maximum allowed preva-
lence.
5.3 Extended Performance Analysis and
Discussion
In this extended analysis both of the replica-
tion/hibernation mechanisms are applied together.
With these experiments we measure the correctness
of replica placement, and the suitable degree of repli-
cation.
Replication Allocation. Obviously, if we can
compute the optimum number of ad hoc partitions or
clusters for a certain point in time, placing a replica
WINSYS 2008 - International Conference on Wireless Information Networks and Systems
208
inside each partition will be the minimum required
number of replicas to keep service/data available
for all participants. Our proposed measure of the
correctness of replica placement is linearly based on
the partition size and the number of already available
replicas in it.
Correctness Ratio CR. In figure 5, the correct-
ness ratio of the placement process is bounded
between 0% and 100%, if there is no replicas in the
partition the ratio should be 0%, else if there are
one or two replicas in the partition the ratio will be
100%, otherwise ,for simplicity, the ratio is linearly
inversely proportional to the number of replicas, the
ratio becomes zero at a number of replicas equals to
the partition size (Pz). Normally, at least one replica
per partition is an optimal case. Finding two replicas
in the same partition is very healthy from replication
point of view.The following equation describes the
value of the correctness ratio CR
t
(P
i
) in an ad hoc
formed partition P
i
at a certain moment of time t:
CR
t
(P
i
) =
0 no(replicas) = 0
1 no(replicas)in
1, 2
Pzno(replicas)
Pz2
no(replicas) > 2
where no(replicas): the available number of repli-
cas in the partition.
Figure 5: A Linear Ratio of the Allocation Correctness.
In Figure 6(a,b,c), both of the correctness ratio of
the replica placement process and the replication de-
gree are investigated against the network size, mobil-
ity index, and the maximum allowed service preva-
lence. Allocation correctness is directly proportional
to bnetwork size as in Figure 6(a). By our approach
of replica placement (based on the client interest), we
can achieve a moderate correctness ratio for moderate
network sizes (about 50 nodes). The higher mobility
increases the correctness ratio, starting from the value
of 50%, for 50 nodes, the correctness ratio is about
35%, as in Figure 6(b). Starting from maximum al-
lowed prevalence ratio equals to 12%, (Figure 6(c)),
the achieved allocation correctness is about 35%. So,
the allocation correctness is increased as the network
Figure 6: Allocation Correctness and Replication Degree.
size increases, the same holds for both higher mobil-
ity and maximum allowed prevalence ratio until cer-
tain values are reached, then it becomes steady vary-
ing about a certain average.
Degree of Replication. The meaning of the degree
of replication here is the average number of replicas
in each network partition per that partition size over
the network operation time. Generally, the optimum
number of replicas is depending on the network
connectivity probability. The global connectivity
probability is based on the network density and
could not be precisely computed (Gianuzzi, 2004).
Rather than creating probabilistic models for estimat-
ing the connectivity probability (Madsen et al., 2004).
Figures 6(a,b,c) show that the proposed protocol
achieves a very low degree of replication, this degree
seems to be varying about some average about 15%
against each of the varying network size, mobility in-
dex, and maximum allowed prevalence.
AN EXTENDED ANALYSIS OF AN INTEREST-BASED SERVICE DISTRIBUTION PROTOCOL FOR MOBILE AD
HOC NETWORKS
209
6 CONCLUSIONS
In this paper we have shown that the proposed pro-
tocol of mobile service distribution in (Hamdy and
k¨onig Ries, 2008) is applicable on a real ad hoc mo-
bile network simulation considering the formation of
several network partitions. Simulation showed that by
using interest measurements of a categorized group of
clients, a certain number of running replicas could be
generated which traversed through the network par-
titions achieving a high service availability and suc-
cess ratio, while showing a low service prevalence
on the network participants. Moreover, we propose
definitions for both correctness of replica placement
regarding the ad hoc formed network partitions and
the replication degree. The proposed protocol shows
promising results for both of these criteria.
REFERENCES
B¨ose, J.-H., Bregulla, F., Hahn, K., and Scholz, M. (2005).
Adaptive data dissemination in mobile ad-hoc net-
works. INFORMATIK 2005 - Informatik LIVE!, Band
2, Beitrge der 35. Jahrestagung der Gesellschaft fr In-
formatik e.V. (GI), pages 528–532.
Derhab, A. and Badache, N. (2005). A pull-based service
replication protocol in mobile ad hoc networks. Euro-
pean Transactions on Telecommunications, 18:1–11.
Derhab, A., Badache, N., and Bouabdallah, A. (2005). A
partition prediction algorithm for service replication
in mobile ad hoc networks. In Second Annual Con-
ference on Wireless On-demand Network Systems and
Services (WONS 2005 ), Swi.
Dustdar, S. and Juszczyk, L. (2007). Dynamic replication
and synchronization of web services for high avail-
ability in mobile ad-hoc networks. Service Oriented
Computing and Applications, 1:19–33.
Fu, A. W.-C. and Cheung, D. W.-L. (1994). A transaction
replication scheme for a replicated database with node
autonomy. In Bocca, J. B., Jarke, M., and Zaniolo, C.,
editors, VLDB’94, Proceedings of 20th International
Conference on Very Large Data Bases, September 12-
15, 1994, Santiago de Chile, Chile, pages 214–225.
Morgan Kaufmann.
Gianuzzi, V. (2004). Data replication effectiveness in mo-
bile ad-hoc networks. In PE-WASUN 04: Proceed-
ings of the 1st ACM international workshop on Perfor-
mance evaluation of wireless ad hoc, sensor, and ubiq-
uitous networks, pages 17–22, Venezia, Italy. ACM.
H¨ahner, J. (2007). Consistent Data Replication in Mobile
Ad Hoc Networks. PhD thesis, Fakult¨at Informatik,
Elektrotechnik und Informationstechnik der Univer-
sit¨at Stuttgart, Stutgart.
Hamdy, M. and k¨onig Ries, B. (2008). A service distribu-
tion protocol for mobile ad hoc networks. In Inter-
national Conference on Pervasive Services (ICPS 08),
Sorrento, Italy.
Hara, T. (2003). Replica allocation methods in ad hoc net-
works with data update. Mobile Networks and Appli-
cations, 8:2003.
Hauspie, M., Simplot, D., and Carle, J. (2001). Replication
decision algorithm based on link evaluation for ser-
vices in manet. Technical report, University of Lille,
France.
Juszczyk, L. (2005). Replication and synchronization of
web services in ad-hoc networks. Master’s thesis,
Technischen Universit¨at Wien.
Lin, G., Noubir, G., and Rajaraman, R. (2004). Mobility
models for ad hoc network simulation. In The 23rd
Conference of the IEEE Communications Society (IN-
FOCOM 2004). College of Computer & Information
Science Northeastern University Boston.
Madsen, T., Fitzek, F., and Prasad, R. (2004). Simulat-
ing mobile ad hoc networks: Estimation of connec-
tivity probability. In The Seventh International Sym-
posium on Wireless Personal Multimedia Communi-
cations (WPMC’04), Abano Terme, Italy.
Moussaoui, S., Guerroumi, M., and Badache, N. (2006).
Data Replication In Mobile Ad Hoc Networks, chapter
Lecture Notes in Computer Science. Springer Berlin /
Heidelberg.
Park, V. D. and Corson, M. S. (1997). A highly adaptive
distributed routing algorithm for mobile wireless net-
works. In INFOCOM (3), pages 1405–1413.
Wang, K. and Li, B. (2002). Efficient and guaranteed
service coverage in partitionable mobile ad-hoc net-
works. In IEEE INFOCOM02., New York,USA.
WINSYS 2008 - International Conference on Wireless Information Networks and Systems
210