Você está na página 1de 9

MOBILITY AWARE ROUTING WITH PARTIAL ROUTE

PRESERVATION IN WIRELESS SENSOR NETWORKS

Sanat Sarangi, Subrat Kar


Bharti School of Telecommunication Technology
And Management, IIT Delhi
New Delhi 110016 India
sanat.sarangi@gmail.com, subrat@ee.iitd.ac.in

ABSTRACT
We propose a novel reactive routing algorithm called PARTROUTE which achieves
energy efficient routing in sensor networks with stationary and mobile nodes.
PARTROUTE uses coarse mobility awareness in each node to reduce the overheads of
reactive routing by intelligently re-using parts of the created on-demand routes beyond
their standard lifetime. We show that the responsiveness of the network improves while
the overall network traffic is reduced. The effectiveness of PARTROUTE with a large
percentage of mobile nodes is also shown.

Keywords: wireless sensor network, routing protocol, PARTROUTE, mobility


aware, AODV, AODVjr

1 INTRODUCTION Recently, in [10], PH-MA-AODV and Agg-AODV


have been proposed that use mobility awareness to
A typical sensor network has a single gateway improve the performance of reactive algorithms with
and a set of connected nodes that participate in high-speed MANETs. In PH-MA-AODV a highly
sensing and forwarding application defined events to mobile node discards a route-request to prevent itself
the gateway. Some of the key driving factors in from participating in route formation. In Agg-AODV
designing routing algorithms for sensor networks are the route-request accumulates the degree of mobility
(a) size of the network (the number of nodes), (b) of each intermediate node and the destination
limited node energy, (c) deployment topology and chooses the route with minimum mobility. However,
(d) provision for node mobility. Traditionally, there both the algorithms assume that the intermediate
have been two approaches to routing– proactive and nodes can precisely measure their speed of
reactive. In proactive routing, routes can be movement. In [11], prediction of broken links is used
computed and updated beforehand so that data is sent to trigger routing updates before links break.
through the best possible route. In reactive routing, Mobility requirements in most sensor networks
routes are created on-demand. The diversity of are different from those of MANETs. The emphasis
sensor network applications have driven the in sensor networks is on minimizing energy
evolution of classes of application appropriate consumption with provision for (often low) mobility
routing algorithms which can be cluster-based [1], within a stationary mesh of nodes where all nodes
hierarchical [2], robust [3], energy-aware [4], talk to a central gateway. We propose a routing
geographic [5], query based [6] or swarm-behaviour- algorithm PARTROUTE for memory-constrained
based [7]. and energy-sensitive sensor networks with stationary
and mobile nodes where each node is explicitly
When some nodes in the network are mobile,
aware of its mobility status at any point of time and
reactive routing becomes necessary. Dynamic Source
this knowledge is used to intelligently preserve
Routing (DSR) [8] and Ad-hoc On-demand Distance
part(s) of the route(s) formed during reactive routing
Vector Routing (AODV) [9] are popular reactive
beyond the standard route lifetime. PARTROUTE
algorithms developed for mobile ad-hoc networks
dynamically adapts itself to changing network
(MANETs).
conditions and helps improve event delivery while
reducing both (a) the overall network traffic and (b)
2 MOTIVATION
latency of route formation (which is critical) for
mobile nodes.
Most reactive algorithms assume that the nodes
With rapid strides in Micro Electro-Mechanical
in the network are mobile. However, mobile nodes
Systems (MEMS), coarse mobility (moving or
are not explicitly aware of their own mobility.
stationary) awareness can now be easily added to
sensor nodes with the help of low-cost discarded if vj  VM. A mobile node uses only M-
accelerometers. Accelerometers have been used to ROUTE-REQUEST to create a route to the gateway.
detect gait postures like walking, standing, lying [12] All stationary nodes in the network except the
[13]. Such information in the context of the gateway use S-ROUTE-REQUEST to reach the
behavioural pattern of the subject can be used to gateway. The gateway uses M-ROUTE-REQUEST
know whether a node is (and to predict if is likely to to find a stationary or mobile node in the network
remain) stationary or mobile. In a sensor network since M-ROUTE-REQUEST is accepted and
containing such semi-mobile nodes, any semi-mobile processed by all nodes. We assume that the network
node that finds itself stationary could temporarily has a single gateway and each node in the network is
behave like any other (always) stationary node. aware of this gateway. The principle can be extended
PARTROUTE is designed to effectively use this to multiple gateways, without loss of generality.
behavior to significantly improve routing In PARTROUTE stationary nodes are allowed to
performance in such networks. behave like mobile nodes under certain conditions. If
The rest of the paper is divided into four sections. the duration for which mobile nodes rest (i.e. become
Section III describes the algorithm. In Section IV we stationary) is large when compared to the duration
evaluate the performance of PARTROUTE for which they move, they can be allowed to
analytically and compare it with the simulation temporarily function as stationary nodes during the
results given in Section V. PARTROUTE is also stationary phase. The target application determines
compared with AODVjr [14] to show the improved (a) the use of this role-switching feature and (b) the
performance under different conditions. duration of rest. Role-switching is discussed in
Section 3.3.
3 THE PARTROUTE ROUTING
ALGORITHM 3.1 Reactive Route Discovery
We use Expanding Ring Search based route
In PARTROUTE, we consider a sensor network discovery [15], an energy efficient method to
with N nodes v1..vN having a set of stationary nodes discover a route to a destination node. The source
VS and a set of mobile nodes VM. |VS| + |VM| = N and sends a route-request message with TTL set to 1 and
VS VM = N. Any node vi belongs to only one of the simultaneously starts a timer that expires after twice
sets and adheres to the policies laid out for that set. the node-traversal time. If the destination is one hop
L(vi, vj) is a link between vi and vj, i j. In away it sends a route-reply message which reaches
PARTROUTE we have two kinds of links– static the source before the timer expires. If a route-reply is
and dynamic. If two nodes vi, vj VS, they are not received, the source increases the TTL value,
permitted to form a static or a dynamic link. In all proportionately increases the timeout and repeats the
other cases the link is dynamic. Links are created and process till either the destination is reached or the
updated only on reception of packets and a table is source gives up. This limits the propagation extent of
used to maintain link information. A route between the flooded route-request thereby saving energy. The
vi and vj with intermediate node vk is denoted by destination sends the route-reply back to the source
P(vi, vk, vj), i j k. A dynamic link has a lifetime of along the route taken by the first route-request it
lifetimeDYNAMIC and for a static link it is lifetimeSTATIC receives. When a node with a valid route to the
with lifetimeDYNAMIC < lifetimeSTATIC. The value for destination sends a route-reply on behalf of the
lifetimeSTATIC is application specific and is further destination, the reply is called gratuitous [15].
discussed in Sections 4 and 5. It depends on factors
like deployment area, number of nodes, mobility 3.2 Partial Route Preservation
behaviour of nodes and their event generation rates. Partial route preservation implies retaining a part
AODV is an established reactive routing of a reactively created route, for a period longer than
algorithm for MANETs and AODVjr is a successor the standard lifetime for the route. The partial route
to AODV which achieves performance comparable is called a trace. Each node on a trace is called a
to AODV without overheads such as sequence representative. A stationary or mobile node which is
numbers, gratuitous RREP, hop count, hello not a representative is called a non-representative.
messages, RERR and precursor lists from AODV. The gateway is considered to be a representative of
AODVjr is therefore more suitable for memory and itself. In PARTROUTE there can be two kinds of
energy constrained sensor networks. We compare route-replies to a route-request: (a) intermediate-
PARTROUTE with AODVjr. node route-reply that can only be generated by a
PARTROUTE uses two kinds of route-requests: representative, (b) destination-node route-reply that
M-ROUTE-REQUEST and S-ROUTE-REQUEST only the gateway can generate. Henceforth, a node
for route creation as discussed in Section 3.4 and that is allowed to send a route-reply for a route-
Section 3.5 respectively. M-ROUTE-REQUEST is request is called a reply-node. It may be noted that an
processed by vi, vj   VS VM whereas S-ROUTE- intermediate-node route-reply is different from a
REQUEST is processed by vi if vj   VS and gratuitous route-reply [15] which is generated by any
node witth a route to t the destinnation based on reppresentative noodes (Section 3.2). The lasst of these
sequence number infoormation alreaady present with w reppresentative caandidates formmed by a routee-reply (i.e.
the node and that recceived with the t route-requuest. nodde 3 in Fig. 1) sends a crreate-trace (C CT) to the
Sequencee numbers are usedd by neiither gateway and recceives an acknnowledgemen nt (CA) to
PARTRO OUTE nor AO ODVjr. Moreeover, unlike the beccome a reprresentative. S Since a reprresentative
gratuitous route-reply, a representative does not send
s canndidate does not
n know whetther it is the laast one on
any com mmunication to the gateeway about the the path of the rooute-reply, the first mobile node that
intermediiate-node routte-reply that itt generates. gets the route-reeply after a sseries of reprresentative
canndidates sendss a special m message called d a create-
tracce-request (CTR) to the preevious node (the( last of
the representativve candidatess) asking it to t send a
create-trace. Witth no intermeddiate mobile nodes,
n the
rouute request oriiginator sendss the create-traace if it is
the last representtative candidaate. Create-traace request
is not
n used here.. The gatewayy on receiving g a create-
tracce acknowledges its recceipt by sen nding an
ackknowledgemennt to the create-trace originator.
A create-tracce message annd its acknowlledgement
connvert all dynam mic links on thhe path betweeen create-
tracce originator and the gatew way into staticc links by
upddating their link lifetimes to lifetimeSTTATIC. The
create-trace origginator makess a suitable number
n of
retrries to send create-trace
c iff it does not receive
r an
ackknowledgemennt beforre giving
g up.
Ack knowledgemeents may not be received either e due
to (a) lossy linnks or (b) role-switching where an
inteermediate stattionary node on the path of create-
Figure 1: Trace formatioon in PARTROU
UTE tracce (or its ackknowledgemeent) message suddenly
beccomes mobile and voluntariily drops the message.
m
Fig. 1 illustrates trrace formationn in the preseence Traansmission of o a create-trace messag ge by a
of stationnary (black) annd mobile (whhite) nodes whhere reppresentative for
f every inttermediate-no ode route-
mobile noden 6 sendss a route-reqquest (RREQ)) to repply that it gennerates can siggnificantly increase the
which thee gateway repplies. Each staationary node that eneergy consumpption level of the network. We limit
the route--reply (RREP P) encounters on its return path
p thiss behaviour byb defined a vvariable CTPEERIOD such
till it reaaches a mobbile node (noode 4 in Fig. 1) thatt every time a representativve sends a creeate-trace,
becomes a representative candidatte. Representaative it does
d not send a create-tracee for any even nt received
candidatees are conveerted to reprresentatives with w witthin the next (lifetimeSTATIIC - CTPERIOD) seconds,
create-traace messages discussed in Section
S 3.3. 0   CTPERIOD lifetimeSTATIC to reduce the
Statioonary nodes (nnode 5 in Fig. 1) that occur in a oveerheads due too create-trace.. Create-trace messages
route afteer a mobile node
n (node 4)
4 cannot becoome cann also becom me carriers of the nod de ids of
representtatives. The mobile
m node (node
( 4) upddates inteermediate reppresentative caandidates to notify the
the routee-reply with thist informatiion to notify the gateway about thheir presence.
stationaryy nodes that process
p the rooute-reply afteer it
about it. Any subsequuent request for f a route to the 3.4 Route Foormation An nd Route Repair for
gateway by a mobile node can bee serviced by the Stationarry Nodes
representtatives which are closer to the event souurce Stationary nodes onlly use S-ROUTE- S
than thee gateway crreating moree traces on the REEQUESTs to create routees to the gatteway. S-
stationaryy nodes. Thiis helps serviice future rooute- ROOUTE-REQUE ESTs only suupport destinaation-node
requests even
e faster. mode operation. So, gateway is the only reeply-node.
Hencceforth, we usse SR and SU to denote thee set If an
a S-ROUTE--REQUEST iis received by y a mobile
of reppresentatives and noon-representattives nodde, it is simplyy discarded. S
So a route creaated to the
respectively. A repressentative is deenoted by Ri and gateway with S-R ROUTE-REQ QUEST is guaaranteed to
non-repreesentative by Ui, i   I+, wheere I+ denotess the havve only statioonary nodes iimplying that the trace
set of poositive integeers. SR   SU = N (numberr of created in the prrocess is the same as the route.
r If a
nodes in thet network). stattionary node fails
f to form a route, it perfforms role
swiitching (Section 3.6) and behaves like a mobile
3.3 Crreate-trace Messages
M nodde (Section 3.55).
A rooute-reply creeates a set off potential noodes
called reppresentative candidates whiich could becoome
3.5 Rooute Formattion and Rooute Repair for RO
OUTE-REQUE ESTs. This iss acceptable to mobile
Mobile Nodes nod
des since theyy would anywaay change their location.
Mobiile nodes usse M-ROUTE E-REQUESTss to Howwever, repressentatives on tthe trace creaated in the
form rouutes to the gateway.
g A mobile
m node can pro
ocess would help
h provide faaster responsees to other
indicate either
e intermeediate-node orr destination-nnode nod
des generatingg M-ROUTE-R REQUESTs.
mode off operation inn the M-ROU UTE-REQUE EST.
The interrmediate-nodee mode is useed by default and 3.6 Role Swittching
destinatioon-node modde is used under speecial A stationaryy node that is uunable to find a route to
circumstaances as discuussed below. the gateway thhrough other stationary nodes n by
nsmitting a S-ROUTE-RE
tran S EQUEST is allowed
a to
behhave like a moobile node (Seection 3.5). Th his allows
stattionary nodees that havve been iso olated by
depployment or have becoome isolated d during
opeeration possiibly due too non-availaability of
neighbouring staationary nodees to be ablee to route
eveents through VS VM insteaad of only VS. Similarly,
a mobile
m node could become stationary if it remains
idlee and the target appllication justiifies this
connversion afterr consideringg the dynamics of the
mobile node as discussed
d in Section 2.
We now expplain using Figg. 3 how role--switching
by a node particiipating in rouute formation affects
a the
behhaviour of PA ARTROUTE aafter the route is created.
Figure 2: M-ROUTE-RE EQUEST from mobile
m node 100 Fouur cases aree used to analyze the different
extends traace L(1, 2) to P(1,2,6)
P circcumstances encountered
e bby a mobile node M
when it createss a route too the gateway G. A
In the interm
mediate-node mode, any commbination of these cases m may be used to t explain
representtative can be a reply-node and send thee M- howw PARTRO OUTE handlles other situations.
ROUTE-REPLY to an a M-ROUTE E-REQUEST T. A Wh herever appliccable, the conncerned nodee switches
successfuully created rooute between the mobile node n its role immediattely after proccessing the M-ROUTE-
M
and a reppresentative coould extend thet previous trrace REEQUEST. Rolle-switching does not ex xempt the
terminatinng at the repreesentative by adding new sttatic nodde from its responsibilityy to processs the M-
links beyond it or create a neew trace if the ROOUTE-REPLY Y it receives for this M-ROUTE-
M
representtative is the gaateway itself as
a shown in Fiig. 1. REEQUEST.
Even duuring trace extension, the create-trrace
message acknowledgement is geenerated by the
gateway instead of thhe intermediatte node. Exissting
traces get extended byy the recursivvely applying this Fiigure 3: Refereence diagram foor role-switching
g analysis
principle.. In Fig. 2 an M-ROUTE-R REPLY generated
from nodde (representaative) 2 in ressponse to the M-
ROUTE-REQUEST frrom node 10 forms f a route P(1, Case 1: Let us assum me a station nary non-
2, 6, 10) to the gatew way and adds static
s link L(22, 6) rep
presentative Uk lies on P(M M, U1, U2, …, UP,
P R1, R2, …,
to extendd the trace L(1, 2) to P(1, 2, 6). RQ,, G). If Uk swwitches its rolle immediately after it
After creating a route in the intermediate-n
i node pro
ocesses M-RO OUTE-REQU UEST, it forregoes its
mode, if the mobile node n discoverrs that it did not chaance to becom me a representtative after th he route is
receive an
a acknowleddgement from the gatewayy for formmed if K = P or all nodes between Uk and a R1 are
an appliccation level event, it staarts sending M- stattionary.
ROUTE-REQUESTs in i destination-node mode after
a Case 2: If R1 that sent the intermed diate-node
a certainn number off retries (givven by EVEN NT- respponse to thee M-ROUTE E-REQUEST becomes
RETRY-COUNT) usinng the intermeediate-node mode m mobile then {M M, U1, U2, …, UP, R1}  SU an nd {R2, …,
as determ mined by thhe target appplication. Since RQ,, G}   SR. R1 cannot ssend intermed diate-node
gateway is the reply-nnode for this mode, a routte is rou
ute-replies to subsequent M-ROUTE-R REQUEST
created between
b the mobile
m node and the gatew way pacckets since it is not a represeentative any more.
m
possibly creating anothher trace in thhe process. Once
O Case 3: If one
o of the othher representaatives Ri, i
applicatioon level comm munication ressumes, the moobile 1 switches its role, its effe fect is not immmediately
nodes swwitches back too using interm mediate-modee M- appparent since M-ROUTE-R
M REQUESTs arre fielded
ROUTE-REQUESTs for f creating suubsequent routtes. by R1 which com mes before Ri. However, now {M, U1,
Sincee M-ROUTE-REQUESTs are processedd by U2, …, UP, Ri}   SU. If Ri, now a mobile node
all nodess they may form shorterr traces thann S- betwween represenntatives Ri-1 aand Ri+1 generrates a M-
ROUTE-REQUEST, Ri-1 could givve a intermediiate- dennse uniform deployment
d oof sensor nod
des in CK
node respponse withoutt realizing thaat its route too the succh that an eveent transmitteed by M can reach the
gateway passes throuugh Ri, thus creating a cyycle gateway in ⎡x/L⎤⎤ hops throughh (⎡x/L⎤−1) inttermediate
between Ri and Ri-1. When mobille (or stationnary) noddes on a linearr path betweenn M and G.
nodes thhat use routes with L (Ri-1, Ri) as an
intermediiate link fail to receive accknowledgem ments
for eventts they sends to the gatewaay, they perceive
this as a broken link problem andd use destinattion-
node M-ROUTE-RE
M EQUESTs (or S-ROU UTE-
REQUES STs) to find new routes to the gatew way
breaking the above cyycle in the prrocess unless the
new routees do not use the
t nodes in thhe cycle. If noodes
in the cyycle are not ussed, the cyclee disappears after
a
link liffetimes expire due to absence of
communiication througgh the links.
Thereefore, rapid roole-switching by
b a mobile node
n
could advversely affect performance of PARTROU UTE.
So we coonstrain fidgetty mobile noddes to always stay
mobile to minimizee the overhheads due such s
conditionns.
Case 4: If a mobilee node UK swiitches its role and
becomes stationary just after proocessing the M- Figure 4: A circullar sensor netwoork deploymentt for
ROUTE-REQUEST it either becomes a anaalysis
representtative provided K = P (sincce UP is adjacent
to R1) or all nodes betw ween UK and R1 are stationnary. 4.2 Estimatin ng Route F Formation Response
Else, UK remains a stattionary non-reepresentative. Time
Suppose at a given instannt T, the mobille node M
4 RFORMANCE
PER E EVALU
UATION OF is at
a a distance x from the G and does not n have a
PAR
RTROUTE rou
ute to the gatewway. It generaates a route-reequest and
exppects a route-rresponse. Delaay (D ) experrienced by
A large delay
d in routee formation iss a bottleneckk for M ata a distance x is the duratiion between thet time at
reactive algorithms
a duee to their on-ddemand naturee. In which it originattes the route-rrequest and th he time at
the earlieer sections, we
w have seenn how traces can which it receives the route-reeply. So the cumulative
c
successfuully help reducce this delay for
f mobile noodes. delay experienceed by M for alll its positionss on a ring
In this Seection we builld an analytical model in order at distance
d x andd (infinitesim
mal) width dx from G is
to characcterize PART TROUTE’s performance
p w
with D . 2πxdx. Hencee, for a given algorithm A, the mean
respect too delay in rooute formatioon. The modeel is Ressponse Time (RTA) experieenced by M positioned
p
used to analyze
a the simulation ressults presentedd in anyywhere in CK is obtainned by averaaging the
Section 5.
5 cummulative delayy over all posiitions at distannce x, 0 
x  K in CK. Thiis is given as
4.1 Asssumptions
The noddes are idenntical with an a ideal circcular K
D .
transmisssion range of o radius L. The analysiss is RT
TA  K 1  
carried ouut with respecct events sent to the gatewaay G
by a mobbile node M moving
m withinn a circular reggion
CK with radius
r K as shown
s in Fig.4. The changge in With the asssumptions in Section 4.1, the route
position of M in thee interval bettween sendinng a req
quest from M is received aat the same tiime by all
route-reqquest and geetting back its responsee is noddes within a circle
c of radiuus L around M. These
negligiblee. All transm
mission, channnel and recepption noddes again broaadcast the rouute-request to repeat the
conditionns are ideal soo a route-respponse is receiived pro
ocess. In AOD DVjr only G iss allowed to respond
r to
for everyy route-requesst. Radio proppagation delaay is the route requestt. So, the routee request wouuld need at
negligiblee when comppared to the trransmission delay leasst x/L   hops to reach G. This is also the worst
at the meessage sourcee. So, the Hopp Traversal Time T casse situation foor PARTROU UTE since thee gateway
(denoted by τ, τ 0) for fo a message to be receivedd by wou uld have to fiield all route-requests in th
he absence
any nodde within a radius L of o the messsage of traces.
t For this case Dx   2ττ x/L . 
transmitteer is essentiaally the same.. Mean queueeing For PARTRO OUTE, the route-reply may y be given
and proceessing delays are consideredd to be a part of τ by an intermediate node if itt is a represen ntative. A
since theyy are applicabble to every noode. We assum me a presentative coould be encouuntered at n ho
rep ops where
1 n x/L x . Mobilityy pattern of M is random, trrace retrransmission of
o the same evvent instead of
o the next
lifetimes are configguration speccific and trrace eveent. A single gateway, a sset (R) of 30 randomly
formationn depends on earlier route formaation depployed non-evvent- generatinng nodes (con
nfigured to
attempts and percenntage of sttationary noodes. be static or mobbile as per reequirement) an nd one or
Assumingg n takes eacch value withhin its range with w more event-generating mobiile nodes aree used to
equal proobability, the mean value ofo Dx is givenn by chaaracterize thee performancce of PART TROUTE.
/L Moovement of mobile nodes iss random way ypoint [8].
Dx    2τ ∑ n / x/L     τ x//L 1 .
In thee best case, all
a route-requeests are fieldeed at Nodes randomlyy choose a loocation and a velocity
h So Dx   2
the first hop. 2τ.  betw ween 1.0m/ss to 2.0m/s. After movin ng to the
Loweer the Respponse Time, lesser is the location with thhe chosen veloocity the nexxt location
possibilitty of the gateway and nodes near it gettting andd velocity aree chosen. It may be notted that a
choked with
w route-requuest packets thhat could degrrade stattionary nodee may be eeither (a) an n always
performaance in a livee scenario. Also,
A decreasee in stattionary node or (b) a role--switching semmi-mobile
Responsee Time imprroves perform mance for high h nodde (discussed in Section 2) which is stattionary for
speed moobile nodes (further discuussed in Secction the duration of siimulation.
5.2.2). We
W also definee a figure of merit
m Trace Inndex
(TI) whicch is used to compare thee effectiveness of
PARTRO OUTE when compared
c to AODVjr.
A Smaaller
the valuue of TI, betterb the reesponsiveness of
PARTRO OUTE.

RTPARTROUTE
P
TI                                       2  
RT
TAODV

4.3 Efffect of Multip ple Event Souurces


The performance
p of PARTROU UTE improvees as
the numbber of event soources increase since (a) more
m
traces arre created and a (b) exissting traces are
preservedd for longer duurations. More traces help give
g
faster ressponses to rouute-requests frrom mobile noodes
thereby controlling
c thhe flooding of routing conntrol
packets. IfI a node in the t network has
h a trace too the Figure 5: A quarteer-circular regioon with random
mly
gateway, for this traace to persisst for the enntire dep
ployed nodes annd a gateway
lifetime of the netwoork it is impportant that it i is
refreshedd with eventts at intervaals smaller than t Nodes are deployed in a quarter-circu ular region
lifetimeSTTATIC. QK (a quarter of CK in Sectionn 4.1) with rad
dius 300m
Deplooyment param meters especiially size of the (Figg. 5). QK is divided
d into 3 sectors withh 10 non-
area and the mean ratee of event genneration shouldd be eveent-generatingg nodes in eacch sector. Thee range of
used to estimate a suittable value forr lifetimeSTATIIC in eacch node (L) is 75m. From m Section 4.1,, an event
order to have
h effective traces. gennerated by a node farthestt (i.e. 300m) from the
gateway needs 3 intermediatee nodes separrated from
5 ULATION
SIMU eacch other by 755m to reach tthe gateway. However,
we use 10 nodees in each seector to (a) su ufficiently
Discrete event simulaation with sim mulation runs for incrrease the density
d of nnodes, (b) keep the
1000 secoonds is used to
t evaluate thhe performancce of disttribution of nodes acrosss the deploym ment area
PARTRO OUTE. The efffect of multipple event souurces hommogeneous evven with mobbility and (c)) create a
on PART TROUTE is analyzed.
a Thee results obtaiined netwwork of reassonable size ffor analysis. It can be
for PARTROUTE aree compared against a AOD DVjr. shoown that Trace Index for CK (in Section n 4.2) also
Finally, Trace
T Index diiscussed in Seection 4 is useed as hollds for each of
o its four commponent quarrter circles
a measurre of the Respponse Time of the networkk for (cenntered at G with
w radius K)) and hence for fo QK. QK
PARTRO OUTE. alsoo serves as a a referrence for analyzing
PAARTROUTE’s performancee in topologiies where
5.1 Coonditions the gateway is at a a corner raather than the centre of
An ideal
i circular radio modell is used forr all the network.
nodes. There
T are no packet lossees and nodes are Each nonn-event-generaating nodess when
always awake.
a For every eventt that the node
n connfigured to bee mobile movves only withiin its own
transmits, it receives an
a acknowleddgement fromm the secctor. Event-generating nodee(s) move in the whole
gateway. Non-receipt of acknowleddgement leadds to depployment areaa. The range oof each node is kept at
75m to have
h sufficientt overlap betw
ween transmission perrcentage of thhe stationary nnodes from R are made
range off nodes. Paacket size iss kept consttant, mobile and the performance
p oof both the alg
gorithms is
lifetimeDYYNAMIC is keptt at 15s, CTPEERIOD is set to 30s commpared by vaarying this perrcentage from m 0 to 80.
and EVE ENT-RETRY Y-COUNT is set to 3. The Fig
gs 7, 8 and 9 show
s variationn of the Respoonse Time,
estimatedd Hop Traverrsal Time is 0.1 seconds. An Tottal Packet Count andd Event Count C for
event transmission is i attempted by an evvent- PAARTROUTE-150 and AODV Vjr for 0-80%
% mobility.
generatinng mobile nodde every 6 secconds. If a moobile It is observeed from Fig.. 7 that for stationary
node in PARTROUTE
P E fails to delivver an event after
a noddes, Responsee Time for PA ARTROUTE-1 150 is less
the retriees defined byy EVENT-RETRY-COUNT T, it thann half of thatt of AODVjr i.e. PARTRO OUTE-150
uses d
destination-no ode M-ROOUTE-REQUE EST is almost twicee as responssive as AOD DVjr. The
(Section 3.5) till it suuccessfully delivers
d an evvent. Ressponse Time remains signiificantly smalller (60%)
Henceforrth we use PARTROUTE-x to referr to eveen when up to 40% nodes arre mobile. As expected,
PARTRO OUTE with thee value of lifeetimeSTATIC ass x. when almost all nodes (80% nodes) are mobile, m the
We use
u the term Total Packet Count to dennote twoo algorithm ms have ssimilar perrformance.
the total number
n of (daata and routingg control) pacckets
transmitteed by all nodees. Event Packet Count is usedu
to denotee the total nummber event paackets transmiitted
by all noodes. Event Count
C refers to
t the numbeer of
events seent by an evennt source for which it receeives
an acknow wledgement. So, the total number
n of pacckets
transmitteed for everyy event (TP PE) successffully
transferreed to the gatewway is given by
b

T  P  C
TPE                3  
E  C

5.2 Reesults
5.2.1 Effect
E of multiiple event souurces
Fromm Fig. 6 it is cllear that multiiple event souurces Figure 7: Variatioon of Response Time (in secon
nds) for
reduce thhe Response TimeT of the neetwork as perr the AO
ODVjr and PAR RTROUTE-150 for different nu umbers of
discussion in Section 4.3. Also, forr a given num mber mobbile nodes (0%: all nodes stationary)
of event sources
s once the
t created traaces get refresshed
periodicaally, any furtther increase in the valuee of For 0% mobility, eventss in PARTRO OUTE-150
lifetimeSTTATIC does not signifiicantly imprrove are sent over (thhe long) tracees created a priori
p over
performaance e.g. forr a single event e source no the stationary nodes whereeas in AOD DVjr, the
significannt improvement in responsee time is obserrved insttantaneously created routees to the gatteway are
by increaasing lifetimeSTATIC
S from 1500 to 200. typically shorterr. So, marginally higher TPE
T levels
for PARTROUT TE-150 when compared to o AODVjr
(Figg 8) are duee to PARTRO OUTE-150’s relatively
hig
gher Event Paccket Count. HHowever, as discussed in
Secction 4.2, wheen the event-ggenerating moobile node
moves at a (rellatively higheer) speed off 4-5 m/s,
PAARTROUTE-150 outperfforms AOD DVjr by
delivering 3% more
m events wiith 21% lowerr TPE due
to its
i significantlly lower Response Time.

Figure 6: Variation of Reesponse Time (in ( seconds) witth


lifetimeSTAATIC (lifetimeSTTATIC takes values 50, 100, 1500,
200 seconds)

5.2.2 Comparison
C off PARTROUT
TE with AODV
Vjr

Respponse Time foor PARTROU UTE for a sinngle


mobile evvent source dooes not vary significantly
s a
after
150s as shown in Figg. 6 So, PAR RTROUTE-150 is Figure 8: TPE forr AODVjr and P
PARTROUTE-150
used forr comparisonn with AOD DVjr. A cerrtain
With mobility, lenngths of tracess shorten so TPE
T
decreasess. For 400% mobiliity, TPE for
PARTRO OUTE-150 is i 74% off AODVjr– an
improvem ment of 26%. With 80% mobility, duee to
lack of sufficient
s stattionary nodess PARTROU UTE-
150 and AODVjr havve similar TP PE levels. Evvent
Count forr PARTROUTE-150 is moore than AOD DVjr
for all caases except with
w 0% mobbility where it i is
almost saame. PARTRO OUTE-150 deelivers up to 15%1
more eveents than AOD DVjr with 40%% mobility.
To seee the effectss of semi-mobbility (Sectionn 2)
with a larrge stationaryy period withiin the durationn of
the simullation we connfigure each node
n in R to keep
k
switchingg its status beetween mobille and stationnary.
The duraation for beinng mobile annd stationary are Figure 10: A grid deployment off nodes
given byy two normallly distributed variables with w
mean vallues of 450s and
a 50s respecctively, each with
w 5.2.3
5 Variationn in Trace Inddex
a standarrd deviation of 10s. Thiis prevents role- r From Sectioon 4.2, theorretical values of Trace
switchingg from happenning synchronnously. Each node
n Ind
dex with L=755 and K=300 aare given as 1, 0.66 and
is initiallly mobile. PAARTROUTE--150 delivers 5% 0.32 for worst, mean and bbest case resspectively.
more eveents, has a 17% % lower respoonse time andd 8% g.11 shows hoow the simullation results for Trace
Fig
lower TP PE compared to AODVjr. Ind
dex for PART TROUTE-1500 compare ag gainst the
theoretical worsst, mean and best case leevels. The
west value of 0.44
low 0 is observed for Trace Index
I with
stattionary nodess. With 40% mobility, the observed
Traace Index is 0..6 which is beetter than meaan value of
0.66.

Figure 9: Event Count foor AODVjr andd PARTROUTE


E-
150
Figure 11: Trace Index
I Analysis
A toopology in Fig. 5 withh random node n
deploymeent has been used
u to characcterize the genneral
behaviouur of PARTR ROUTE. How wever, a typpical 6 CONCLUSIION
commonlly used configuration wouuld look like Fig.
10 wheree 50% of the stationary noodes in R (i.e. 15
nodes annd the gatew way) form a connected grid Wee have shown how logicallly dividing the network
through which
w events can be easily routed and the intoo stationary and
a mobile noodes with parrtial route
remainingg 50% move throughout thhe grid usingg the (traace) preservattion over stattionary nodess can help
random-w waypoint mobbility model discussed
d earrlier. incrrease the efffectiveness oof reactive routing
r in
There is one evennt-generating mobile noode. mobility awaree sensor nnetworks wh here the
Remaininng conditionss are the same s as earrlier. imp portance of energy-efficciency is paramount.
Transmisssion range of nodes annd the distaance Perrformance of PARTROU UTE under different
between adjacent nodees on a row or column is 75m. 7 connditions has been show wn to illusstrate its
PARTRO OUTE-150 delivers 12% more m events when
w appplicability too different situations. Mobility
comparedd to AODVjr at a 20% low wer response time
t chaaracteristics of
o the nodes comprising the t sensor
and 20% lower TPE. With W appropriiate adjustmennt of netw work, naturee of event traffic and d typical
node trannsmission rangge, the configuuration also helps
h depployment topoologies couldd be used to configure
in carryinng out localizaation [16] [17]]. PAARTROUTE to bring about significant
s
improvements on all routing performance parameters. on Systems and Networks Communications,
PARTROUTE can also find significant application Washington, DC, USA: IEEE Computer
with Body Sensor Networks where the mobility Society, pp. 25–29 (2009).
pattern of nodes can be determined with sensors like [11]S. Ju and J. Evans: Mobility-aware routing
accelerometers. protocol for mobile ad-hoc networks, IEEE Intl.
Conf. on Communications Workshop, 2009, pp.
7 REFERENCES 1–6 (Jun. 2009).
[1] W. R. Heinzelman, A. Ch, and H. Balakrishnan: [12]M. Mathie, B. Celler, N. Lovell, and A. Coster:
Energy-efficient communication protocol for Classification of basic daily movements using a
wireless microsensor networks, IEEE Proc. triaxial accelerometer, Medical and Biological
Hawaii Intl. Conf. Sys. Sci., pp. 1–10 (Jan. Engineering and Computing, vol. 42, pp. 679–
2000). 687, 2004.
[2] A. Manjeshwar and D. P. Agrawal: Teen: A [13]Q. Li, J. Stankovic, M. Hanson, A. Barth,
routing protocol for enhanced efficiency in J. Lach, and G. Zhou: Accurate, fast fall
wireless sensor networks, Intl. Parallel and detection using gyroscopes and accelerometer-
Distributed Processing Symp., vol. 3, pp. derived posture information, Sixth Intl.
2009–2015 ( 2001). Workshop on Wearable and Implantable Body
[3] D. Ganesan, R. Govindan, S. Shenker, and Sensor Networks, pp. 138–143 (2009).
D. Estrin: Highly-resilient, energy-efficient [14]Chakeres, I. D., and L. Klein-Berndt: Aodvjr,
multipath routing in wireless sensor networks, aodv simplified, SIGMOBILE Mob. Comput.
SIGMOBILE Mob. Comput. Commun. Rev., Commun. Rev., vol. 6, no. 3, pp. 100–101
vol. 5, no. 4, pp. 11–25 (2001). (2002).
[4] R. C. Shah and J. M. Rabaey: Energy aware [15]C. E. Perkins, E. M. Belding-Royer, and S. Das:
routing for low energy ad hoc sensor networks, Ad hoc on-demand distance vector (aodv)
Proc. of the IEEE Wireless Communications routing, RFC 3561 (Jul. 2003).
and Networking Conf. (WCNC), Orlando, [16]S. Sarangi and S. Kar: Performance analysis of
Florida ( 2002). an improved graded precision localization
[5] B. Karp and H. T. Kung: Gpsr: greedy perimeter algorithm for wireless sensor networks, Intl. J.
stateless routing for wireless networks, of Computer Networks & Comm. (IJCNC),
MobiCom ’00: Proc. of the 6th Annual Intl. vol. 2, no. 4, pp. 150–159 (Jul. 2010).
Conf. on Mobile Computing and Networking, [17] S. Sarangi and S. Kar: Location estimation with
New York, NY, USA: ACM, pp. 243–254 reactive routing in resource constrained sensor
(2000). networks, Intl. Conf. on Sensors and Related
[6] C. Intanagonwiwat, R. Govindan, and D. Estrin: Networks (SENNET’09), vol. II, Vellore, India,
Directed diffusion: A scalable and robust pp. 563–567 (Dec. 2009).
communication paradigm for sensor networks,”
in MOBICOM, ACM, pp. 56–67 (2000).
[7] L. Wang, R. Zhang, and S. Geng: An energy-
balanced ant-based routing protocol for
wireless sensor networks, WiCOM’09: Proc. of
the 5th Intl. Conf. on Wireless communications,
Networking and Mobile Computing,
Piscataway, NJ, USA: IEEE Press, pp. 3556–
3559 (2009).
[8] D. B. Johnson and D. A. Maltz: Dynamic source
routing in ad hoc wireless networks, Mobile
Computing, ser., The Kluwer International
Series in Engineering and Computer Science,
T. Imielinski and H. F. Korth, Eds, Springer
US, vol. 353, pp. 153–181 (1996).
[9] C. Perkins and E. Royer: Ad-hoc on-demand
distance vector routing, Proc. of Second IEEE
Workshop on Mobile Computing Systems and
Applications, 1999. WMCSA ’99., pp. 90–100
(Feb. 1999).
[10]Y. Khamayseh, O. M. Darwish, and S. A.
Wedian: Ma-aodv: Mobility aware routing
protocols for mobile ad hoc networks,
ICSNC ’09: Proc. of the 2009 Fourth Intl. Conf.

Você também pode gostar