USC Computer Science Technical Reports, no. 707 (1999)

PDF

Download

Open document

Flip pages

Download a page range

Christopher Ho, Katia Obraczka Gene Tsudik, Kumar Viswanath. "Flooding for reliable multicast in multi-hop ad hoc networks." Computer Science Technical Reports (Los Angeles, California, USA: University of Southern California. Department of Computer Science) no. 707 (1999).

Transcript (if available)

Content Flo o ding for Reliable Multicast in MultiHop Ad Ho c Net w orks
Christopher Ho Katia Obraczk a
Gene Tsudik Kumar Visw anath
y
USC Information Sciences Institute
Admiralt yW a y Marina del Rey CA USA
T el F ax April Abstract
Ad Ho c Net w orks are gaining p opularit y as a result of adv ances in smaller more v ersatile and p o w erful
mobile computing devices The distinguishing feature of these net w orks is the univ ersal mobilityof all
hosts This requires reengineering of basic net w ork services including reliable m ulticast comm unication
This pap er considers the sp ecial case of highly mobile fastmo ving ad ho c net w orks and argues that
for suchnet w orks traditional m ulticast approac hes are not appropriate Flo o ding is suggested as a
p ossible alternativ e for reliable m ulticast and sim ulation results are used to illustrate its eects The
exp erimen tal results also demonstrate a rather in teresting outcome that ev en o o ding is insucien t for
reliable m ulticast in ad ho c net w orks when mobilityis v ery high Some alternativ e more p ersisten t
v ariations of o o ding are sk etc hed out
In tro duction
Recen t adv ances in p ortable computing devices and wireless comm unication tec hnology ha vemade
it p ossible to stay c onne cte d anywher e anytime In the near future users will b e able to mo v e
freely and still ha v e seamless reliable and highsp eed net w ork connectivit yP ortable computers
and handheld devices will do for data comm uni cation what cellular phones are no w doing for v oice
comm unic ation
T raditional net w ork mobilit y fo cused on roaming whic hisc haracterized b y hosts connecting to the
xedinfrastructure in ternet at lo cations other than their w ellkno wn home net w ork address Hosts
can connect directly to the xed infrastructure on a visited subnet through a wireless link or a
dialup line These socalled traditional or xedinfrastructure mobile net w orks raise issues suc h
as address managemen t but do not require signican tc hanges to core net w ork functions suchas
routing
Multihop AdHo c Net w orks AHNs refer to mostly wireless net w orks where all net w ork com
p onen ts are mobile In an AHN there is no distinction b et w een a host and a router since all net w ork
Con tact author email k atiaisiedu
y
This w ork is b eing supp orted b y NSFNGI gran tn um b er ANI

hosts can b e endp oin ts as w ell as forw arders of trac In con trast with xedinfrastructure net
w orks AHNs require fundamen tal c hanges to net w ork routing proto cols including m ulticast routing
and pac k et forw arding
Little has b een done todate as far as pro viding supp ort for m ulticast comm unication in AHNs A
ma jor c hallenge lies in ac hieving reliable m ulticast comm unication in en vironmen ts with univ ersal
mobilit y and frequen t no de outages and failures
T o this end this pap er explores the limits of reliable m ulticast in v ery dynamic highmobilit y AHNs
and motiv ates the need for new m ulticast routing proto cols aimed sp ecically at suc h net w orks W e
fo cus on studying o o ding as an alternativ eto m ulticast routing in v ery dynamic AHNs Flo o ding
emphasises minimal state and high reliabilit y whic h mak es it v ery attractiv e for highly dynamic
fastmo ving AHNs W e presen t preliminary sim ulation results that test the limits of o o dings
reliabilit y as a function of mobilit y Multicast in AHNs
Regardless of the underlying net w ork en vironmen t net w orklev el m ulticast is the fundamen tal en
abling tec hnology for collab orativ e and more generally group comm uni cation applications In b oth
military eg command and con trol in battleeld scenarios and civilian eg disaster relief en vi
ronmen ts grouporien ted services suc h as teleconferencing and data distribution are an ticipated to
b e some of the k ey applications for AHNs
The c hallenges raised bym ulticast routing and pac k et forw arding in AHNs are essen tially due to
their unconstrained mobilityc haracteristics Unconstrained mobilit y implies the follo wing
Individual host b eha vior indep enden t of other hosts
Essen tially no limit on host sp eed
No constrain ts on direction of mo v em en t
High probabilit y of frequen t temp orary net w ork partitions
The ab o v e translates in to frequen t top ology c hanges whichmak es it dicult for a host to main tain
timely m ulticastrelated state information mem b ership other than its o wn F urthermore in man y
t yp es of AHNs eg where hosts are handheld devices b oth storage capacit yand po w er are sev erely
limited This is y et another reason to a v oid main taining and exc hanging m ulticast state
Another imp ortan t consideration has to do with the mission of AHNs In the critical en vironmen ts
AHNs are most often deplo y ed b oth in military and other emergency situations robustness
and high qualit yofservice are of paramoun t concern Th us m ulticast mec hanism s ho w ev er
attractiv e otherwise that cannot pro vide the highest deliv ery guaran tees ma y not b e appropriate
Only recen tly routing proto cols for AHN m ulticast ha vebecomean activ e area of researc h Ex
amples of curren t researc h eorts include the OnDemand Multicast Routing Proto col ODMRP
from UCLA the m ulticast extensions to CED AR b y UIUC and the Ad ho c Multicast Rout
ing proto col utilizing Increasing idn um b erS AMRIS from National Univ ersit y of Singap ore and

GeorgiaT ec h The common denominator among these approac hes is that they all require state
to b e main tained bynet w ork elemen ts In con trast o o ding results in minim al state reten taion and
high reliabilit y whichmak es it a viable candidate for m ulticast proto cols in v ery dynamic AHNs
W e also note that a single cureall m ulticast solution for all AHNs is highly unlik ely Relativ ely stable
AHNs with few host failures outages and infrequen tmo v em en ts will lend themselv es to approac hes
dieren t from those b estsuited for highly dynamic and highly v olatile AHNs Consequen tly one of
the longterm c hallenges is in determining the b est m ulticast approac h when faced with a sp ecic
AHN conguration and parameters Ween visage a suite of solutions eac h geared to w ards and
nearly optimal in the con text of a sp ecic AHN t yp e and host mobilit y pattern This is v ery
m uc h in line with m ulticast dev elopmentin xed net w orks suchasthe In ternet The t womodes of
Proto colIndep enden t Multicast PIM sparse and dense mo des are a case in p oin t
Desired Multicast Prop erties
Multicast routing in highlymobile AHNs m ust emphasize the follo wing
Robustness v ersus Eciency Manym ulticast routing approac hes rely on state in routers to k eep
trackof m ulticast group mem b ers This coupled with the high v olume of routing information
exc hanges and slowcon v ergence mak e traditional m ulticast approac hes un tenable in highly dynamic
AHNs comp osed of anemic lo wp o w er lo w storage capacit y hosts Therefore new tec hniques that
stress rapid and robust deliv ery m ust b e dev elop ed
Adaptabili t y AHN b eha vior can c hange o v er time ie a v ery mobile AHN can stabilize in part
or as a whole or similarly a relativ ely static AHN can suddenly b ecome v ery mobile Dieren t
m ulticast mec hanisms are appropriate for highmobilit yand lo wmobilit y AHNs The same holds
for sparse and dense AHNs Ideally hosts should b e able to adapt to AHN b eha vior c hanges b y
dynamically switc hing among dieren tm ulticast mec hanisms This m ust b e done with the minim um
of b oth eort and incon v enience eg pac k et loss Also o v er time an AHN can exp erience drastic
c hanges in terms of its collectiv e mobilit y or con v ersely its stabilit y Unlimit ed Mobilit y Some m ulticast solutions are geared to w ards discr ete mobilit y whereb y p erio ds
of mo v em en t are in tersp ersed with p erio ds of rest Some others assume limits on direction sp eed
and n um b er of sim ultaneously mo ving hosts In con trast w e stress c ontinuous and high mobilit y
of al l AHN comp onen ts
In tegrated Multicast Multicast solutions for AHNs will most lik ely dier substan tially from those
for xed net w orks one of the main reasons is the mark ed dierence in transmission rates In
order to oer seamless and in tegrated m ulticast service new mec hanisms m ust b e dev elop ed for
in terop eration of xed and wireless m ulticast solutions

Problem Scop e
As men tioned ab o v e our fo cus is on m ulticast routing in highlymobile AHNs This is a c hallenging
and un til recen tly relativ ely littleexplored area W e note that in con trast m ulticast routing in
xedinfrastructure mobile net w orks presen ts only a few engineering obstacles These issues whic h
mainly ha v e to do with the last hop deliv ery and mem b ership trac king are not discussed in this
pap er
Multicast mec hanisms for other less dynamic t yp es of AHNs ha v e b een prop osed As discussed
in Section relativ ely slo wmo ving AHNs lend themselv es to adaptations of traditional state
based m ulticast metho ds suc h as PIM Alternativ elyslo wmo ving AHNs can b e amenable
to m ulticast extensions of disco v erybased or ondemand routing metho ds suc h as DSR Also m ulticast in hierarc hical or clustered AHNs with fairly static clusterlev el top ology and little
in tercluster migration can b e accommo dated b y traditional m ulticast metho ds for in tercluster
routing Our starting h yp othesis is that reliable m ulticast in v ery dynamic highsp eed AHNs is un tenable
with traditional m ulticast mec hanisms whic h are based on state built b y routers Ondemand
AHN m ulticast metho ds op erate b y amortizing the cost of o o dingbased destination disco v ery
o v er subsequen t data pac k ets Ho w ev er collecting paths to a set of destinations is useless if it is
all but certain that the destinations will mo veb y the time the next pac k et is sen t Proactiv e state
based metho ds deriv ed from PIM eg STWM suer from the same problem Keeping accurate
state ab out the m ulticast group mem b ership of all no des neigh b ors is v ery dicult if the set of
neigh b ors c hanges at a v ery high rate
In ligh t of the ab o v e it seems natural to consider plain o o ding whic h although hea vyhanded in
terms of o v erhead t ypically obtains the b est results in terms of reliable deliv ery Ho w ev er w e
p ostulate that in the t yp es of AHNs w e consider o o ding do es not pro vide the high reliabilit y
comparable to that it pro vides in static net w orks or more stable AHNs Highsp eed no de mo v eme n t
can preclude reception of a pac k et
Sim ulations allo w us to demonstrate this h yp othesis b y deriving concrete AHN scenarios where
plain o o ding fails Informally fails means the follo wing
Assume that p represen ts p er hop pac k et transmission delayand d is the maxim um AHN
diameter Apac k et is sentat time t
and o o ded throughout the AHN Ev ery no de
that receiv es the pac k et broadcasts it exactly once to all imme diate neigh bors Let
t
t
dp Assuming that the AHN sta ys con tin uously connected
bet w een time t
and t
there exists at least one no de that has not receiv ed the pac k et
P arameters
Sim ulations are in v aluable when ev aluating the p erformance of net w ork proto cols They allo w
extensiv e exploration of a proto cols design space This includes sub jecting the proto col to extreme
Am ulticast a v o red DSR w ould use p ersource o o ding to build a path tree to m ulticast group mem b ers
Connected means that there is alw a ys a path b et w een anyt wonodes
or b oundary conditions whic h are often hard to repro duce in reallife or liv e exp erimen ts
Th us when sim ulating a proto col it is essen tial to determine b oth the metrics b y whichto ev aluate
the proto cols p erformance as w ell as the dimensions of the proto cols design space The latter are
t ypically used as parameters of the sim ulator represen ting knobs that can b e indep enden tly turned
in order to sub ject the proto col to a range of conditions
This exercise is particularly hard when ev aluating m ulticast proto cols for AHNs AHNs add mo v e
men t a complex dimension to the already in tricate m ulticast proto col design space Since AHNs
are a relativ ely incipien t researc h area w ellestablished mobilit y patterns to b e used when sim ulating
AHNs are not curren tly a v ailable
Aside from no de mobilit y the other design space dimensions of m ulticast routing in AHNs are listed
b elo w W e note that this is not an exhaustiv e parameter list
Applicationsp ecic parameters include total n um b er of messages sentb y eachnode orthe
no des transmit rate and the in terv al no des w ait b efore forw arding a pac k et they receiv e
Multicast group parameters include n um b er of no des or no de densit y Since w e are sim ulating
broadcast the total n um b er of no des is equal to the n um b er of recipien ts In the case of
m ulticast these v alues are t ypically dieren t W e set the n um b er of no des high enough to b e
in teresting but again lo w enough to readily sim ulate W e sp ecically a v oided a n um ber so
lo w as to allo w longliv ed net w ork partitioning
T errainrelated parameters include the dimensions of the eld p erimeter In the case of
threedimensional terrains one can also sp ecify heigh t This is useful when sim ulating ph ysical
obstacles ho w ev er mobilit ycapable net w ork sim ulators lik e GloMoSim or ns with mobilit y
extensions b y CMU curren tly do not supp ort it
No de capabilit y parameters include the no des transmission p o w er range and bandwidth
In Section w e discuss the sim ulation parameters used in our exp erimen ts in more detail
Sim ulation En vironmen t
F or the sim ulations p erformed in this studyw e used the Global Mobile System Sim ulator or
GloMoSim dev elop ed at the Univ ersit y of California Los Angeles GloMoSim is a librarybased
sequen tial and parallel sim ulator for wireless net w orks It is designed as a set of library mo dules
eachof whichsim ulates a sp ecic wireless comm uni cation proto col in the proto col stac k The library
has b een dev elop ed using P ARSEC a Cbased parallel sim ulation language whic hcan be used
to program new proto cols and mo dules that can b e added to the library GloMoSim has b een
designed to b e extensible and comp osable The comm unication proto col stac k for wireless net w orks
is designed using a la y ered approac h where eac hla y er has with its o wn API Mo dels of proto cols
of one la y er in teract with those of other la y ers via these APIs

Our Mo dications to GloMoSim
F or our sim ulations w e used GloMoSim v ersion the v ersion of GloMoSim a v ailable at the
time w e started our exp erimen ts The only mobilit y mo del supp orted in GloMoSim kno wn
as the drunken mobility mo delw orks as follo ws A t the start of the sim ulation eac h mobile no de
is assigned a random p osition within the eld When the no de is next considered for mo v em en t
the mobilit y mo dule c hec ks all the p ossible directions in whic h the no de can mo v e to ensure that
it sta ys within the eld b oundaries With probabilit ygiv en b y mobility pr ob ability the no de then
mo v es in the direction randomly c hosen from the set of p ossible directions Eac hnode mo v es b y
one unit distance in that direction during the mobility interval Both the mobilit y probabilit y and
in terv al are parameters of the sim ulator This mo v eme n t pattern is also kno wn as r andom walk When w e traced no de mo v emen t during the sim ulation w e found that no des generally exhibited
oscillatory mo v eme n t patterns ie they tended to mo vebac k and forth ab out their original p ositions
in the eld Consequen tly no des nev er mo v ed signican tly in an y direction
Too v ercome this problem w e mo died GloMoSims original mobilit y mo del in order to bias no de
mo v em entto w ards the direction randomly c hosen when the no de is rst selected for mo v em en t
When the no de reac hes the eld b oundary it c ho oses another direction and pro ceeds
Another mo dication w e made to the mobilit y mo del w as to parameterize the distance co v ered b y
a no de during eac h mobilityin terv al This allo w ed us to increase the distance no des co v er in a
mobilityin terv al and therefore their sp eed Besides the original one meter distance w e also used
and meters in our sim ulations
Flo o ding Application
Our o o ding application sim ultaneously runs on n no des whic h are randomly placed in the sim ulated
eld Throughout the course of the sim ulation eac h of the no des attempts to broadcast m messages
to all other no des in the net w ork Th us the application exp ects the no des to collectiv el y receiv e
m n messages When a no de receiv es a pac k et it w aits a uniformly distributed time in terv al
bet w een and f l ooding inter v al b efore it broadcasts the pac k et n mand f l ooding inter v al
are parameters of the sim ulator
W e counteac h time a message from an y sender fails to reachan y receiv er The dierence b et w een
the exp ected and the actual n um b er of messages collectiv ely receiv ed divided b y the total n um ber
of pac k ets sen t is the p acket loss factor Note that w e are actually p erforming broadcast a sp ecial
case of m ulticast
Sim ulation Platform P arameters and Metho dology
W e ran the sim ulations on a Sun Ultra with Mb ytes of memory running Solaris T able summarizes the sim ulation parameters w e used The total sim ulation time w as seconds
A total of no des w ere randomly placed in a eld of a x meters Eac h no de generated
messages using a Mbitsec c hannel with p o w er range of meters The rationale for using
this p o w er range is explained in Section b elo w In our sim ulations f l ooding inter v al is set to

P arameter V alue Description
number of nodes total n umberofsim ulated no des
num pack ets n um b er of messages sentb y eachnode
f l ooding inter v al random w ait b efore o o ding in msecs
f iel d r ang e x sim ulation elds Xco ordinate
filed r ang e y sim ulation elds Yco ordinate
pow er rang e no des p o w er range
bandw idth no des bandwidth in bitssec
mobil ity pr obabil ity probabilityof node mo ving
mobil ity inter v al
distance unit
simul ation time sim ulation time in nanosecs
node pl acement RANDOM no de placemen t p olicy
pr opag ation f unc FREESP A CE propagation function
r adio ty pe RADIONOCAPTURE capture eect
mac pr otocol CSMA MACla y er proto col
netw or k pr otocol FLOODING net w ork la y er proto col
tr anspor t pr otocol UDP transp ort la y er proto col
T able Sim ulation P arameters
msecsF ree space propagation w as used to determine whether no des that are in the transmitters
range are able to receiv e data The proto col stac kw e used consisted of CSMA
o o ding and UDP
as the MA C net w ork and transp ort la y er proto cols resp ectiv ely GloMoSim has t w o mobilit y parameters the mobilit yin terv al measured in milli seconds and the
mobilit y probabilit y a real n um ber b et w een and Throughout the sim ulation a no de is considered
for mo v em entev ery mobil ity inter v al milliseconds Eac h time a no de is considered for mo v eme n t
it mo v es with a probabilityof mobil ity pr obabil ity T o maximize no de mo v eme n t w e set the
mobilit y probabilityto T o study the eect of mobilit yw ev ary b oth the mobilityin terv al and
the distance unit as describ ed in Section W e ran eac h sim ulation main taining all sim ulation parameters the same times eac h time with
a dieren t seed v alue Seeds v aried from to in steps of The graphs presen ted in
Section b elo w plot the a v erage and standard deviation across all runs
Results
Setting the P o w er Range
A critical factor inuencing pac k et loss in AHNs is the eectiv e transmission range of eachnode W e rst studied this parameter to get a b etter understanding of its eect and also to establish
optim um v alues whic h could b e used for our sim ulations Figure sho ws the results obtained
The sim ulations w ere p erformed for no des randomly placed in a x eld using GloMoSims recen t release also implemen ts the standard whic h is one of the reasons w e are curren tly p orting our sim ulations
to use GloMoSims new v ersion

50 100 150 200 250 300 350 400
50
55
60
65
70
75
80
85
Packet loss Vs Power−Range: Total Terrain range = 1000 x 1000 m
Power Range (meters)
Packet Loss %
Figure P ac k et Loss as a function of the P o w er Range
dieren tpo w er ranges from to meters The mobility interval for all the sim ulations w as set
to millisec onds Eac h no de transmitted messages at random times during the sim ulations
A t eachpo w er the test w as run times using dieren t random n um b er seeds F rom Figure it
can b e observ ed that the sw eet sp ot in the curv e is in the range from to meters This is
due to the fact that at lowpo w er ranges pac k et losses are high largely due to no de disconnectivit y In the other extreme ie for higher p o w er ranges losses are mostly due to collisions
Figure represen ts the pac k et loss factor as a function of the mobility interval for v e dieren t
po w er ranges These sim ulations are for no des randomly distributed in a x eld The
mobility interval for eachv alue of the p o w er range w as v aried from to milli seconds A teac h
mobility interval the test w as run times with dieren t random seeds as b efore W e observ ed
that at lo w er p o w er ranges the pac k et loss factor do es not v ary signican tly with the mobility
interval This can b e attributed to the fact that since the p o w er is lo w relativ e to the spacing
bet w een the no des individual no des can b e isolated and small groups can b e partitioned from the
remainder of the net w ork In these cases the pac k et loss is basically due to the ab o v e partitioning
eect and the loss due to the mobilit y itself is relativ ely lo w In some cases w e found that increased
mobilit y had the opp osite eect of reducing the pac k et loss factor since the increased mobilit y
caused the isolated no des to mo vein to receiving range
The Eects of Mobilit y
Our goal w as to study ho w no de mobilit y aects the p erformance of o o ding in wireless net w orks
More sp ecically w ew ere in terested in ev aluating the eects of mobilit y on o o dings abilit yto
deliv er pac k ets reliably W e use the pac k et loss factor as dened in Section to measure the
p ercen tage of pac k ets lost
Recall that o o ding w orks byha ving eachnode forw ard a pac k et it receiv es out on ev ery link except
the one from whic h it receiv ed the pac k et In wired net w orks o o ded pac k ets can b e lost due to
net w ork partitions caused b y link or no de failure that happ en b efore the o o ding w a v e and is

0 10 20 30 40 50 60 70 80 90 100 110
55
60
65
70
75
80
Plot of Percent Loss Vs Mobility for different Power Ranges
Mobility (msecs)
Percent Loss
Power Range 40m
Power Range 50m
Power Range 60m
Power Range 90m
Power Range 100m
Figure P ac k et Loss for dieren tv alues of P o w er Range
not mended in time In AHN en vironmen ts a no de ma y miss a pac k et if it mo v es out of range as
the o o ding w a vegoes b y Unless the no de subsequen tly migrates to a region of the net w ork where
the w a v e has not gone b y it will miss the pac k et
W e ran sim ulations for distanc eunits of and meters F or eac h distanc eunit v alue w ev aried
the mobilityin terv al from no mobilit y to msecs in steps of msecs Eac h exp erimentw as run
times with dieren trandom n um b er seeds and the pac k et loss factor w as then a v eraged o v er
the en tire set for a giv en mobility interval Figure sho ws o o dings pac k et loss factor as a function
of no de mobilit y Figure a sho ws that the pac k et loss factor do es not v ary considerably with lo w mobilit yThis can
b e attributed to the fact that since the no des only mo v e meter ev ery mobility interval they do
not mo v e sucien tly to go out of range and miss the o o ding w a v e
As w e increase no de sp eeds w e start noticing greater v ariation in pac k et loss as a function of
mobilit y Figure b sho ws the pac k et loss factor when no des mo v e meters ev ery mobilityin terv al
this corresp onds to kms v elo cit y While for no mobilit y the loss factor is for msec
mobility interval the pac k et loss factor is The no des high sp eeds cause them to mo v e out
of receiving range from an y of their neigh b ors resulting in higher pac k et loss It should b e noted
that pac k et loss is also inuenced b y no de densit y In sparse groups ev en little mobilit y can cause
no des to mo v e out of range resulting in m uc h higher loss v alues
P ac k et loss v alues for distanc eunit meters Figure c are not considerably dieren t than
the v alues in Figure b W e b eliev e that this is due to the fact that no des mo v e greater distances
during eac h mobilityin terv al Therefore most no des reac h the eld b oundary and b ounce bac k
to w ards the cen ter resulting in pac k et losses that are not signican tly higher than the case where
distanc eunit meters

0 500 1000 1500 2000
10
20
30
40
Packet loss Vs Mobility Interval : Mobility Unit = 1 meter
Mobility Interval (msec)
Packet Loss %
a distanc euni t meter
0 20 40 60 80 100
20
30
40
50
60
Packet loss Vs Mobility Interval :Mobility Unit = 5 meters
Mobility Interval (msec)
Packet Loss %
b distanc eunit meters
0 20 40 60 80 100
20
30
40
50
60
Packet loss Vs Mobility Interval :Mobility Unit = 10 meters
Mobility Interval (msec)
Packet Loss %
c distanc eunit meters
Figure P ac k et loss v ersus mobilit y Additional Observ ations
The o v erhead in v olv ed in o o ding is an imp ortan t criterion for ev aluating the p erformance of o o d ing as a reliable m ulticast proto col Since o o ding has to main tain minimal amoun t of state infor
mation its o v erhead in terms of state reten tion is lo w In o o ding the o v erhead is primarily due
to m ultiple copies of the same pac k et circulating in the net w ork Figure represen ts the n um ber
of duplicate pac k ets receiv ed byeac h no de as a function of the mobilit y The a v erage n um ber of
duplicate pac k ets receiv ed b y a no de is seen to decrease as the mobilit y increases In our sim u
lations with no des in a x eld eac h no de transmits messages whic h means that
eac h no de should ideally receiv e unique messages F rom the gure it can b e observ ed that
eac h no de receiv es duplicate messages for a mobilit yin terv al of msecs and this increases
to messages for a mobilit yin terv al of msecs Th us on an a v erage a no de receiv es ab out
duplicates of eac h transmitted pac k et when the mobilityin terv al is msecs and this increases
to pac k ets when the mobilityin terv al is msecs W eh yp othesize that this decrease in the
n um b er of duplicate pac k ets receiv ed b y a no de with increased mobilit y is due to the fact that the
no des receivea lo w er p ercen tage of the transmitted pac k ets as they mo v e faster
Related W ork
There are a n um b er of prop osed m ulticast proto cols orien ted to w ards AHN en vironmen ts These
can b e group ed in t w o main categories proactiv e proto cols that main tain routing state and reactiv e
proto cols that acquire routes on demand In this section w e briey describ e sev eral proto cols in
b oth categories

10 20 30 40 50 60 70 80 90 100
2720
2740
2760
2780
2800
2820
2840
2860
2880
2900
Number of Duplicate Packets Received by a Node Vs Mobility Interval : Mobility Unit = 5 meter
Mobility Interval (msec)
Number of Duplicate packets
Figure Duplicate pac k ets as a function of mobilit y Proactiv e Proto cols
The Ad Ho c Multicast Routing Proto col AMRoute builds a bidirectional shared tree to m ul
ticast data to mem ber nodes Comm unications b et w een these no des is la y ered on con v en tional
unicast messaging Coreno des are resp onsible for mem b er detection and group setup but unlik e
c or eb asedtr e es these core no des do not pass data and can migrate dynamically in resp onse to
mem b ership and connectivityc hanges AMroute relies on the underlying unicast proto col to handle
top ology c hanges with a tunneling mec hanism used for trac b et w een m ulticast islands tra v ers
ing regions where m ulticast is not deplo y ed This is adv an tageous in that in termediate routers need
not run anym ulticast proto col and o v erhead is conned to no des participating in the m ulticast
groups The p enalt y for this userlev el approac h is p o orer eciency in m ulticast pac k et replication
and in pac k et deliv ery dela y Although directed acyclic graphs tree structures are con v en tionally used for ecien tm ulticast
on wired net w orks the CoreAssisted Mesh Proto col CAMP prop oses multic ast meshes for
impro v ed robustness on mobile net w orks T op ology c hanges do not necessarily trigger m ulticast
reconguration Extending the idea of corebased trees CAMP builds on a traditional arc hitecture
for wired m ulticast but uses rev erse shortest paths the shortest path from receiv er to source
instead of a topdo wn tree This impro v es eciency o v er the shared m ulticast tree approac h
He artb e ats are used to main tain the rev erse shortest paths In con trast to F GMP CAMP
sp ecically a v oids the need to o o d the net w ork with either data or con trol messages whic hisseen
as unscalable
The MCED AR proto col is la y ered as a m ulticast extension on top of CoreExtraction Dis
tributed Ad Ho c Routing CED AR In CED AR a set of hosts are selected as the c or e of the
net w ork appro ximating a minimum dominating setEv ery host within this core is resp onsible for
route computation for its lo cal mobile hosts not in the core Longliv ed links are preferred when
building the core graph th us lo cal top ology c hanges infrequen tly cause global up dates Linkup
transitions are distributed slo wly but linkdo wn conditions are propagated quic klyA c orebr o adc ast
mec hanism is impleme n ted using reliable unicast messages that scales linearly with the n um ber of

net w ork no des CED AR emphasizes qualit yofservice QoS requiremen ts
LikeCAMPMCED AR uses the idea of m ulticast meshes adopting the notation of k connectivit y where k denotes the maxim um n um b er of logical connections at eachm ulticast mem b er Ho w ev er
m ulticast trac is distributed using only a spanning tree subgraph of the k connected mesh Th us
MCED AR claims the adv an tages of b oth m ulticast trees eciency and m ulticast meshes reliabilit y Join op erations tra v el through the core un til they reac h a core no de in the appropriate
m ulticast mesh and are accepted up to the k connectivityfactor Lea v e op erations up date the
m ulticast mesh appropriately and ma y b e delib erate or caused bynet w ork partition
The Ad Ho c Multicast Routing proto col utilizing Increasing idn um b erS AMRIS builds a
shared deliv ery tree for the m ulticast group It assigns an iden tication n um ber to eachm ulticast
member in the group the root of the tree has the lo w est n um b er in the tree with n um b ers increas
ing radially out w ards Reco v ery from brok en links is done lo cally within one p erio d of neigh bor
disco v ery with route reconstruction fa v ored o v er route disco v ery This is accomplished via an ex
panding ring searchb y the higher n um b ered no de Recall the higher n um b ered no de is further from
the ro ot so this is similar to a receiv eriniti ated join request Lik e CED AR longliv ed links are
preferred when building the tree to reduce later reconguration
OnDemand Proto cols
The Ad Ho c On Demand Distance V ector A OD V routing proto col prop oses b oth a unicast
and m ulticast mec hanism for data deliv eryIn con trast to the previous proactiv e approac hes the
ondemand trait means routes are only created for activ e trac and th us o v erhead is minim iz ed for
routing table space pro cessing and transmission bandwidth A ODVresem bles traditional linkstate
or distancev ector routing proto cols but mo died for the AHN en vironmen t to send few er messages
Route request messages are broadcast in an expanding ring to nd a path to a destination no de
route replies are unicast bac k T o guard against the coun ting to innit y con v ergence problem
A OD V stamps routing information with sequence n umbers later sequence n um b ers are alw a ys
preferred A OD V implem en ts neigh bor disco v ery using IP datagrams whic h cause o v erhead
The OnDemand Multicast Routing Proto col ODMRP also uses an ondemand reactiv e ap
proac h based on forwar ding gr oups Although not as formally structured as meshes or cores F GMP
uses a subset of the connected no des to b e resp onsible for forw arding m ulticast trac This is
referred to as sc op edo o ding Mem b ership in these forw arding groups require receiv erinitiated
con trol messages Lik eA OD V routes are constructed only on a demand basis Because of its lo w
o v erhead ODMRP targets high mobilit y net w orks and also large net w orks with sparse m ulticast
requiremen ts
Discussion
All the ab o vem ulticast proto cols main tain state some more than others in the form of a m ulticast
tree or mesh whic h requires o v erhead to construct and main tain Ev en the ondemand proto cols re
quire a no de to ha v e kno wledge of its m ulticast neigh b ors whic hma y b e instan taneously accurate
but so on stale In scenarios in v olving extremel y high mobilit y shortcomings b ecome eviden t

b ecause no de mobilit y causes con v en tional m ulticast trees to rapidly b ecome outdated F requen t
state c hanges require constan t routing up dates p ossibly nev er con v erging to accurately p ortraythe
curren t top ology Eac h proto col optimizes dieren t considerations in situations with high mobilit y
w e prop ose disp ensing en tirely with the structure of the m ulticast hierarc h y and using a o o ding
strategy Summary and F uture W ork
In closing this pap er discussed a n um b er of features and c haracteristics of highly mobile fastmo ving
AHNs and argued that high sp eed and frequen t mobilityin terv als necessitate radical approac hes
for reliable m ulticast In particular o o ding is prop osed as a stateless and top ologyindep enden t
mec hanism for reliable m ulticast
The sim ulation results illustrate that o o ding is quite eectiv e Ho w ev er when mobilit yin terv als
are small and no de sp eed is sucien tly high ev en o o ding b ecomes unreliable This is a surprising
outcome whichmotiv ates further in v estigation of more robust and more p ersistentv ariations of
o o ding suitable to suc hv olatile and dynamic AHNs
Sp ecically the follo wing are some of the issues to b e addressed
More robust forms of o o ding w ould lik ely incur some p er pac k et state in net w ork no des
Although not as longterm in nature as the state k ept in treebased m ulticast mec hanisms the
amoun t of state m ust b e k ept to a minim um Extensiv e exp erimen ts and heuristic adjustmen ts
are needed to shed some ligh t on this issue
The o v erhead of o o ding and its v ariations needs to b e carefully measured W e note that the
o v erhead of o o ding in xed net w orks is comparativ ely trivial to measure a pac k et tra v erses
a giv en net w ork link at most once In AHNs the o v erhead is lik ely to b e more elusiveto
measure and quan tify since high mobilit y can result in m ultiple receptions of the same pac k et
and o o ding w a v es w ould tak e longer to complete due to mobilit yinduced pac k et loss

References
R Bagro dia and R Mey err P ARSEC A parallel sim ulation en vironmen t for complex system T ec hnical rep ort UCLA
A Ballardie P F rancis and J Cro w croft Core based trees CBT An arc hitecture for scalable in terdomain m ulticast
routing In Pr o c of A CM SIGCOMM E Bommaiah M Liu A McAuley and R T alpade AMRoute Adho c m ulticast routing proto col IETF manet draft
talpademanetamroutetxt C Chiang and M Gerla Ondemand m ulticast in mobile wireless net w orks In Pr o c IEEE ICNP
C Chiang M Gerla and L Zhang Shared tree wireless net w ork m ulticast In IEEE Internationa l Confer enceon
Computer Communic ations and Networks ICCCN Septem b er PIM dev elopmen t group Proto col indep enden tm ulticast Av ailable from h ttpnet w ebuscedupim No v em b er J GarciaLunaAcev es and E Madruga A m ulticast routing proto col for adho c net w orks In Pr o c of INF OCOM M Gerla and SJ Lee Ondemand m ulticast routing proto col for mobile adho c net w orks Av ailable from
h ttpwwwcsuclaeduNRLwireless IEEE Std D Johnson and D Maltz Dynamic source routing in ad ho c wireless net w orks In T omasz Imielinski and Hank Korth
editors Mobile Computingc hapter Klu w er Academic Publishers C P erkins and E Ro y er Ad ho c on demand distance v ector A OD V routing IETF manet draftietfmanetao dvtxt
CMU Monarc h Pro ject Mobilit y extensions to ns Av ailable from h ttpwwwmonarc hcscm uedu S Ramanathan and M Steenstrup A surv ey of routing tec hniques for mobile comm unications net w orks A CMBaltzer
MONET pages Octob er P Sinha RSiv akumar and V Bhargha v an CED AR Core extraction distributed ad ho c routing In Pr o c of IEEE
INF OCOMM P Sinha R Siv akumar and V Bhargha v an MCED AR Multicast core extraction distributed adho c routing In Pr o c
of the Wir eless Communic ations and Networking Confer enc e C W u Y T a y and C T oh Ad ho c Multicast Routing proto col utilizin g Increasing idn um b erS AMRIS IETF manet
draftietfmanetamrissp ectxt X Zeng R Bagro dia and M Gerla GloMoSim a library for parallel sim ulation of largescale wireless net w orks In
Pr o c of P ADS

Asset Metadata

Creator Ho, Christopher (author), Obraczka, Katia (author), Tsudik, Gene (author), Viswanath, Kumar (author)

Core Title USC Computer Science Technical Reports, no. 707 (1999)

Alternative Title Flooding for reliable multicast in multi-hop ad hoc networks (title)

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)

Tag OAI-PMH Harvest

Format 14 pages (extent), technical reports (aat)

Language English

Unique identifier UC16271044

Identifier 99-707 Flooding for Reliable Multicast in Multi-Hop Ad Hoc Networks (filename)

Legacy Identifier usc-cstr-99-707

Format 14 pages (extent),technical reports (aat)

Rights Department of Computer Science (University of Southern California) and the author(s).

Internet Media Type application/pdf

Source 20180426-rozan-cstechreports-shoaf (batch), Computer Science Technical Report Archive (collection), University of Southern California. Department of Computer Science. Technical Reports (series)

Access Conditions The author(s) retain rights to their work according to U.S. copyright law. Electronic access is being provided by the USC Libraries, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Repository Email csdept@usc.edu

Inherited Values

Title Computer Science Technical Report Archive

Description Archive of computer science technical reports published by the USC Department of Computer Science from 1991 - 2017.

Coverage Temporal 1991/2017

Repository Email csdept@usc.edu

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)

Linked assets

Computer Science Technical Report Archive