Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
Computer Science Technical Report Archive
/
USC Computer Science Technical Reports, no. 655 (1997)
(USC DC Other)
USC Computer Science Technical Reports, no. 655 (1997)
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Alternate P ath Routing and Pinning for In terdomain Multicast
Routing
USC Computer Scienc eT e chnic al R ep ort Daniel Zappala
danielisiedu
h ttpnet w ebuscedudaniel
Univ ersit y of Southern California
Information Sciences Institute
Admiralt yW a y Flo or
Marina del Rey CA Deb orah Estrin
y
estrinuscedu
h ttpnet w ebusceduestrin
Computer Science Departmentand
Information Sciences Institute
Univ ersit y of Southern California
Los Angeles CA Scott Shenk er
z
shenk erparcxero xcom
XeroxP alo Alto ResearchCen ter
Co y ote Hill Road
P alo Alto CA Abstract
Man y researc hers ha v e explored enhancemen ts of the In ternets b esteort service mo del
that allo w realtime and other inelastic applications to obtain preferen tial Qualit y of Service
Ho w ev er these applications are limited to utilizing the opp ortunistic shortestpath routes pro
vided b y the curren t routing infrastructure T o b etter supp ort realtime applications this pap er
in tro duces extensions to in terdomain m ulticast routing to scalably compute and install alternate
paths and nonopp ortunistic or pinned routes W e presen t a simple m ulticast setup proto col
for installing alternate paths and discuss ho w it prev en ts lo ops F urthermore w e include the
results of a sim ulation study to demonstrate the viabilit y of using lo calized route construction
to nd adequate alternate paths
In tro duction
The In ternet has b een extremely successful supp orting elastic applications with b esteort service
Cla Ho w ev er b est eort service can result in large and widely v arying endtoend pac k et
dela ys if the links and routers tra v ersed are hea vily loaded T o b etter supp ort realtime and other
inelastic applications for whic h suchv agaries are detrimen tal the researc h comm unit y has prop osed
This material is based up on w ork supp orted b y the Air F orce Oce of Scien tic Researc h under Aw ard No
D AAHG
y
This material is based up on w ork supp orted b y the National Science F oundation under Gran t No NCR
z
This researchw as supp orted in part bythe Adv anced Researc h Pro jects Agency monitored b yF ort Huac h uca
under con tracts D ABTC The views expressed here do not reect the p osition or p olicy of the US
go v ernmen t
extensions to the In ternets service mo del and arc hitecture These extensions w ould allowo ws or
o w aggregates to obtain preferen tial qualities of service QoS b y marking pac k ets or b y using a
resource reserv ation proto col
While m uc h researc h has b een dev oted to the dev elopmen t of admission con trol sc heduling and
queueing mec hanisms relativ ely little atten tion has b een paid to upgrading the routing infrastruc
ture of the In ternet to supp ort these extensions Previous w ork Bre EZL
addressed some
asp ects of the problem for unicast routing This pap er is concerned with addressing the additional
complexities that arise when considering m ulticast routing supp ort for in tegrated services net w orks
F urthermore w e fo cus on in terdomain routing for whic h issues of scaling are more acute than for
the in tradomain case
The routing infrastructure of the In ternet has b een designed primarily to supp ort b esteort
service Curren t routing proto cols Hed Mil Hed IDR RL Mo yb use opp ortunistic
shortestpath routing for all applications By opp ortunistic w e mean that routing alw a ys utilizes
the curren t shortest path ev en if the previous shortest path is still functioning By shortest p ath
w e mean that routing uses a single cost metric often just hopcoun t and then c ho oses the
leastcost path
In this pap er w e fo cus on t w o particular problems that face realtime or other p erformance
sensitiv e applications that op erate o v er curren t routing proto cols
F aile d primary p ath pr oblem Because curren t routing proto cols use a single path an appli
cation has no alternativ e route to try if it do es not receiv e acceptable service along this path
This ma y happ en for example if a o wdoes not ac hiev e acceptable dela y along an unreserv ed
path or if it attempts to mak e a resource reserv ation along the shortest path and is denied
Th us man yo ws ma y b e denied service ev en though other paths could accommo date their
service requiremen ts
Opp ortunistic r outing pr oblem When curren t routing proto cols adapt to a new shortest path
an application ma y exp erience a service disruption F or example if a o w has obtained a
go o d route and then the route c hanges p ortions of the new route ma ynot ha v e the necessary
capacit y If the o w is unable to completely reestablish its desired service on the new path
service will b e unnecessarily disrupted
Man y in the researchcomm unityha v e prop osed QoS r outing tec hniques to solv e these prob
lems A t ypical QoS routing sc heme globally distributes top ology link resource a v ailabilit ygroup
mem b ership and in some cases p ero w resource usage A sources rsthop router then uses
this information to compute a m ulticast tree that is kno wn a priori to ha vea v ailable resources
Finally this same router uses a sourceinitiated setup proto col to install the m ulticast tree in the
net w ork The bulk of this w ork attempts to minimize tree cost primarily for static m ulticast groups
BKJ W ax Cho KPP The In ternet and A TM comm unities ha v e b egun applying these
results to linkstate m ulticast routing proto cols R GW ZSSC PNN
Because these QoS routing approac hes require global distribution and sync hronization of suc h
rapidly v arying quan tities w e do not b eliev e they are applicable to in terdomain routing where
issues of scale are paramoun t
A global database of top ology alone scales linearly with the size
of the net w ork neither group mem b ership nor the n um ber of o ws is limited b y the size of the
net w ork Additionally the o v erhead required to main tain a consisten t view of this data dep ends
on application b eha vior suc h as group mem b ership c hanges and resource usage W e b eliev e this
QoS routing ma y still b e used within a domain
com bination of storage and pro cessing o v erhead rules out the use of cen tralized computation and
installation of routes as w ell as global optimization of these routes
One alternativ e to QoS routing that do es scale adequately for in terdomain routing is to compute
m ultiple paths for eac h destination based on relativ ely static routing metrics In this approac h
called QoR r outing the metrics reect the Qualit y of Route using static service c haracteristics
suc h as maximal bandwidth or minimal latency to indicate link capabilities The routing proto col
main tains a separate routing table for eac h metric and applications indicate the their desired QoR
when sending data This is similar to the congestionsensitiv et yp eofservice routing describ ed
in MS but routes adapt only to top ology c hanges not resource usage QoR routing could
pro vide signican t b enets for b esteort service b y allo wing for instance in teractiv e applications
to a v oid routes in v olving satellite links while enabling applications in v olving async hronous bulk
data transfers to seek out maximal bandwidth paths F or similar reasons QoR routing could b enet
realtime and other inelastic applications
Because QoR routing metrics are static this approac h has none of the scaling problems of the
more dynamic QoS routing prop osals Ho w ev er the essence of the failed primary route problem
remains an application could only a v ail itself of one minimallatency route and one maximal
bandwidth route etc If a service requiremen tw as denied along one of these precomputed routes
there w ould b e no w a y of utilizing a v ailable bandwidth along other routes with similar prop erties
Th us while QoR routing can increase the c hance that an application will b e satised with the
primary route receiv ers need routing to install alternate p aths in to a m ulticast tree on demand
Lik ewise QoR routing do es not solv e the opp ortunistic routing problem since QoR routes adapt
as the metrics c hange Therefore receiv ers need routing to also install pinnedr outes routes that
will not adapt unless they fail in to a m ulticast tree on demand
In this pap er w ein tro duce a routing arc hitecture in whic h alternate paths and pinned routes
are installed bya m ulticast route setup mec hanism Existing route setup mec hanisms use sender
orien ted route setup or require routers to kno w the iden tities of do wnstream group mem b ers DB FBZ UNI whic h renders them un usable for in terdomain routing W e describ e a simple
scalable route setup mec hanism named MORF and sho w that it prev en ts lo ops while establishing
and rerouting a m ulticast tree
The k ey to the viabilit y of this arc hitecture is whether routers can nd adequate alternate paths
T o scale to the in terdomain lev el w e prop ose to use route construction that is b oth decen tralized
and querydriv en Routers with lo cal receiv ers nd alternate paths for their receiv ers ondemand
Moreo v er these routers do not use global distribution of top ology to nd routes Rather they nd
routes using a partial map of the net w ork whic h they build using heuristics to query the routing
tables of other no des In this pap er w e presen t the results of a sim ulation study demonstrating the
viabilit y of using lo calized route construction to nd alternate paths around b ottlenec ks Our in ten t
in presen ting these results is to demonstrate the utilityof sev eral lo wcost pro ofofconcept route
construction heuristics th us v alidating our arc hitecture As others nd b etter route construction
heuristics routers ma y incremen tally deplo y the impro v emen ts
In our approac h to route construction w e are designing for the case when congestion is not
widespread and th us do not attempt to nd paths where resource a v ailabilit y is kno wn a priori
W e view this as the most feasible approachfor in terdomain routing for sev eral reasons During
times of high load when congestion is common using alternate paths can degrade net w ork utilization
KZ Aki Moreo v er it is impracticable to design a scalable route computation metho d for
nding the pro v erbial needle in a ha ystac k the one route that has a v ailable resources among a
v ery large n um b er of routes that do not
Router
Multicast
Routing
Unicast
Routing
Local Route
Construction
Route Setup
Protocol
Routing
Reservation
Protocol
Host
Reservation
Protocol
Application
QoS Mgr
Figure Routing Arc hitecture
Th us our solution to the failed primary route and opp ortunistic routing problems consists
of t w o enhancemen ts to the routing infrastructure a m ulticast route setup proto col and lo calized route construction W e b egin the rest of this pap er b y rst discussing our m ulticast routing
arc hitecture in Section sho wing ho w routing ma y install alternate paths and pinned routes on
b ehalf of lo cal receiv ers Then in Section w e describ e MORF our m ulticast route setup proto col
and analyze its lo opfreedom In Section w e describ e in terdomain route construction heuristics
weha v e dev elop ed and presentsim ulations sho wing their eectiv eness in nding alternate paths
Section discusses related w ork and Section presen ts our conclusions
Multicast Routing Arc hitecture
Ov erview
Our m ulticast routing arc hitecture utilizes four primary comp onen ts to build m ulticast trees Fig
ure The m ulticast routing proto col constructs m ulticast trees based on routes obtained from
the unicast routing proto col
The route setup proto col installs alternate paths and pinned routes
It op erates separately from the unicast and m ulticast routing proto cols but in teracts with them
to o v erride their computations Routers at the endp oin ts of the net w ork ie near hosts ma y
also incorp orate a lo cal route construction agen t whic h nds alternate paths for the route setup
proto col
Figure also sho ws ho w applications and the reserv ation proto col in teract with the routing
arc hitecture Applications access route setup and reserv ation setup through t w o separate in terfaces
in con trast to man y QoS routing prop osals where there is a single in terface Note that there is
no application in terface to the lo cal route construction agen t the applications rsthop router
con tacts the agen t when it needs a route By not allo wing the application to determine the routes
b eing used w e prev en t a malicious malfunctioning or misguided user from pro viding its o wn route
and undermining the in tegrit y or eciency of a giv en tree
In the most simplistic mo del of ho w this arc hitecture w ould b e used applications w ould in terface
directly with routing and the reserv ation proto col resp ectiv ely In fact giv en the complexit y of the
service options a v ailable w e assume that man y op erating systems will oer some form of supp ort
D VMRP WPD uses its o wn in ternal unicast routing proto col but this is equiv alen t for the purp oses of our
arc hitecture
W e are indebted to our colleagues Stev e Deering and V an Jacobson for emphasizing that end systems should not
con trol routing
Host
Application
QoS Mgr
Route Pinning or
Alternate Path Request
Router
Local Route
Construction
Route Setup
Protocol
Route Calculation
Request
Route
Figure Example Use of Route Arc hitecture
for managing qualit y of services Sucha QoS manager could act as an agen t on b ehalf of an
application managing the reserv ation establishmen t and the routing services pro curemen t pro cess
A user or a monitoring program ma y simply indicate that service is unacceptable The QoS
manager could then c ho ose from a n um b er of actions includin g asking for one of the a v ailable
routing services dep ending on the application Th us the QoS managemen t function could reside
either in the application itself or in a QoS manager or in some other form of op erating system
supp ort All of these p ossibiliti es t within our prop osed arc hitecture w e are dening the services
a v ailable to an endhost not the organization of soft w are on an endhost
T o initiate alternate path setup an application or a QoS manager prompts routing for a dieren t
route Figure This signal ma y b e prompted byman y factors including a users unhappiness
with pac k et dela ys or an admission con trol failure along the shortest path The rsthop router
then con tacts the lo cal route construction agen t whic h returns an explicit route
that meets the
receiv ers criteria The setup proto col installs and pins this route reconguring the m ulticast tree
at eac h hop
T o pin an existing opp ortunistic route the application or QoS manager prompts routing to pin
the curren t route Figure This signal ma y b e prompted byan y indication that the application
is satised with the curren troute andw an ts to minimize its c hances of disruption for example if
pac k et loss is lo w or it has succeeded in making a reserv ation along the route The rsthop router
prob es the m ulticast tree to determine the curren t route and enco des this route as an explicit
route The setup proto col then installs and pins this route using the same mec hanism as for an
alternate path
Note that the extra trip for probing the route allo ws routing to prev en t lo ops
during pinning b y using an explicit route see Section Multicast Route Setup Proto col
An um b er of reserv ation proto cols p erform route setup at the same time as installing a reserv ation
FBZ DB Sti Ho w ev er applications not using reserv ations mayw an t to utilize route setup
F or example use of video and v oiceconferencing o v er the m ulticast bac kb one is widespread to da y
CD but eac h receiv er is limited to using the shortest path When congestion o ccurs on this
path a receiv er mayw an t to use route setup to install an alternate path y et still use b esteort
service o v er that path if it yields adequate p erformance In this con text applications could use a
reserv ation proto col as y et another separate enhancemen t of this service mo del This requires that
W e could use the term source route but the route lists hops from the receiv er to sender and is installed bya
router near the receiv er
In fact a pinned route is really an alternate path in the sense that it is an alternativ e to the opp ortunistic route
already in place
wenot em b ed route setup in the reserv ation proto col itself but rather incorp orate it in to the basic
routing infrastructure
The function of the m ulticast route setup proto col then is to install explicit routes on b ehalf
of receiv ers o v erriding the opp ortunistic routes used bya m ulticast tree An y route it installs is
pinned so that it remains in place un til it fails at whic h time the m ulticast tree migrates backto
an opp ortunistic route
The setup proto col represen ts all routes as strict explicit routes listing all hops in order from
the receiv er to the sender This restriction enables the setup proto col to more easily guaran tee the
lo opfreedom of the m ulticast tree see Section Lo cal Route Construction Agen t
One of the k ey c hallenges of this m ulticast routing arc hitecture is to dev elop a route construction
proto col that ma y b e applied to the in terdomain scale W e distribute route computation to routers
lo cated near receiv ers and do not attempt to nd routes based on link resource a v ailabilit y Instead
routers rst use the shortest path and then as needed nd alternate paths that a v oid an y b ottle
nec ks This design follo ws the unied routing mo del ERH in whic h commonlyused routes are
precomputed and routes used less frequen tly are computed ondemand
Using lo calized route construction reduces the problem of constructing m ulticast routes to that
of constructing unicast routes Eac h route construction agen t only needs to nd a route b et w een
a lo cal receiv er and the sender rather than ha ving to takein to accoun t the en tire m ulticast tree
This approac h scales w ell to large m ulticast groups and allo ws agen ts to use existing unicast routing
proto cols as the basis for a route construction algorithm
In addition using lo calized route construction has sev eral other adv an tages First b ecause a
route construction agen t do es not need to co ordinate its routing decisions with other agen ts it can
utilize information that is not captured b y the routing proto cols static metrics This information
can include the status of lo cal reserv ation requests or resource a v ailabilit y information that is only
kno wn lo cally In addition route construction agen ts do not need to globally agree on a metric or
algorithm This allo ws div ersit y in the ev olution of route construction tec hniques
One consequence of using lo calized route construction is that routers mayc ho ose conicting
routes Because an ynode in a m ulticast tree m ust ha v e a single paren t routes using conicting
paren ts m ust b e resolv ed In the m ulticast routing arc hitecture the route setup proto col resolv es
suc h conicts as it installs eac h route Section discusses this pro cedure in more detail
The MORF Multicast Route Setup Proto col
Weha v e designed the MORF m ulticast route setup proto col to install routes pro vided bylocal
route construction agen ts A router initiates installation of an explicit route b y generating a
Setup message con taining the route T able MORF forw ards the Setup message along the route
creating a Setup T ree that it main tains separately from the shortestpath tree built b y the m ulticast
routing proto col Where the Setup T ree conicts with the shortestpath tree MORF o v errides the
shortestpath tree and the m ulticast routing proto col prunes the conicting branc hes Figure a
MORF also adjusts forw arding table en tries so that the resulting m ulticast tree reects the path
installed b y MORF Figure b The m ulticast tree ma y b e for a single sender DEF
or
m ultiple senders ma y rendezv ous via a core DEF
BF C In either case the proto col is the
same in the follo wing discussion w e refer to senderbased trees for simplicit y
T able MORF Proto col Messages
Messages P arameters
Setup Gr oup T ar get R oute Gr oup m ulticast group
F ailure Gr oup T ar get SetupR t T r e eR t T ar get sender or core
T eardo wn Gr oup T ar get R oute explicit route
SetupR t route from Setup
T r e eR t route used bytree
Prune
Setup
a) Setup builds Setup Tree, prunes
multicast tree
b) re-configured Setup tree and
multicast tree
multicast tree
Setup Tree
Figure Using a Setup Message to Install a Route
Since the Setup T ree o v errides default opp ortunistic routing eac h router in the Setup T ree m ust
ha v e a mec hanism to detect failures of an alternate path or a pinned route The setup proto col
ma y rely on a unicast routing proto col to exc hange query messages with its neigh b ors to determine
whether they are aliv e or it ma y use its o wn similar mec hanism Once a failure is detected MORF
sends a T eardo wn message b oth upstream and do wnstream of the failure to remo v e failed branc hes
from the Setup T ree Figure a A teac h hop MORF noties the m ulticast routing proto col of
the branchesit isremo ving The m ulticast routing proto col rebuilds the m ulticast tree to reect
its metric often the shortest path to the sender Figure b
The ab o v e examples represen t the simplied case when a Setup do es not conict with the rest
of the Setup T ree Ho w ev er the setup proto col m ust also resolv e Setup messages from dieren t
lea v es that use conicting routes b ecause leaf routers ma y use indep enden t route construction
agen ts MORF resolv es conicts b yc ho osing the rst route that is installed for an ygiv en branc h
of the tree Where subsequen t routes meet this branc h they m ust conform to the route used from
that p ointup w ard to w ard the source If the setup proto col do es not follo w this restriction then a
n um b er of lo oping scenarios ma y arise Section discusses these cases and the manner in whic h
they are prev en ted
Figure sho ws an example of ho w all Setup messages except the rst one m ust matc h the route
already used b y the Setup T ree When a Setup message adds a no de to the Setup T ree it cac hes
the route it will use to tra v el from that no de up w ard to w ard the sender If a subsequen t Setup
message arriv es at that no de it compares the remaining route it m ust tra v el to the route cac hed
lo cally If the routes do not matc h the no de stops pro cessing the Setup and sends a F ailure message
do wnstream Figure a The F ailure message con tains the route used b y the failed Setup and the
a) Teardowns remove AP-tree branches
after failure, Join re-builds multicast tree
b) re-configured Setup Tree and
multicast tree
multicast tree
Setup Tree
Teardown
Teardown
Join
Figure Using a T eardo wn to Remo veaF ailed Route
a) Setup does not match, triggers
Failure
b) Setup matches
Setup(<1,2,3,4,6,S>)
Failure(<1,2,3,4,6,S>,<4,5,6,S>)
Setup(<1,2,3,4,5,6,S>)
S
1
3 2
54
6
S
1
3 2
54
6
<4,5,6,S> <4,5,6,S>
multicast tree
Setup Tree
Figure Matc hing Setup Messages
route used b y the tree from the rejecting no de up w ard T able A router receiving a F ailure
message merges the t w o routes it con tains to construct a new route that will matc h the tree then
sends a second Setup with this route Figure b
It is from this mec hanism Match or F ail that MORF deriv es its name By using this
restriction MORF ma y increase the setup latency but it prev en ts an y lo ops from forming while
the tree is constructed The remainder of this section discusses p oten tial lo oping scenarios and
analyzes the tradeos of MORF v ersus other p oten tial solutions for prev en ting lo ops
Prev en ting Lo ops
When Setup messages are not restricted to matc hing the rest of the Setup T ree a n um b er of p ossible
lo oping scenarios arise Figure a sho ws t w o Setups eac h using an explicit route Based on their
order of arriv al as sho wn if the Setups merge they form a lo op This lo op can b e prev en ted if
no des and compare the routes and detect the lo op will form Ho w ev er when three joins are
in v olv ed as in Figure b a single no de cannot prev en t the lo op from forming without ha ving more
information a v ailable
T o prev en t lo ops a no de can use one of t w o strategies
Before adding a paren t the no de c hec ks all its descendan ts to b e sure the paren t is not already
a descendan t
5
2
3 1
S
4 6
a) Loop formed by two Setups b) Loop formed by three Setups
Setup #1 <4,1,2,S>
reaches 1 first
Setup #2 <5,2,3,S.>
reaches 2 first
Setup #3 <6,3,1,S>
reaches 3 first
Sender:
Loop:
S
1-2-3-1
Setup #1 <4,1,2,3,S>
reaches 1 first
Setup #2 <6,3,1,S>
reaches 3 first
Sender:
Loop:
S
1-2-3-1
5
2
3 1
S
4 6
Figure Lo ops F ormed b y Setup Messages
a) Setup triggers Merge sent upstream b) Setup triggers Merge sent downstream
multicast tree
Setup Tree
Setup(<1,2,3,4,6,S>) Setup(<1,2,3,4,6,S>)
S
1
3 2
54
6
S
1
3 2
54
6
Merge(<1,2,3,4,6,S>)
Merge(<1,2,3,4,5,6,S>)
Figure Merging Setup Messages Instead of Matc hing
Before adding a c hild the no de c hec ks all its ancestors to b e sure the new c hild is not already
an ancestor
W e discuss eac h of these in turn Due to the dynamic nature of m ulticast trees a no de ma y
not kno w all of its ancestors or descendan ts While a no de kno ws the route em b edded in the
Setup message it has sen t upstream that message ma yha v e merged with another Setup carrying
a dieren t route Lik ewise other Setups ma yha vemergeddo wnstream adding new descendan ts
One approachtok eep no des up dated concerning upstream and do wnstream merges is to dis
tribute information after eachmerge F ollo wing solution ab o v e eac h Setup that merges can
send a Merge message upstream con taining its route Figure a Ev ery no de can then knowall
its descendan ts and thereb y detect an y lo ops Alternativ elyin k eeping with solution ab o v e
eac h Setup that merges can send a Merge message do wnstream con taining the upstream p ortion
of the route it merged with Figure b This allo ws ev ery no de to detect lo ops b y kno wing all its
ancestors W e denote these t womec hanisms as Mer ge Up and Mer ge Down resp ectiv ely In b oth
of these approac hes information distributed b y the Merge messages ma y b e stale so lo ops suc has
that sho wn in Figure ma y still form temp orarily b efore b eing brok en
As opp osed to these solutions whic h in some cases will only detect lo ops after they ha v e b een
formed the strategy w e use in MORF prev en ts an y lo ops from forming By requiring eac h Setup
to matc h the upstream route already in place on the tree MORF in eect enforces solution b y
requiring that eac h no de kno w its ancestors b efore it is added to the tree Once a no de is added to
the m ulticast tree its ancestors do not c hange The cost of this strategy is that eac h Setup ma y
T able Comparison of Setup Mec hanisms
Mec hanism Message Storage Setup Lo op
Name Ov erhead Ov erhead Latency Handling
MORF O Oa or trips Prev en t
Merge Do wn O Oa trip DetectBreak
Merge Up Od Od trip DetectBreak
tak e an extra roundtrip b et w een itself and the rest of the tree
T able compares MORF to the Merge Do wn and Merge Up mec hanisms when building a single
m ulticast tree assuming there is no pac k et loss and that one receiv er joins the tree at a time The
columns listing message and storage o v erhead consider the b eha vior of eachmec hanism at a single
no de Ov erhead in these cases is expressed in terms of a the n um b er of ancestors of a no de or d the n um b er of descendan ts of a no de The setup latency column lists time in terms of the n um ber
of trips tak en b et w een a receiv er and the m ulticast tree
Clearly the Merge Up mec hanism do es not scale w ell b ecause eachnode m ust store eachde scendentas w ell as send one message upstream for eac h descendan t The adv an tages of using a
receiv erorien ted mec hanism are lost with Merge Up a senderorien ted mec hanism has the same
message o v erhead but only the sender m ust store the descendan ts
The MORF and Merge Do wn mec hanisms ha v e a similar o v erhead in this situation The MORF
mec hanism mayha v e a longer setup latency but in return has the distinct adv an tage that it ma y
prev en t rather than just detect lo ops as discussed ab o v e
Unicast Route Setup
Previous w ork has studied the ecacy of using explicit routing to supp ort unicast realtime ap
plications Bre One w a y to use explicit routes to pro vide alternate paths or pinned routes
is to em b ed the explicit route in an applications pac k ets EZL
HLFT DH Assuming
the route will b e inserted b y the senders nearest router no mo dications to applications will b e
needed Ho w ev er b ecause man y routers curren tly dela y pro cessing of explicitly routed pac k ets
this mec hanism ma y not b e applicable to applications with strict dela y requiremen ts
An alternativ e is for the senders nearest router to insert a lab el in the applications pac k ets
rather than an explicit route This lab el w ould reference an alternate path or pinned route that is
installed using MORF
Because unicast applications in v olv e only one receiv er the setup latency
will only b e trip
In terdomain Route Construction Heuristics
Giv en a scalable in terdomain route setup proto col the imp ortan t issue wem ust address is whether
w e can construct useful alternate paths at a reasonable cost Due to the scaling problems inheren t
in distributing global top ology information at the in terdomain lev el of the net w ork w epropose
The lab el could in fact b e a m ulticast group address reducing unicast alternate path routing to a sp ecial case of
m ulticast
Backbone
Sender
Receivers
Route
Construction Agent
Backbone
Sender
a) Route construction agents find alternate
paths around bottlenecks for local receivers
b) Agent near receiver queries agent near sender
when necessary
Figure Route Construction Agen ts Serv e Lo cal Receiv ers
that a route construction agen t use heuristics to partially explore the in terdomain top ology and
nd routes
Weha v e dev elop ed sev eral lo wcost pro ofofconcept heuristics that do not require c hanges to
routing proto cols th us allo wing incremen tal deplo ymen t at the edges of the net w ork T o determine
their eectiv eness in nding alternate paths w eha v e conducted a sim ulation study o v er v arious
t yp es of random top ologies F or the purp oses of our sim ulations w eha v e concen trated on v alidating
our approachb y trying to nd routes around a single o v erloaded in terdomainlev el link
Approac h
Our approac h relies on route construction agen ts to serv e lo cal receiv ers Figure a When a re
ceiv er needs an alternate path its lo cal agen t uses heuristics to nd a route around an y b ottlenec ks
If the lo cal agen t is unable to nd an alternate path it ma ycon tact an agen t near the sender for
a route Figure b
Because the agen ts do not ha v e full top ology information they m ust build a partial map of the
top ology to nd alternate paths W eha v e fo cused on metho ds for gathering top ological information
that is a v ailable with existing routing proto cols Weha v e dev elop ed the follo wing algorithm for
exploring paths
An agen t explores the curren t path from itself to a small set of Initial No des in the net w ork
These no des ma y b e randomly c hosen or ma y consist of all the no des within n hops The
agen t initializes its map with these paths
When the route setup proto col requests an alternate path from the agen t the request iden ties
am ulticast tree a group and a sender or core and a b ottlenec k link The agen t prob es the
m ulticast tree for the requesting receiv ers curren t path to the sender adds this information
to its map and marks the b ottlenec k link
The agen t then runs a Dijkstra computation o v er its curren t map to nd an alternate path
around the b ottlenec k link If one is found it returns this path to the route setup proto col
If a path is not found the agen t augmen ts its map b y exploring the curren t path from some
other no de in its map a thirdpart y to the sender The thirdpart yisc hosen using a
T able Ov erhead and P ath Length Bounds for Route Construction Heuristics
Heuristic Ov erhead Bound
NHop c c
N NHopSender c c N MRandom M
MRandomSender M
NHopMRandom M c c N NHopMRandomSender M c c
N breadthrst searc h of the map starting from the agen t and nding the rst Initial No de not
already used as a thirdpart y for the sender in question Once a thirdpart y is found that
adds new links to the map the agen t returns to step to rerun the Dijkstra computation If
no suc h no de is found then no alternate path can b e found lo callyA t this p oin t the agen t
ma y optionally con tact an agen t that serv es the sender of the m ulticast tree
The v ariable parameters in this algorithm are the metho d b y whic h no des are selected to
initialize the agen ts map either random or within n hops and the option to query another
agen t near the sender of the tree when a route cannot b e found lo cally Com bining the v ariations
of these parameters yields the follo wing set of heuristics
NHop Initialize using all no des within N hops
NHopSender Same as ab o v e but query the senders agen t if unable to nd a route
lo cally MR andom Initialize using M random no des
MR andomSender Same as ab o v e but query the senders agen t if unable to nd a route
lo cally NHopMR andom Initialize using all no des within N hops and M random no des not
including an y no des within N hops
NHopMR andomSender Same as ab o v e but query the senders agen t if unable to nd a route
lo cally W e can b ound the o v erhead of eac h of the ab o veh ueristics in terms of the n um b er of thirdpart y
queries p erformed for a single alternate path searc h T able lists these b ounds in terms of c the
maxim um degree of connectivit y of the net w ork N the n um b er of hops explored initially and M
the n um b er of random no des explored initially In the algorithm giv en ab o v e thirdpart y queries
are limited to the set of Initial No desth us o v erhead is b ounded b y the size of this set In the
case of the NHop heuristic this b ound is c c
N deriving the b ounds for the other heuristics
is straigh tforw ard By k eeping N small ie or hops w e can limit the o v erhead of all of the
heuristics to a small n um b er of thirdpart y queries
a) 100-node flat random network b) 100-node transit-stub network
Figure Generated Net w orks
Sim ulation Mo del
Weha v e implemen ted the route construction algorithm giv en ab o v e within the LBNL net w ork sim u
lator MFF to ev aluate eectiv eness in nding alternate paths Our primary goal is to c haracterize
the p erformance of the heuristics according to a v aried set of top ologies W e are also in terested in
measuring the path length of alternate paths computed using the heuristics they should not b e
man y hops longer than the shortest path
T op ology
W e generated v arious large top ologies using the Georgia T ec h ITM top ology generator ZCB CZ
W e used one at random net w ork of no des Figure a using the DoarLeslie edgeconnection
metho d DL to generate edges that mostly connect no des near eac h other The a v erage degree
of connectivit y for this net w ork is W e also created transitstub top ologies whic h consist
of a bac kb one net w ork and connected stub net w orks with eac h subnet w ork generated randomly Figure b sho ws a no de transit stub net w ork w e used ha ving an a v erage degree of connectivit y
of T o determine howw ell our heuristics scale to larger top ologies w e also generated three
no de transitstub net w orks
W orkload
Giv en a single top ology the p erformance of the route construction heuristics will dep end on the
lo cations of the agen t the sender and the b ottlenec k link Weha v e designed a w orkload that
c haracterizes the eectiv eness of a heuristic for a giv en top ology byv arying the placemen t of these
three en tities F or eac h top ologyasim ulation b egins b y randomly c ho osing a sender and a receiv er
It then iterates through eac h of the links b et w een the sender and receiv er assuming in turn that
eac h link is the b ottlenec k link The route construction agen t whic hw e assume is colo cated with
the receiv er then tries to nd an alternate path around that link
The output of a single sim ulation using a single senderreceiv er pair is the n umberofattempts
made the n um b er of times an alternate path is found the length of the alternate paths and the
n um b er of route queries needed to compute the paths If the heuristic is based on MR andom then w e rep eat eac h sim ulation times to compute a v erages for these n um b ers Other heuristics
only need to b e run one time F or eac h of the generated top ologies w e ran sim ulations for senderreceiv er pairs
T able Success Rate of Heuristics on no de Flat Random Net w ork
Heuristic Av erage Success Rate
Hop HopSender Random RandomSender HopRandom HopRandomSender Hop HopSender Sim ulation Results
Weranabattery ofsim ulations using the ab o vew orkload for eac h of the route construction heuris
tics W e then ev aluated eac h of the heuristics based on success rate and path length
Success Rate
T o determine the eectiv eness of the heuristics w e compute the success rate b y dividing the n um ber
of successes v ersus the n um b er of attempts W edo not coun t attempts where there is no alternate
path a v ailable ie where ev en an algorithm with full kno wledge of the top ology will not nd
an alternate path It is imp ortan t to note that our exp erimen ts ha v e underestimated the success
rate b ecause a giv en route serv er is nding alternate paths to only a small n um b er of senders In
practice a route serv er ma y handle requests for routes to a large n um b er of senders and an y single
request ma y b enet from routes learned for other requests This will particularly b e true when
nding routes around lo cal b ottlenec ks
T able sho ws the success rate of some of the route construction heuristics for a no de at
random net w ork F or the at random net w ork b oth the Hop and Random heuristics are able
to nd an alternate path about of the time An ycom bination of these heuristics with eac h
other or with querying the sender is eectiv eo ver of the time
Although the a v erage success rate for the Hop and Random heuristics is nearly iden tical
the distribution of the success rate among v arious senderreceiv er pairs is quite dieren t Figure sho ws a histogram of the success rate for these t w o heuristics F or these histograms w e group the
alternate path attempts of eac h senderreceiv er pair and calculate the o v erall success rate of eac h
pair Based on these histograms querying lo cal routers will alw a ys b e helpful for some sender
receiv er pairs and will nev er help for others On the con trary querying a single random router will
alw a ys help to nd some alternate paths in a at random net w ork
While the histograms are useful for determining the distribution of a single heuristic graphing
the cum ulativ e distribution of the senderreceiv er paired success rates is useful for comparing man y
dieren t heuristics F or ease in reading these graphs w eha vecon v erted the cum ulativ e distribution
function F
x
a! P x ! ain to a diminutive distribution function DDF F
x
a! P x ! a
Th us for a giv en p oin t on the graph the y v alue represen ts the p ercen tage of senderreceiv er pairs
whose success rate is greater than or equal to the x v alue F or example Figure a sho ws that for
the Hop heuristic of the senderreceiv er pairs nd an alternate path of the time and
0
10
20
30
40
50
60
70
80
90
100
0 10 20 30 40 50 60 70 80 90 100
Percentage of Sender/Receiver Pairs
Percentage of Successful Alternate Path Requests
1-Hop Heuristic
0
10
20
30
40
50
60
70
80
90
100
0 10 20 30 40 50 60 70 80 90 100
Percentage of Sender/Receiver Pairs
Percentage of Successful Alternate Path Requests
1-Random Heuristic
Figure Hop and Random Heuristics on No de Flat Random Net w ork
0
10
20
30
40
50
60
70
80
90
100
0 10 20 30 40 50 60 70 80 90 100
DDF of Paired Success Rate
Sender-Receiver Paired Alternate Path Success Rate
100-node Flat Random Network
1-Hop
1-Random
1-Hop+1-Random
0
10
20
30
40
50
60
70
80
90
100
0 10 20 30 40 50 60 70 80 90 100
DDF of Paired Success Rate
Sender-Receiver Paired Alternate Path Success Rate
100-node Flat Random Network
1-Hop
1-Hop+Sender
1-Hop+1-Random+Sender
2-Hop
a) Adding 1-Hop and 1-Random b) Expanding from 1-Hop to 2-Hop
Figure DDF of Heuristics on No de Flat Random Net w ork
alw a ys nd an alternate path This gure also sho ws that for the Random heuristic all of
the pairs are successful at least of the time but only ha v e a success rate higher than
The HopRandom heuristic com bines the p oten tial for high success of Hop with the lo w er
b ound of Random
Because Hop has suc h a high p oten tial success rate for a senderreceiv er pair w e are in terested
in the v alue of com bining it with other heuristics Figure b compares the eectiv eness of adding
Random and a query to the Sender to Hop v ersus expanding Hop to Hop Adding just a
query to the Sender to Hop increases the p ercen tage of pairs that alw a ys nd an alternate path
from to Adding Random to this com bination raises the lo w er b ound on success rate
from to While Hop still has a lo w er b ound of only of the pairs fall in this category Almost all of the other pairs alw a ys nd an alternate path
While man y of the heuristics p erform w ell on the at random net w ork they all p erform sub
stan tially w orse on the no de transitstub top ology T able sho ws the success rate of the same
T able Success Rate of Heuristics on no de T ransitStub Net w ork
Heuristic Av erage Success Rate Decline F rom Flat Random
Hop HopSender Random RandomSender HopRandom HopRandomSender Hop HopSender 0
10
20
30
40
50
60
70
80
90
100
0 10 20 30 40 50 60 70 80 90 100
DDF of Paired Success Rate
Sender-Receiver Paired Alternate Path Success Rate
100-node Flat Random Network
1-Hop
2-Hop
1-Hop+Sender
2-Hop+Sender
Figure DDF of Heuristics on No de T ransitStub Net w ork
heuristics run o v er the transitstub net w ork along with the dierence b et w een this rate and that
for the at random net w ork
In particular the Random heuristic nds an alternate path only of the time compared
to for the at random net w ork The reason for this decline can b e seen b yb y examining the
no de transit stub top ology in Figure b Nearly all of the no des out of are lo cated
in one of the stub net w orks eac h of whic h has an a v erage of no des Th us when the sim ulation
randomly c ho oses a sender and a receiv er most lik ely they will b e lo cated within t w o dieren t
stubs Then when the agen t randomly explores the path to one of the no des it is lik ely to c ho ose
a no de that is in y et a third stub With this map the agen t will not explore an y part of the sender
and receiv ers stub net w orks except for the shortest paths through them Th us the only times the
agen t is lik ely to nd an alternate path will b e the rare o ccasions when the third part y is within
the same stub as the sender or receiv er or when the b ottlenec k is in the bac kb one
The NHopSender heuristic on the other hand is b etter able to nd alternate paths around
lo cal b ottlenec ks either within the vicinit y of the receiv er or near the sender Th us this heuristic
is m uc h more successful on the transitstub net w ork than those using random no des Figure demonstrates the eectiv eness of querying the Sender in this top ology sho wing the DDF for Hop
and Hop b oth with and without a query to the sender Clearly the b enet of querying the sender
T able Generation P arameters F or T ransitStub Net w orks
Net w ork Num ber Num ber T ransit Stubs Stub
Name No des T ransits Size T ransit Size
far out w eighs the b enet of adding an extra hop to the heuristic Lik ewise increasing the n um ber of
hops has more impact when the heuristic also queries the sender since b oth the sender and receiv er
are expanding the radius of their searc h T o further conrm the suitabilit y of NHopSender for
nding lo cal b ottlenec ks w e reanalyzed the data for this top ology considering only b ottlenec ks
within a stub In this scenario the success rates of HopSender and HopSender rise to
and resp ectiv ely Our sim ulations on no des th us conrm the follo wing results
T op ology aects the p erformance of the heuristics
Querying lo cal no des helps nd alternate paths around lo cal b ottlenec ks
Querying the sender alw a ys helps to nd alternate paths particularly for lo cal b ottlenec ks
near the sender
W e b eliev e that nding routes around lo cal b ottlenec ks is an imp ortan t case b ecause w e exp ect
the bac kb one of the net w ork to b e w ellengineered whereas a connecting domain ma y exp erience
temp orary o v erload
T o observehoww ell these heuristics scale to larger net w orks w e rep eated our sim ulations using
three no de transitstub net w orks W e generated eac h of these net w orks b y building on the
no de transit stub top ology and making a dieren t mo dication for eac h of the three net w orks
T able lists the generation parameters for all of the transitstub net w orks whic hha vean a v erage
degree of connectivit y of and resp ectiv elyF or Net w ork w e sp ecied
a larger transit net w ork resulting in a larger highlyconnected bac kb one and more stub net w orks
eac h no de in the bac kb one retains the same n um ber of a v erage stub net w orks F or Net w ork w e sp ecied more transit net w orks resulting in a larger hierarc hical bac kb one and lik ewise more
stubs Finally for Net w ork w e sp ecied larger stubs whic h retains a v ery small bac kb one
and the same n um b er of stubs T ok eep the no de degree lo w for these stubs w e used the DoarLeslie
edge connection metho d T able sho ws the success rate of some of the heuristics run o v er these larger top ologies as
compared with the no de net w ork Because the HopSender and HopSender heuristics
nd alternate paths around lo cal b ottlenec ks they p erform b est on Net w ork T in whic h
the size of the stubs dominate the net w ork Lik ewise these same heuristics do not p erform as
w ell when the n um b er of transit net w orks is increased in Net w ork T These t w o heuristics
con tin ue to nd alternate paths around lo cal b ottlenec ks either within a stub or a transit net w ork
but do not nd alternate paths around distan t b ottlenec ks when the connection b et w een net w orks
is hierarc hical On the other hand when the bac kb one consists of a large at transit net w ork as
T able Success Rate of Heuristics on no de T ransitStub Net w orks
Av erage Success Rate
Heuristic Net T Net T Net T Net T
HopSender HopSender HopRandomSender HopSender Stubs HopSender Stubs T able P ath Length of Alternate P aths
Av erage Num b er of Extra Hops p er Alternate P ath
Heuristic Net F Net T Net T Net T Net T Global
Hop
Random
HopSender
HopSender
HopRandomSender
with Net w ork T these heuristics can also nd alternate paths within the bac kb one The last
t w o lines of the table emphasize the abilit y of the NHopSender heuristics to nd routes around
lo cal b ottlenec ks If o v erall p erformance on a v ariet y of hierarc hical net w orks is a consideration
then the HopRandomSender heuristic b ycom bining lo cal queries with random queries is
able to consisten tly nd routes around b oth lo cal and distan t b ottlenec ks
P ath Length
W e measured the length of all of the alternate paths found byeac h heuristic and compared the
length to the alternate paths found b y the algorithm using global kno wledge of the top ology The
global algorithm nds the shortest a v ailable alternate path th us b y taking the dierence in path
length w e can determine the n um b er of extra hops in the alternate paths found b y the heuristics
T able lists the a v erage n um b er of extra hops for some of the heuristics on eac h of the top ologies
Weha v e group ed the data in to sev eral categories for ease in comparing the results The rst group
lists the a v erage n um b er of extra hops for the global algorithm Compared to this the Hop
heuristic generally has comparable paths with Random ha ving longer paths The last group
of data includes the three heuristics whose success rates are the highest In almost all cases the
a v erage n um b er of extra hops is b elo w one
These results indicate that the heuristics often nd an alternate path whose length is equal
to that of the shortest path If this also holds true for realw orld net w orks then a distance
v ector unicast routing proto col lik e BGP RL could pass through some equalcost paths to route
serv ers simplifying route computation If these paths are used frequen tly then it will b e c heap er to
distribute them rather than compute them individuall y at eac h route serv er A route serv er w ould
still need to use the route querying heuristics describ ed within this pap er to nd lessfrequen tly
used and sligh tly longer alternate paths when other paths are not adequate This division of lab or
bet w een precomputed paths and ondemand computed paths has b een prop osed earlier as a part
of the unied routing arc hitecture ERH Related W ork
The MORF proto col wepresen t herein is no v el in that it is receiv erorien ted making it scalable
for m ulticast and uses a simple mec hanism to prev en t lo ops The A TM F orum has published
a setup mec hanism for m ulticast trees that includes receiv erorien ted joins but it requires the
source to establish all branc hes of the m ulticast tree UNI In the In ternet comm unit yboth
the STI I DB DHHS and T enet FBZ BF G
proto cols establish m ulticast trees using
primarily senderorien ted mec hanisms STI I sp ecies additional mec hanisms for receiv erorien ted
joins but only along shortestpath routes and no des in the tree m ust kno w the iden tit y of all lea v es
in their subtree More recentw ork allo ws receiv erorien ted setup of nonopp ortunistic branc hes
in a m ulticast tree PGLA but again only along shortestpath routes and at the exp ense of
considerably more state to prev en t lo oping Finally Guerin et al ha v e recen tly prop osed a sender
orien ted route pinning mec hanism R G but the approac h is applicable only to unicast routing
and uses only shortestpath routes
Weha v e discussed sev eral route construction heuristics that do not require global distribution
of top ology Alaettinoglu explored route construction using aggregation and hierarc hical heuris
tics for querying remote aggregates for more detailed information Ala Hotz has studied route
construction heuristics based on route fragmen ts and triangulation of graph p osition Hot In
con trast to the ab o vew ork QoS routing proto cols globally distribute top ology and group infor
mation A large b o dy of QoS m ulticast routing literature uses Steiner trees to optimize tree cost
for small static groups BKJ W ax ChoKPP Salama et al compare these and other
approac hes SR V The In ternet and A TM comm unities ha v e prop osed m ulticast proto cols using
related tec hniques for more dynamic groups R GW ZSSC PNN Because of scaling limits
this w ork is applicable for in tradomain routing
As a means for pro viding a v ariet y of paths to meet application requiremen ts QoR routing is
not a no v el idea Sev eral routing proto cols suc h as OSPF Mo yb allo w the creation of m ultiple
routing table en tries p er destination and a T yp e of Service eld in IP pac k ets has b een designed
to index those tables Alm Recen tw ork b y Matta and Shank ar MS indicates that suc han
approac h can impro veo v erall endtoend dela ys in the net w ork While their approac h uses metrics
based on measured dela y and utilization the form ulations of their metrics result in relativ ely static
measures that th us do not exhibit the oscillation seen in the ARP ANET KZ
The use of alternate path routing during p erio ds of congestion is common in the telecomm u
nications industry AKK GKK MS HSS MG MGH AH T runk reserv ation is
used to limit the use of alternate path routing under high load so that it do es not decrease through
put Aki Lik ewise the In ternet comm unit y has explored the use of adaptiv e routing to a v oid
congestion and impro v e throughput A tt NSC W C Bre Estrin et al in tro duced the
use of hopb yhop routing for commonlyused routes coupled with explicit routing for sp ecialized
routes ERH W ork in the academic comm unit y has pro vided more scalable receiv erorien ted joins for A TM BDG but do es
not pro vide the details of their proto col mec hanisms
Multicast routing in the In ternet w as pioneered b y Deering who dev elop ed pro cedures for
hosts to notify routers of their group mem b ership Deea and m ulticast routing metho ds for
b oth distancev ector and linkstate proto cols Dee D VMRP Deeb WPD is based on the
o o dandprune distancev ector metho d and MOSPF Mo ya Mo yc extends OSPF to include
group mem b ership with linkstate adv ertisemen ts PIM DEF
EFH
in tro duces explicit join
messages for sparselydistributed groups and along with CBT BF C BJR allo ws creation of
m ulticast trees that are shared b y all senders Comparisons of these proto cols include WE BCF G
Conclusions
Weha v e dev elop ed t wok ey enhancemen ts for m ulticast routing that are needed for an in tegrated
services arc hitecture First w eha v e designed a simple lo opfree and scalable route setup proto col
that routers can use to install alternate paths in to a m ulticast tree on b ehalf of lo cal receiv ers
This allo ws receiv ers to utilize routes other than the precomputed shortest path and to install
those routes as nonopp ortunistic paths Second w eha v e armed that route serv ers can use
lo calized route construction tec hniques to compute useful alternate paths without requiring global
distribution of top ologyWeha v e demonstrated the utilityof sev eral lo wcost pro ofofconcept
heuristics as others nd b etter route construction heuristics routers ma y incremen tally deplo y the
impro v emen ts
Because b oth the route setup and route construction heuristics are receiv erorien ted they ma y
co op erate to reduce the impact of alternate paths on the net w ork W eha v e sho wn that the route
construction heuristics nds alternate paths whose a v erage length is not m uc h longer than that of
shortest paths If a computed alternate path matc hes what is already in the m ulticast tree then its
impact on o v erall tree size will b e negligible If it do es not matc h then the route setup mec hanism
will return the path in use b y the m ulticast tree and require the originating router to mo dify its
alternate path Th us a router will alw a ys knowthe en tire path from itself to the ro ot of the tree
allo wing it to c hec k the impact a path will ha v e on the tree b efore installing it
Ac kno wledgme n ts
Bob Braden and fello w researc hers at ISI ha v e encouraged and supp orted this w ork since its in
ception John Heidemann help ed to nd an eectiv ew a y of presen ting sim ulation data Ellen
Zegura and Ken Calv ert pro vided the GTITM top ology generation soft w are along with excellen t
do cumen tation and examples of its use
References
AH Gerald R Ash and BaoSheng D Huang An Analytical Mo del for Adaptiv e Routing
Net w orks IEEE T r ansactions on Communic ations No v em b er Aki J M Akinp elu The Ov erload P erformance of Engineered Net w orks With Nonhier
arc hical and Hierarc hical Routing A TT Bel l L ab or atories T e chnic al Journal Septem b er
AKK G H Ash A H Kafk er and K R Krishnan Servicing and RealTime Con trol
of Net w orks with Dynamic Routing Bel l System T e chnic al Journal Octob er
Ala Cengiz Alaettinoglu Sc alable InterDomain R outing with T oS Policy and T op olo gy
R esolution PhD thesis Univ ersit y of Maryland Alm P Almquist T yp e of Service in the In ternet Proto col Suite RF C July A tt R A ttar A Distributed Adaptiv e MultiP ath Routing Consisten t and Conicting
Decision Making In Fifth A nnual Berkeley Workshop on Distribute d Data Manage
ment and Computer NetworksF ebruary BCF G
T om Billhartz J Bibb Cain Ellen F arreyGoudreau Doug Fieg and Stephen Gordon
Batsell P erformance and Resource Cost Comparisons for the CBT and PIM Multicast
Routing Proto cols IEEE Journal on Sele ctedA r e as in Communic ations April
BDG Ric k Bub enik John DeHart and Mik e Gaddis Multip oin t Connection Managemen t
in High Sp eed Net w orks In IEEE INF OCOM BF C A J Ballardie P F F rancis and J Cro w croft Core Based T rees In A CM SIG
COMM August BF G
R Bettati D F errari A Gupta W Hener W Ho w e M Moran Q Nguy en and
R Y a v atk ar Connection Establishmen t for MultiP art y RealTime Comm unication
In Fifth International Workshop on Network and Op er ating System Supp ort for Digital
A udio and Vide o April BJR A J Ballardie N Jain and S Reev e Core Based T rees CBT Multicast Proto col
Sp ecication w ork in progress No v em ber BKJ Kadaba BarathKumar and Jerey M Jae Routing to Multiple Destinations in
Computer Net w orks IEEE T r ansactions on Communic ations Marc h Bre Lee Breslau A daptive Sour ceR outing of R e alTime T r ac in Inte gr ate d Servic es
Networks PhD thesis Univ ersit y of Southern California Decem b er
CD S Casner and S Deering First IETF In ternet Audio cast A CM SIGCOMM Com
puter Communic ation R eview July Cho CH Chou On Multicast P ath Finding Algorithms In IEEE INF OCOM Cla Da vid D Clark The Design Philosoph y of the D ARPAIn ternet Proto cols In A CM
SIGCOMM August CZ Ken Calv ert and Ellen Zegura Georgia T echIn ternet w ork T op ology Mo dels
h ttpwwwccgatec hedufacEllenZeguragraphsh t ml
DB L Delgrossi and L Berger In ternet Stream Proto col V ersion ST Proto col
Sp ecication V ersion ST RF C August
Deea Stephen Deering Host Extensions for IP Multicasting RF C Ma y Deeb Stephen Deering Multicast Routing in In ternet w orks and Extended LANs In A CM
SIGCOMM August Dee Stephen Deering Multic ast R outing in a Datagr am Internetwork PhD thesis Stan
ford Univ ersit y DEF
Stephen Deering Deb orah Estrin Dino F arinacci V an Jacobson ChingGung Liu
and Liming W ei An Arc hitecture for WideArea Multicast Routing In A CM
SIGCOMM August DH S Deering and R Hinden In ternet Proto col V ersion IPv Sp ecication RF C
Decem ber DHHS Luca Delgrossi Ralf Guido Herrt wic h F rank Oliv er Homann and Sib ylle Sc haller
Receiv erInitiated Comm unication with STI I T ec hnical Rep ort IBM Eu
rop ean Net w orking Cen ter DL Matthew Doar and Ian Leslie Ho w Bad Is Naiv e Multicast Routing" In IEEE
INF OCOM EFH
Deb orah Estrin Dino F arinacci Ahmed Helm yDa vid Thaler Stephen Deering Mark
HandleyV an Jacobson Ching gung Liu Puneet Sharma and Liming W ei Proto col
Indep enden t Multicast Sparse Mo de PIMSM Proto col Sp ecication w ork in
progress Marc h ERH Deb orah Estrin Y ak ovRekh ter and Stev e Hotz A Scalable In terDomain Routing
Arc hitecture In A CM SIGCOMM August EZL
Deb orah Estrin Daniel Zappala T onyLi Y ak ovRekh ter and Kannan V aradhan
Source Demand Routing P ac k et F ormat and F orw arding Sp ecication V ersion RF C Ma y FBZ Domenico F errari Anindo Banerjea and Hui Zhang Net w ork supp ort for m ultimedia
A Discussion of the T enet Approac h Computer Networks and ISDN Systems GKK R J Gibb ons F P Kelley and P BKey Dynamic Alternativ e Routing Mo delling
and Beha vior In Pr o c e e dings of the International T eletr ac Congr ess June Hed C Hedric k Routing Information Proto col RF C STD June Hed Charles L Hedric k An In tro duction to IGRP T ec hnical rep ort The State Univ er
sit y of New Jersey Octob er HLFT S Hanks T Li D F arinacci and P T raina Generic Routing Encapsulation GRE
RF C Octob er Hot Stev en Hotz R outing Information Or ganization to Supp ort Sc alable Inter domain
R outing with Heter o gene ous Path R e quir ements PhD thesis Univ ersit y of Southern
California
HSS B R Hurley C J R Seidl and W F Sew ell A Surv ey of Dynamic Routing Metho ds
for CircuitSwitc hed T rac IEEE Communic ations Magazine Septem ber
IDR In ternational Standards Organization Pr oto c ol for the Exchange of InterDomain
R outing Information among Interme diate Systems to Supp ort F orwar ding of ISO PDUs KPP V P Komp ella J C P asquale and G C P olyzos Multicasting for Multimedia
Applications In IEEE INF OCOM KZ A tul Khanna and John Zinky The Revised ARP ANET Routing Metric In A CM
SIGCOMM Septem ber MFF Stev e McCanne Sally Flo yd and Kevin F all LBNL Net w ork Sim ulator
h ttpeelblgo vns
MG Debasis Mitra and Ric hard J Gibb ens StateDep enden t Routing on Symmetric
Loss Net w orks with T runk Reserv ations Analysis Asymptotics Optimal Design
T ec hnical Rep ort A T#T Bell Lab oratories July MGH Debasis Mitra Ric hard J Gibb ens and B D Huang Analysis and Optimal Design of
AggregatedLeastBusyAlternativ e Routing on Symmetric Loss Net w orks with T runk
Reserv ation In Pr o c e e dings of the th International T eletr ac Congr ess June Mil D L Mills Exterior Gatew a y Proto col for F ormal Sp ecication RF C April
Mo ya J Mo y Multicast Extensions to OSPF RF C Marc h Mo yb J Mo y OSPF V ersion RF C Marc h Mo yc John Mo y Multicast Routing Extensions for OSPF Communic ations of the A CM August MS Debasis Mitra and Judith Seery Comparativ eEv aluations of Randomized and Dy
namic Routing Strategies for CircuitSwitc hed Net w orks IEEE T r ansactions on
Communic ations Jan uary MS Ibrahim Matta and A Uda y a Shank ar T yp eofService Routing in Datagram Deliv ery
Systems IEEE Journal on Sele ctedA r e as in Communic ations Octob er NSC Don J Nelson Khalid Sa y o o d and Hao Chang An Extended LeastHop Distributed
Routing Algorithm IEEE T r ansactions on Communic ations April PGLA Mehrdad P arsa and J J GarciaLunaAcev es A Proto col for Scalable Lo opF ree
Multicast Routing IEEE Journal on Sele ctedA r e as in Communic ations April PNN A TM F orum PNNI SW G PNNI Dr aft Sp e cic ation R R G S Herzog R Guerin S Kamat QoS P ath Managemen t with RSVP w ork in
progress Marc h
R GW A Orda R Guerin and D Williams QoS Routing Mec hanisms and OSPF Exten
sions w ork in progress Marc h RL Y Rekh ter and T Li A Border Gatew a y Proto col BGP RF C July
SR V Hussein F Salama Douglas S Reev es and Y annis Viniotis Ev aluation of Multicast
Routing Algorithms for RealTime Comm unication on HighSp eed Net w orks IEEE
Journal on Sele ctedA r e as in Communic ations April Sti Burkhard Stiller A Surv ey of UNI Signalling Systems and Proto cols for A TM Net
w orks A CM SIGCOMM Computer Communic ation R eview April UNI A TM F orum T ec hnical Committee Signalling Sub w orking Group UNI Signal ling
W ax Bernard M W axman Routing of Multip oin t Connections IEEE Journal on Sele cte d
A r e as in Communic ations Decem ber W C Zheng W ang and Jon Cro w croft Shortest P ath First with Emergency Exits In
A CM SIGCOMM Septem ber WE Liming W ei and Deb orah Estrin The T radeos of Multicast T rees and Algorithms
In International Confer enc e on Computer Communic ations Networks Septem ber
WPD D W aitzman C P artridge and S Deering Distance V ector Multicast Routing
Proto col RF C No v em ber ZCB Ellen W Zegura Ken Calv ert and S Bhattac harjee Ho w to Mo del an In ternet w ork
In IEEE INF OCOM ZSSC Z Zhang C Sanc hez B Salk ewicz and E Cra wley Qualit y of Service Extensions
to OSPF w ork in progress June
Linked assets
Computer Science Technical Report Archive
Conceptually similar
PDF
USC Computer Science Technical Reports, no. 606 (1995)
PDF
USC Computer Science Technical Reports, no. 614 (1995)
PDF
USC Computer Science Technical Reports, no. 656 (1997)
PDF
USC Computer Science Technical Reports, no. 648 (1997)
PDF
USC Computer Science Technical Reports, no. 657 (1997)
PDF
USC Computer Science Technical Reports, no. 605 (1995)
PDF
USC Computer Science Technical Reports, no. 673 (1998)
PDF
USC Computer Science Technical Reports, no. 731 (2000)
PDF
USC Computer Science Technical Reports, no. 674 (1998)
PDF
USC Computer Science Technical Reports, no. 613 (1995)
PDF
USC Computer Science Technical Reports, no. 565 (1994)
PDF
USC Computer Science Technical Reports, no. 644 (1997)
PDF
USC Computer Science Technical Reports, no. 678 (1998)
PDF
USC Computer Science Technical Reports, no. 727 (2000)
PDF
USC Computer Science Technical Reports, no. 690 (1998)
PDF
USC Computer Science Technical Reports, no. 723 (2000)
PDF
USC Computer Science Technical Reports, no. 585 (1994)
PDF
USC Computer Science Technical Reports, no. 560 (1993)
PDF
USC Computer Science Technical Reports, no. 599 (1995)
PDF
USC Computer Science Technical Reports, no. 631 (1996)
Description
Daniel Zappala, Deborah Estrin, Scott Shenker. "Alternate path routing and pinning for interdomain multicast routing." Computer Science Technical Reports (Los Angeles, California, USA: University of Southern California. Department of Computer Science) no. 655 (1997).
Asset Metadata
Creator
Estrin, Deborah
(author),
Shenker, Scott
(author),
Zappala, Daniel
(author)
Core Title
USC Computer Science Technical Reports, no. 655 (1997)
Alternative Title
Alternate path routing and pinning for interdomain multicast routing (
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
24 pages
(extent),
technical reports
(aat)
Language
English
Unique identifier
UC16270731
Identifier
97-655 Alternate Path Routing and Pinning for Interdomain Multicast Routing (filename)
Legacy Identifier
usc-cstr-97-655
Format
24 pages (extent),technical reports (aat)
Rights
Department of Computer Science (University of Southern California) and the author(s).
Internet Media Type
application/pdf
Copyright
In copyright - Non-commercial use permitted (https://rightsstatements.org/vocab/InC-NC/1.0/
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)
Copyright
In copyright - Non-commercial use permitted (https://rightsstatements.org/vocab/InC-NC/1.0/