Computer Science Technical Report Archive

USC Computer Science Technical Reports, no. 530 (1992)

(USC DC Other)

USC Computer Science Technical Reports, no. 530 (1992)

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Analysis of a Resequencer Mo del for Multicast
o v er A TM Net w orks
Liming W ei
F ongChing Lia w
Deb orah Estrin
Allyn Romano w
T om Ly on
Computer Science Departmen t Univ ersit y of Southern California
Los Angeles CA Sun Microsystems Computer Corp oration Moun tain View CA Abstract Multicast deliv ery sa v es bandwidth and oers logical ad
dressing capabilities to the applications The receiv ers of a m ulticast
group need to dieren tiate cells sentb y dieren t sources This dem ulti
plexing requiremen t can b e satised in an A TM en vironmen t using m ul
tiple dedicated p oin ttom ultip oin t virtual c hannel connections V Cs
but with certain shortcomings This pap er discusses an alternativ erese quencing mo del to solv e this problem It scales w ell in large net w orks
Three resequencing metho ds are are dev elop ed and sim ulation results
rep orted The strategy is useful for applications spanning large regions
where it is desirable to mix streams of cells from dieren tburst y sources
on to the same virtual c hannel
Appeared in the Proceedings of the rd Internat ion al Workshop on Network
and OS Support for Digital Audio and Video San Diego Nov In tro duction
It is imp ortan t for future A TM net w orks to ha vem ulticast capabilit y as they will
supp ort suc h applications as teleconferencing and information distribution ser
vices Applications based on m ulticast ha vet womajor adv an tages o v er unicast
logical addressing and bandwidth sa vings Deering With logical addressing an
application uses a single m ulticast group address to send and receiv e data It is
not necessary for senders and receiv ers to knowthe n um b er or lo cation of group
mem b ers Multicast pro vides signican tsa vings in bandwidth b ecause the sender
transfers only a single cop y of the data Data cells are not replicated un til they
W ork supp orted b y SUN summer in tership at SMCC Moun tain View California
Wew ould lik e to thank Stev e Deering Bry an Lyles Lixia Zhang and the referees for
helpful discussions and commen ts
reac h a branc hing p ointinthe m ulticast deliv ery tree at a lo cation closer to the
destinations
The results of researc h and exp erimen tal implemen tations of m ulticast in
IP net w orks ha v e demonstrated the b enets of supp orting m ulticast in public
data net w orks RF C MOSPF Although it w ould b e desirable to adopt
the implemen tatio n approac hes successfully used in IP net w orks the dierences
bet w een an A TM net w ork and a con v en tional pac k et switc hed net w ork makeit
dicult to directly map m ulticast mo dels for IP in to A TM net w orks
In IP net w orks for a m ulticast group ha ving N senders N m ulticast deliv ery
trees are established eac h tree deliv ering pac k ets from a source to all desti
nations in onetoman y fashion Eac hIP pac k et carries a source address and
a destination ie m ulticast group address in its header RF C T o route
m ulticast pac k ets b oth destination address and source addresses are used for
lo okup in the routing table A receiv er uses the source address eld of the IP
header to distinguish pac k ets from dieren t senders Th us m ulticast pac k ets can
b e arbitrarily in termixed with eac h other on the forw arding path
A TM ho w ev er is a connection orien ted tec hnologyin whic h connections
m ust b e explicitly setup b efore data can b e transferred An A TM connection
can b e view ed as a cac hed route LLR in the sense that eac h cell carries only
a small routing tag the virtual path iden tier and virtual c hannel iden tier
VPIV CI Unfortunately if cells from dieren t sources are m ultiplexed on to
the same virtual c hannel they will carry the same routing tag or VPIV CI
up on arriv al at a receiv er There is not a straigh tforw ard w a y to distinguish cells
sen t from dieren t sources This is often referred to as the c el l demultiplexing
problem Although sev eral solutions ha v e b een prop osed they ha v e shortcomings
and most imp ortan tly they do not scale w ell
This w ork prop oses a solution that b oth solv es the cell dem ultiplexing prob
lem and scales w ell The follo wing section describ es curren tly prop osed solutions
to A TM m ulticasting and discusses their shortcomings The third section dev el
ops the prop osed resequencer mo del In section four sim ulation results for the
p erformance of sev eral dieren t metho ds of resequencing are presen ted
Strategies for Implemen ting Multicast in A TM
The cell dem ultiplexing problem can b e a v oided if one VPIV CI is used for
eac h source In this case eac hm ulticast VPIV CI is a onetoman y connection
represen ting one m ulticast tree and is indep enden t from other VPIV CIs Eac h
tree can b e optimized according to some criterion suc h as least dela y or least
cost
The VPI is an bit eld in the cell header If it is used to iden tify a m ulticast
group it restricts the n umberofm ulticast connections to ev en less than the
n um ber a v ailable with V CIs In addition VP switc hing ma y b e used to bundle
large n umberofV C connections or to separate V Cs of diering qualit y of service
classes Therefore w e assume that the VP will not b e a v ailable for use as a
m ulticast tree iden tier F or the remainder of this discussion w e refer to the use
Fig Onev cp ersource m ulticast mo del
The onev cp ersource strategy is compatible with the curren tIP m ulticast
mo dels RF C pro vided that the switc h signaling mec hanisms ha v e access
to the installed IP m ulticast routing tables LLR The sc heme is useful for a
applications with small n um b ers of sources b lo cal m ulticast groups where
there is not a shortage of V Cs relativ e to the demand and c applications
where all sources con tin uously transmit eg video
Ho w ev er this sc heme uses a large n umberofV CIs and therefore do es not
scale w ell to net w orks with man y no des and large n um b ers of longliv ed m ulticast
groups
In particular the sc heme has a n um b er of limitatio ns The n um b er of vir
tual c hannels is to o limited to accomo date a sizable n um b er of large longliv ed
groups F urthermore if bandwidth is reserv ed or allo cated on a p er VCba sis the large n um ber of V Cs resulting from a m ultcast group will quic kly use
up the a v ailable bandwidth Similarl y In the public net w ork if the cost of cell
switc hing dep ends on the n um b er of virtual c hannels the additional V Cs used
in p ersourceV C will b e a costly disadv an tage In addition since eac h sender
denes a tree when a new receiv er joins a group it m ust b e added to all existing
trees
It is imp ortantfor A TM m ulticast to scale to large size Applications in v olv
ing h undreds or ev en thousands of end users globally are not to o far distan t
Th us it is desirable to explore alternativ es in whic h the n umberofa v ailable V Cs
in a switc h or host in terface is not an upp er b ound on the n umberofsources an
application can supp ort
Recen tly sev eral mo dels ha v e b een prop osed for A TM m ulticast aimed at
solving the V C depletion problem and ac hieving high throughput and scalabilit y
BGD GREL YL SBO KPP One prop osal GREL YL is for all sources in
am ulticast group to share one virtual c hannel and to use collision detection
tec hniques suc h as Aloha to a v oid in termixing cells from dieren t sources The
CR C will detect when a cell is out of order ie not from the same sender and
the sender retransmits In this sc heme the common virtual c hannel is treated as
a shared media similar to an ethernet cable with some of the same adv an tages
and disadv an tages ethernet has The sc heme is useable for lo cal lo w bandwidth
and non realtime applications
In another sc heme BGD an iden tier in the A TM pa yload eld is used to
distinguish cells from dieren t sources It is not clear ho w simple or complex
the pro cedures for negotiating unique iden tiers among sources w ould b e This
sc heme w as dev elop ed for AAL in whic h there is a MID m ultiplexing ID eld
in the cell information header Ho w ev er curren tly manydev elop ers of A TM
plan to use AAL SEAL whic hdoes not ha v e sucha dem ultiplexing eld in the
header In an y case the width of the MID eld whic h is bits puts a limit on
the n um b er of sources allo w ed in a group Another cost is that the host in terface
is required to ha v e additional dem ultiplexing hardw are
The Resequencer Solution to Multicast
The sc heme wein tro duce solv es the V CI depletion and bandwidth allo cation
problems without requiring sp ecialized bandwidth allo cation sc hemes for certain
common cases of m ulticast The trac classes for whic h this approachisw ell
suited are a transmission p er source is not con tin uous b small to mo dest
size PDUs c large m ulticast groups First w e describ e the basic resequencing
strategy follo w ed b y sev eral extensions and renemen ts to the basic idea Next
w e presen t a group tree mo del whic h extends the sc heme to a larger net w ork
Resequencers
Toac hiev e b etter bandwidth sharing without complicating the V Cbased re
source managementmec hanisms and to a v oid the VCa v ailabilit y dilemm a cells
from dieren t sources are m ultiplexed on to a single V C A designated source in
the group is elected as a r ese quenc er All other sources send their m ulticast A TM
cells to the designated resequencer whic h buers incoming cells from eac h source
b efore all cells of a PDU are receiv ed After the last cell of a PDU has b een re
ceiv ed the resequencer forw ards all cells of that PDU on to a single outgoing V C
without in terlea ving it with cells from other sources
Note that the outgoing V C of the resequencer is a p oin ttom ulti pointcon nection Cells are duplicated at the branc hing p oin ts and the order of cells is
preserv ed across all links in this onetoman y connection If a sender is also a
receiv er there is a branc h of the onetoman yV C leading bac k to the sender
Although this single resequencer sc heme pro vides the b enets of m ulticast it
has sev eral problems The pro cessor and link sp eed of the source designated as
a resequencer could b ecome a b ottlenec k Cho osing the resequencer is complex
and there ma y b e situations where no ideal selection of designated resequencer
exists This sc heme is b est when all senders are close to the resequencer
T o extend the sc heme to accommo date the case where mem b ers are spread
o v er dieren t regions and eac h region has man y mem bers mem b ers in eac h
region elect a source to act as a resequencer for lo cal senders A onetoman y
V C is created for eac h resequencer whic h deliv ers cells to all other resequencers
of the group and cells are therefore deliv ered to all receiv ers The n um ber of
V Cs required is no w reduced to the n um b er of participating regions instead of
the n um b er of sources N onetoman yV Cs for N resequencers
Tomak e the sc heme more ecien t resequencing can b e done in switc hes
instead of in hosts Normally switc hes sustain higher aggregate cell rates T o
participate in m ulticast the host only needs to kno w the address of its nearest
m ulticast resequencer In a lo cal area A TM the lo cation of a m ulticast rese
quencer can b e either statically assigned or dynamicaly disco v ered so that the
nativenodes are k ept up dated ab out the a v ailable m ulticast serv ers T oa v oid
long call setup latency the resequencer information can b e cac hed in the host
This approac h eliminates the necessit y of going through a decision pro cess to
c ho ose a host resequencer
Multicast Group T rees
Our nal renemen t of this approac h is to further reduce the n um ber of V Cs
main tained for a group Wedo this b y building a group tree instead of p er
source or p ersourceregion ro oted trees Connections among the resequencers
are bidirectional A resequencer serializes all cells destined for a common group
and duplicates them on to the appropriate outgoing p orts with correct VPIV CIs
some lead to other resequencers some go to lo cal mem b ers Figure depicts
a group tree of resequencers and receiv ers The group tree is a deliv ery path in
common for cells from all sources As Figure sho ws eac h source establishes a
p oin ttop oin t connection to its resequencer and sends cells to w ards a m ulticast
group along that connection as if it w ere a unicast destination
Either p oin ttop oin t or p oin ttom ultip oi n t connections can b e used to dis
tribute cells from a resequencer to its lo cal receiving mem b ers dep ending on
supp ort from the lo cal net w ork If a mem b er is b oth a sender and a receiv er a
bidirectional p oin ttop oin t connection is used The connections b et w een s s
and Resequencer M in g are suc h examples T o reduce the bandwidth used
in lo cal m ulticast cell deliv ery lo cal switc hes m ust supp ort p oin ttom ultip oin t
connections The connection b et w een resequencer M and r r and r in g is suc h a p oin ttom ultip oi n t connection where r r and r are all receiv ers
Note also that there ma ybe manypure A TM switc hes b et w een anyt w o
resequencers and b et w een an y resequencer and its lo cal mem b ers The paths
dra wn in gure are logical c hannels
Note that it is not suggested to replace the rev erse c hannels of s s and s with a
single p oin ttom ultip oin t connection unless the senders dont care if they receiv e
duplicated cells lo op ed backb y resequencer M
Fig The r ese quenc er and gr oup multic ast tr e e mo del
Discussion
The resequencer approac h requires algorithms dieren t from those used for IP
m ulticast to compute the group m ulticast tree The tree will not b e optimal
in terms of shortest path or least dela yHo w ev er sub optimal solutions should
suce for a n um b er of applications KPP CBT
Since this resequencer and group tree mo del can b e built up on onetoone
and onetoman y virtual c hannel connections it can co exist with the onev cp er
source mo del An application can c ho ose one of the t w o mo dels F or example
video applications where sources transmit con tin uously mayc ho ose to use the
onev cp ersource mo del whereas video conferencing with compressed streams
and m ultiple c hanging sources mayc ho ose the resequencer approac h The re
sequencer and group tree mo del is also exible in that it w orks b oth with and
without p oin ttom ultip oi n t virtual c hannels The dierence is in the bandwidth
sa vings of the lo cal cell deliv ery paths This approac h turns out to b e esp ecially
w ell suited for v oice conference applications where one bandwidth allo cation for
the whole m ulticast group ma y b e sucien t and is indep enden t of the n um ber
of sources In an audio conference it is un usual to ha vem ultiple sim ultaneous
transmitters Usually only one or t w o p ersons sp eak at a time as is consisten t
with higher lev el h uman proto cols
In order to setup and main tain the group tree a higher la y er proto col either a net w ork la y er proto col or a call con trol lev el proto col is needed Dier
en t proto cols ma y b e appropriate for this purp ose One p ossibilit yis to adapt
Deerings host mem b ership proto col and m ulticast tree setup proto cols Deering
The resequencer and group tree m ulticast mo del has a nice scaling prop ert y Only one virtual c hannel is needed for eac h receiv er to receiv e cells from all
sources When a new mem b er b e it a sender or a receiv er or b oth joins it only
needs to do the setup necessary to reac h a resequencer already in the group The
op erations of join and lea v e are inheren tly lo cal gro wing a branchorc hopping
o a branc h do es not ha veto in v olv e a distan t part of the tree
Tw o p erformance parameters maxim um throughput and dela y will b e used
to assess the resequencer and group tree mo del and will determine the range of
applications of this mo del Dieren t implemen tation metho ds result in dieren t
maxim um throughputs ranging from tens of megabits p er second to that ap
proac hing the maxim um unicast rate W e will sho w in the next section that as
the needs for higher sp eed m ulticast service app ear there are w a ys to increase
the maxim um throughput without c hanging the end system
Three Approac hes to Resequencing
In the resequencer and group tree mo del resequencers are the crucial comp o
nen ts that determine p erformance W e describ e three dieren tsc hemes for re
sequencing cell streams from dieren t sources in soft w are at the net w ork la y er
in hardw are and an optimized hardw are sc heme The p erformance of these ap
proachesisev aluated with resp ect to throughput dela y and dela y jitter
Soft w are Resequencing Multicast cells from the same source can b e reassem
bled at the resequencer in the same w a y that signaling messages are reassem
bled An A TM switc h with signaling capabilit y has asso ciated CPU and memory
to pro cess and forw ard signaling messages QB Similarly for resequencing
of m ulticast cells the reassem bled PDUs are passed to the con trol pro cessor
whichmak es routing decisions b y insp ecting the address eld in the header of
the PDU The forw arding pro cess is sho wn in gure a The route lo okup and
PDU forw arding are done in soft w are similar to a datagram router
T o examine p erformance as sho wn in gure b the cell forw arding dela y
can b e decomp osed in to three parts the dela y to collect cells of the PDU
the dela y due to the route lo okup algorithm and transmission dela y W e observ e that the cell collection time and transmission time of consecutiv e
PDUs can b e o v erlapp ed Therefore route lo okup is the dominan t factor aecting
throughput
With certain optimizations in the implemen tation the soft w are forw arding
sc heme can ac hiev e high throughput Consider a switc h equipp ed with a MIPS pro cessor Assume its m ulticast routing table has en tries and it
PDU queuing dela y needs not b e considered here if w e assume no congestion
Fig Soft w are PDU forw arding
tak es ab out instructions to forw ard one PDU
The p er PDU route lo okup
and forw arding time will b e s F or b yte PDU this corresp onds to a
maxim um throughput of Mbps if the cell collection times and transmission
times of consecutiv e PDUs are completely o v erlapp ed
The dela y exp erienced byeac h individual PDU dep ends on a n um b er of fac
tors In the onev cp ersource mo del a PDU is not reassem bled un til reac hing an
end host Using the soft w are resequencing metho d a PDU has to b e reassem bled
in ev ery resequencer it passes The additional dela y is the sum of cell collection
dela ys and routing dela ys the PDU exp erienced in all resequencers along its
path The PDU cell collection dela y dep ends on the PDU size and link sp eed
Figure sho ws the PDU cell collection dela y under dieren t link sp eeds and
PDU sizes Multimedia applications with dela y jitter or other realtime require
men ts tend to use small pac k et sizes b et w een to b ytes Ev en with larger
PDU sizes Bytes and slo w er link sp eed Mbps the w orst case aggre
gate reassem bly and routing delayinthe abo v e example can still b e less than
This is a rough estimation If the op erating system in A TM switc hes are b etter
designed for comm unication proto col pro cessing the n um b er of instructions w ould
b e smaller
ms if less than resequencers are in v olv ed in the longest path
100.0 300.0 500.0 700.0 900.0 1100.0
PDU size (Byte)
0.0
20.0
40.0
60.0
PDU Reassembling Delay (us)
Link Speed=155Mbps
Link Speed=622Mbps
Fig PDU reassem bly dela y
Th us the p erformance of the soft w are resequencing metho d in terms of sp eed
and dela y is go o d for applications requiring less than Mbps and in v olving less
than thousands of regions pro vided the longest path of the m ulticast group
tree has less than a few h undred resequencer s Hardw are Resequencing When higher throughput is required w e prop ose
a metho d using hardw are supp ort in the switc h Instead of a soft w are route
lo okup m ulticast cells can b e forw arded directly b y hardw are switc h fabrics
using mec hanisms similar to that used for forw arding p oin ttop oin t connection
cells Incoming V Cs carrying m ulticast cells are treated separately from those for
normal p oin ttop oin t connections they are queued in buers neither forw arded
nor reassem bled The switc h monitors the End of P ac k et EoP cell When the
EoP cell arriv es the hardw are dumps all cells in that PDU bac ktobac kon to
outgoing links
Hardw are resequencing requires extra buers and buer managemen tmec h
anisms in the resequencing switchhardw are The amoun t of buer required
dep ends on a v ariet y of factors including the PDU size trac c haracteristics
congestion con trol algorithms etc W e can estimate the n um b er of buers re
quired b y a resequencer in the restricted case assuming no congestion and that
sources ob ey negotiated resource allo cations In this case a switc h has sucien t
bandwidth to carry the incoming trac and the queue size for eac h incoming
V C should not growbey ond PDUs More buer space is needed only when
there is congestion If the maxim um n um b er of activem ulticast groups is N the
maxim um n um b er of activ e source streams coming in to a resequencer for eac h
activ e group is S and the maxim um PDU size used b y all m ulticast sources is
P the maxim um amoun t of buer required in a resequencer is N S P Optimized Hardw are Cell F orw arding The third metho d is an optimiza
tion of the second When the resequencer receiv es the rst cell of a PDU from
a source if there are no cells from other sources queued or b eing forw arded
it directly forw ards all cells on the outgoing links without queuing them rst
un til it forw ards the End of P ac k et cell During the forw arding action if cells
from other sources arriv e ie on other incoming links they will b e queued and
pro cessed according to either the soft w are or hardw are resequencing metho d
T o reduce delayv ariance after an EOP cell has b een sen t the switc h should
try to pro cess queued cells b efore pro cessing newlyarriving cells These latter
cells are queued
This optimized sc heme is particularly suited for certain applications suc has
v oice conferences where normally only one or v ery few participan ts are activ ely
transmitting to the group at the same time Both queuing time and route lo okup
time for the preferred groups are eliminated This sc heme ac hiev es almost the
same cell deliv ery throughput and delayasonev cp ersource metho d ho w ev er
it uses only a single V C for the whole group This metho d is less suitable for
groups with man y mem b ers sim ultaneously transmitting data at a constantrate
to the group
Sim ulation Studies
Since this mo del is targeted to w ards applications with realtime constrain ts w e
ha v e built an ev en t driv en sim ulator to study the cell forw arding dela ys and
dela y jitters Endtoend dela y and v ariance are dep endan t on p er switc h dela ys
and other parameters Wesim ulate and measure the switc hes under dieren t
trac conditions and with eac h of the resequencing metho ds describ ed
The sim ulator uses a queuing mo del to c haracterize the b eha viors of dieren t
buers in a switc h The input queues buer incoming cells the PDU queue is
used b y the soft w are resequencer to sa v e reassem bled PDUs b efore routelo okup
the output queues store outgoing cells when the cell arriv al rate is larger than
the link sp eed F or simplicit y a FIF O algorithm is used to sc hedule the resources
whenev er comp etition o ccurs The link sp eed link dela ynet w ork connectivit y soft w are PDU forw arding sp eed endtoend route and m ulticast group informa
tion etc are en tered b y the user and stored statically A simple trac mo del has b een used to appro ximate AudioVideo st yle trac
sources in the sim ulations The trac mo del assumes that eac h source generates
data at a constan t rate and ev ery t seconds a PDU is ready to send Then the
PDU is c hopp ed in to A TM cells and senton to the net w ork at the link sp eed
while at the same time the application slo w er than the net w ork is generating
data for the next PDU The source stops generating data after sending sev eral
PDUs and sta ys silen t for a randomly selected p erio d of time Then it starts
sending again
A link during the in terv al b et w een deliv ery of t w o consecutiv e PDUs is lik e
av acuum This partially explains wh y in the follo wing sim ulation ev en when
the total activ e sending time of all the relativ ely slowmem b ers of the group
exceeds the dela y caused b y resequencing dieren t PDUs in collision is
not v ery signican t Of course there is a c hance that the n um b er of collisions can
b e larger if the slo w sources together with the link delaymak e dieren t PDUs
frequen tly arriv e at a resequencer at the same time
A fully connected no de net w ork including hosts and switc hes w as
constructed A total of m ulticast groups w ere created The resequencer switc h
lo cated at the net w ork b ottlenec k lo cation w as setup for measuremen ts Figure sho ws the distributions of the cell dela ys across that switc h The same trac
sources are used for the three runs with dieren t resequencing metho ds The
measured path has four sources sending to the same group and whose p ercen t
age of time activ ely sending are Although the source rates
are b elo w the saturation p oin t they do collide inside the resequencer and get
queued and reordered on exits The tails of the curv es are caused b y the queu
ing dela ys The horizon tal distance b et w een the Hardw are resequencing curv e
and the soft w are resequencing curv e reects the route lo ok up dela y s
The horizon tal distance b et w een the optimized and the hardw are resequencing
curv es reects the cell collection time related to link sp eed and PDU size The
dierences in heigh ts of the three curv es depict the fact that the more time a
PDU sp ends in a resequencer the more c hance it ma y collide with another PDU
The widths of the dela y distribution curv es sho ws the jitter the cell exp erienced
Most of the sources sim ulated use cell PDUs except for one that uses
smaller cell PDU and one that uses larger cell PDUs With larger PDU
sizes the curv es in the picture will mo v e horizon tallyCurv es and will mo v e
to w ards the righ t indicating larger cell collection dela y and more c hances of
collision w ould o ccur The lines at the measured resequencer are hea vily utilized
at to of a v ailable bandwidth
Conclusion
Weha vein tro duced a resequencer and group tree m ulticast mo del for m ulticast
o v er A TM net w orks and the trac classes for whic h this approachisw ell suited
References
GREL YL D H Greene and J Bry an Lyles R eliability of A daptation L ayers rd
IFIP W G w orkshop on proto cols for highsp eed net w orks Sto c k
holm Ma y RF C S Deering Host Extensions for IP multic astingRF C Aug Deering S Deering Multic ast R outing in a Datagr am Internetwork Stanford Uni
v ersit y PhD Thesis Dec
0.0 100.0 200.0
Switch Cell forwarding delay (Microseconds)
0.0
1000.0
2000.0
3000.0
4000.0
5000.0
6000.0
Number of cells
1
2
3
1. Software Resequencing
2. Hardware Resequencing
3. Optimized Resequencing
Fig Distribution of cell forw arding dela ys
BGD R Bub enik M Gaddis J DeHart Communic ating with Virtual Paths and
Virtual Channels Pro ceedings of Info comm KPP V Komp ella J P asquale G P olyzos Multic asting for Multime dia Appli
c ations Pro ceedings of Info comm MOSPF J Mo y Multic ast Extensions to OSPFIn ternet Draft No v em ber SEAL T Ly on Simple and Ecient A daptationL ayerSEAL Prop osal
TS to ANSI w orking group TS Aug CSZ D Clark S Shenk er L Zhang Supp orting R e alTime Applic atio ns in an
Inte gr ate d Servic es Packet Network A r chite ctur e and Me chanism Pro ceed
ings of SigComm LLR T Ly on F Lia w A Romano w Network L ayer A r chite cturefor A TM Net
worksIn ternet Draft June Lyles B Lyles Private Conversation June QB CCITT draft text QB sp e cic atio n Marc h SBO A Segall T Barzilai and Y Ofek R eliable Multiuser T r e e Setup with L o c al
Identiers INF OCOMM CBT J Cro w croft A Ballardie P Tsuc hiy a Cor e BasedT r e esIn ternet Draft
August

Linked assets

Computer Science Technical Report Archive

Conceptually similar

PDF

USC Computer Science Technical Reports, no. 560 (1993)

PDF

USC Computer Science Technical Reports, no. 565 (1994)

PDF

USC Computer Science Technical Reports, no. 599 (1995)

PDF

USC Computer Science Technical Reports, no. 613 (1995)

PDF

USC Computer Science Technical Reports, no. 608 (1995)

PDF

USC Computer Science Technical Reports, no. 614 (1995)

PDF

USC Computer Science Technical Reports, no. 690 (1998)

PDF

USC Computer Science Technical Reports, no. 657 (1997)

PDF

USC Computer Science Technical Reports, no. 672 (1998)

PDF

USC Computer Science Technical Reports, no. 655 (1997)

PDF

USC Computer Science Technical Reports, no. 723 (2000)

PDF

USC Computer Science Technical Reports, no. 745 (2001)

PDF

USC Computer Science Technical Reports, no. 667 (1998)

PDF

USC Computer Science Technical Reports, no. 682 (1998)

PDF

USC Computer Science Technical Reports, no. 644 (1997)

PDF

USC Computer Science Technical Reports, no. 750 (2001)

PDF

USC Computer Science Technical Reports, no. 703 (1999)

PDF

USC Computer Science Technical Reports, no. 774 (2002)

PDF

USC Computer Science Technical Reports, no. 735 (2000)

PDF

USC Computer Science Technical Reports, no. 700 (1999)

Description

Liming Wei, Fonching Liaw, Deborah Estrin, Allyn Romanow, Tom Lyon. "Analysis of a resequencer model for multicast over ATM networks." Computer Science Technical Reports (Los Angeles, California, USA: University of Southern California. Department of Computer Science) no. 530 (1992).

Asset Metadata

Creator Estrin, Deborah (author), Liaw, Fonching (author), Lyon, Tom (author), Romanow, Allyn (author), Wei, Liming (author)

Core Title USC Computer Science Technical Reports, no. 530 (1992)

Alternative Title Analysis of a resequencer model for multicast over ATM 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 12 pages (extent), technical reports (aat)

Language English

Unique identifier UC16270558

Identifier 92-530 Analysis of a Resequencer Model for Multicast over ATM Networks (filename)

Legacy Identifier usc-cstr-92-530

Format 12 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)