Computer Science Technical Report Archive

USC Computer Science Technical Reports, no. 678 (1998)

(USC DC Other)

USC Computer Science Technical Reports, no. 678 (1998)

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Virtual In terNet w ork T estb ed Status and Researc h Agenda
Sandeep Ba ja j Lee Breslau Deb orah Estrin Kevin F all
Sally Flo yd P adma Haldar Mark Handley Ahmed Helm y John Heidemann P olly Huang Satish Kumar Stev en McCanne
Reza Rejaie Puneet Sharma Scott Shenk er Kannan V aradhan
Haob o Y u YaXu Daniel Zappala
USC Computer Science DepartmentT ec hnical Rep ort July
Abstract
Sim ulation is an imp ortan t to ol in net w ork proto
col dev elopmen t pro viding an eectiv e w a y to p er
form con trolled exp erimen ts consider alternativ ede signs understand proto col in teractions and examine
scales and top ologies that are dicult to create in
the lab oratory Ho w ev er the scale and heterogene
ityof toda ys net w orks create c hallenges for net w ork
sim ulation In particular conguring sim ulation in
puts prop erly mo deling the m yriad of in teracting
proto cols and pro cessing and visualizing sim ulation
output is b ecoming increasing dicult The VINT
pro ject is dev eloping a set of to ols cen tered around ns
sim ulator and nam net w ork animator to meet these
c hallenges This pap er describ es the curren t status
and future directions of this ongoing w ork
In tro duction
The rapid diusion of in ternet w orking tec hnology
brings t w o ma jor sources of stress to the underly
ing proto col mec hanisms and asso ciated design meth
o ds scale and heterogeneit y Scale m ust b e consid
This researc h is supp orted b y the Defense Adv anced Re
searc h Pro jects Agency D ARP A through the VINT pro ject
at LBL under D ARP A order E at USCISI under D ARP A
gran t ABTC at Xero xP AR C under D ARPAgran t
D ABTC
ered in ev aluating b oth the correctness and p erfor
mance of wide area in ternet proto cols at ev ery lev el
from routing to transp ort and application proto
cols Heterogeneit y of applications translates in to
a larger n um ber of in teracting proto cols service re
quiremen ts and trac patterns
Sim ulation is a v aluable to ol in designing and ev al
uating proto cols for large heterogeneous net w orks
It allo ws rep eatable and con trolled exp erimen tation
while a v oiding the cost and complexit y of net w ork
testb eds The VINT Virtual In terNet w ork T estb ed
pro ject is dev eloping the sim ulation to ols and sup
p orting infrastructure to address the problems of
scale and heterogeneit y across a range of net w orking
proto cols and studies
The soft w are platform for our eorts is ns a dis
crete ev en t sim ulator comp osed of a C
core an
ob jectorien ted extension to the Tcl scripting lan
guage called OTcl and a gro wing framew ork of net
w orking proto cols and building blo c ks Up on this
base w e are dev eloping the follo wing tec hniques and
to ols to enable systematic studies of net w ork proto col
p erformance as scale and proto col in teractions gro w
Libraries of net w ork proto cols Ns pro vides a
ric h set of proto cols with whic h to build sim u
lations Due to its ob ject framew ork customiza
tion of the existing proto cols and addition of new
proto cols is straigh tforw ard The framew ork en

courages extension and mo dication b yusers to
promote in v estigation of v arian ts to existing pro
to cols and testing of new alternativ es
Libraries of top ologies and trac generators
The load mix of the In ternet is shifting rapidly
no w to WWW and expanding to audio and
video and the top ology has alw a ys b een in
ux it is imp ortan ttopro vide researc hers with
the abilit y to explore the implications of these
top ology and load shifts W e are dev eloping a
library of net w ork top ologies and trac gener
ators for ns to facilitate proto col p erformance
testing o v er a wide range of conditions This
common library of sim ulation en vironmen ts
will also allo w b etter comparison bet w een
dieren t sim ulations easing the v erication of
the sim ulation results
Abstraction tec hniques and to ols It is hard to
iden tify the relev an t phenomena buried within
the moun tain of data generated b y a detailed
net w ork sim ulation F urthermore resource lim
itations often constrain the n um ber of net w ork
ob jects ie no des and links that can b e sim
ulated A crucial asp ect of making sim ulations
b oth practical and v aluable is to pro vide the abil
ityto v ary the lev el of abstraction in b oth the
analysis of the data and in the sim ulation itself
Abstraction tec hniques will allo w users to iden
tify relev an t phenomena using highlev el sim ula
tions and then use detailed sim ulations to study
the phenomena more extensiv ely Key to this
approac h is crossv alidation of detailed and high
lev el sim ulations
Visualization tec hniques Man y relev an t net w ork
phenomena remain in visible when only aggregate
statistics are collected Visualization tec hniques
are crucial in enabling users to iden tify in ter
esting asp ects of the sim ulation The visualiza
tion to ols w e are dev eloping ha v e pro v en to be
extremely useful for debugging sim ulations and
understanding proto col b eha vior as w ell as for
helping users frame their design questions
Em ulation in terface Ns includes an em ulation in
terface whic h pro vides a gatew aybet w een real
w orld net w ork no des and the sim ulator Em ula
tion enables a running sim ulator to send pac k ets
to and receiv e pac k ets from op erational net
w ork hosts and routers This allo ws more thor
ough testing of implemen tations prior to deplo y
men t and will allo wsim ulations to b e driv en b y
more accurate net w ork inputs
The purp ose of this pap er is t w ofold First w e
w an t to describ e the kinds of problems net w ork re
searc hers use sim ulation to study and in doing so
demonstrate some of the functionalit y in ns Second
wew an t to presen t the tec hniques w e are dev eloping
to address the c hallenges of using sim ulation to study
these algorithms and proto cols in large heterogeneous
net w orks
W e b egin with an o v erview of the motiv ations and
design decisions resulting in the currentsoft w are ar
c hitecture of the sim ulation en vironmen t W e then
pro ceed to illustrate a wide range of net w ork researc h
areas in whichin v estigators ha veemplo y ed ns in pro
ducing imp ortan t researc h results W e touc h up on
scenario generation scaling and visualization eac h
activ e areas of dev elopmen t W e conclude with a brief
discussion of prior w ork in net w ork sim ulation and
c hallenges for the future
Comp osable sim ulation
framew ork
The fundamen tal abstraction pro vided b y the ns soft
w are arc hitecture is programmable comp osabilit y
In this mo del sim ulation congurations are expressed
as a pr o gr am rather than as a static conguration
or through a sc hematic capture system A sim ula
tion program c omp oses ob jects dynamically in to arbi
trary congurations to eect a sim ulation congura
tion By adopting a full edged programming mo del
for sim ulation conguration the exp erimen talist is
free to extend the sim ulator with new primitiv es or
program in dynamic sim ulation ev en t handlers
that in teract with a running sim ulation to c hange its
course as desired
Rather than adopt a single programming language
that denes a monolithic sim ulation en vironmen t w e

ha v e found that dieren tsim ulation functions require
dieren t programming mo dels to pro vide adequate
exibilit y without unduly constraining p erformance
In particular tasks lik elo wlev el ev en t pro cessing or
pac k et forw arding through a sim ulated router require
high p erformance and are mo died infrequen tly once
put in to place Th us they are b est serv ed byanim plemen tation expressed in a compiled language lik e
C

On the other hand tasks lik e the dynamic
conguration of proto col ob jects and the sp ecication
and placemen t of trac sources are often iterativ ely
rened and undergo frequen tc hange as the researc h
task unfolds Th us they are b est serv ed b y an im
plemen tation in a exible and in teractiv e scripting
language lik e Tcl
T o this end ns exploits a split pr o gr amming mo del where the sim ulation k ernelie the core set of high
p erformance sim ulation primitiv esis implemen ted
in a compiled language C
while sim ulations are
dened congured and con trolled b y writing an ns
sim ulation program expressed in the Tcl scripting
language This approac h is a b o on to longterm pro
ductivit y b ecause it cleanly separates the burden of
sim ulator design main tenance extension and debug
ging from the goal of sim ulation itselfthe actual
researc h exp erimen tsb y pro viding the sim ulation
programmer with an easy to use recongurable and
programmable sim ulation en vironmen t Moreo v er it
encourages a programming st yle that leads to an im
p ortan t separation of mec hanism and p olicy core
ob jects that represen t simple and pure op erations are
free of builtin con trol p olicies and seman tics and can
th us b e easily reused
In our split programming mo del negrained sim
ulation ob jects are implemen ted in C
and are
com bined with Tcl scripts to eect more po w erful
higherlev el macroob jects F or example a sim u
lated router is comp osed of dem ultiplexers queues
pac k et sc hedulers and so forth By implemen ting
eac h primitiv e in C
and comp osing them using
Tcl a range of routers can be sim ulated faithfully W e can string together the lo wlev el dem ultiplexers
queues and sc hedulers to mo del an IP router p er
haps with m ulticast forw arding supp ort or instead
arrange them in to a conguration that mo dels a high
sp eed switc h with a new sc heduling discipline In the
latter case the switc h could b e easily extended with
proto col agen ts implemen ted en tirely in Tcl that
mo deled an exp erimen tal signaling proto col P er
formance also guides our split programming mo del
Lo wlev el ev en tlev el op erations lik e route lo okups
pac k et forw arding and TCP proto col pro cessing are
implemen ted in C
while highlev el con trol op era
tions lik e aggregate statistics collection mo deling of
link failures route c hanges and lo wrate con trol pro
to cols are implemen ted in Tcl Careful design is nec
essary to obtain a desirable tradeo b et w een p erfor
mance and exibilit y and the division often migrates
during the course of a proto col in v estigation
This comp osable macroob ject mo del is naturally
expressed using ob jectorien ted design but unfortu
natelyatthe time w e designed ns Tcl did not pro
vide supp ort for ob jectorien ted programming con
structs nor did it pro vide v ery eectiv e programming
constructs for building reusable mo dules Th us w e
adopted an ob jectorien ted extension of Tcl Of the
sev eral Tcl ob ject extensions a v ailable at the time w e
c hose the Ob ject Tcl OTcl system from MIT b ecause it required no c hanges to the Tcl core and
had a particularly eleganty et simple design W e fur
ther adopted a simple extension to OTcl called TclCL
for Tcl with classes that pro vides ob ject scaold
ing bet w een C
and OTcl and thereb y allo ws an
ob jects implemen tation to be split across the t w o
languages in congruence with our split programming
mo del
With the OTcl programming mo del in place eac h
macroob ject b ecomes an OTcl class and its com
plexit y is hidden b ehind a simpletouse set of ob ject
metho ds Moreo v er macroob jects can b e em b edded
within other macroob jects leading to a hierarc hical
arc hitecture that supp orts m ultiple lev els of abstrac
tion As an example highlev el ob jects migh t repre
sentanen tire net w ork top ology and set of w orkloads
while the lo wlev el ob jects representcomponen ts lik e
dem ultiplexers and queues As a result the sim ula
tion designer is free to op erate at a high lev el eg
b y simply creating and conguring existing macro
ob jects at a middle lev el eg b y mo difying the b e ha vior of an existing macroob ject in a deriv ed sub
class or at a lo wlev el of abstraction eg b yin tro
ducing new macroob jects or split ob jects in to the ns

core Finally class hierarc hies allo w users to sp e
cialize implemen tations at an y one of these lev els for
example extending a v anilla TCP class to imple
men t TCP Reno The net eect is that sim ulation
users can implemen t their sim ulation at the highest
lev el of abstraction that supp orts the lev el of exi
bilit y required th us minimizing exp osure to and the
burden asso ciated with unnecessary details
Proto col Design and
In teraction
The ns sim ulator has a large p opulation of users and
it and its ancestors ha v e b een used in man y successful
researc h eorts
In this section w e review four areas
of researc h activit y that ha v e used ns and describ e
existing functionalit y of the sim ulator These areas
include TCP congestion con trol queue managemen t
m ulticast routing and reliable m ulticast transp ort
This is merely a sample of researc h using ns and is b y
no means an exhaustiv e list In addition w e describ e
a recen tly dev elop ed em ulation in terface in ns
TCP Congestion Con trol
The ns sim ulator and its ancestors ha v e b een used
to study a n um ber of algorithmic c hanges to the
TCP proto col In v estigations of TCP error and con
gestion con trol algorithms ha v e led to the dev elop
men t of sev eral new algorithms for TCP including
selectiveac kno wledgmen ts forw ard ac kno wl
edgmen ts and explicit congestion notication
ECN Sim ulation studies using ns rev eal ho w
common TCP algorithms p erform p o orly when sub
jected to mo derate to hea vy pac k et loss and ho wal gorithms in end no des suc h as New Reno and selec
tiv e rep eat can help to impro v e b eha vior signican tly
under suc h conditions In addition TCP can also
be mo died to a v oid pac k et bursts Suc h bursts
whic h can ha v e harmful eects on the net w ork often
result after the successful reconstruction of the data
stream follo wing a series of pac k et losses or after
Ns is deriv ed from REAL whic h is deriv ed from
NEST The currentv ersion of ns v ersion is a v ailable
at h ttpwwwmashcsb erk eley eduns
idle connections The a v ailabilit y of a public
domain sim ulator including sev eral v arian ts of TCP
greatly facilitates the ev aluation of prop osed proto col
enhancemen ts
Queue Managemen t and
Sc heduling P olicies
Ns has also b een used to study routerbased algo
rithms Random Early Detection RED queue man
agement whic hw as dev elop ed on one of the an
cestors to ns is supp orted as a standard queue man
agemen t tec hnique in ns RED queue managemen t
reacts to congestion prior to o v erload RED has b een
sho wn to ac hiev e signican t impro v emen ts o v er tradi
tional FIF O droptail queue managemen t strate
gies b y allo wing for some burst y beha vior but also
pro viding implicit signaling to w ellb eha v ed net w ork
o ws prior to buer exhaustion In addition its ran
dom approac h helps to a v oid undesirable phase eects
ie throughput bias for particular connections in
net w orks of droptail queues
Other in v estigations in to router trac manage
men t emplo ying the ns sim ulator include Class Based
Queueing CBQ a tec hnique in whic h pac k ets
are treated as mem b ers of classes Classes are as
signed a maxim um bandwidth allo cation and prior
itylev el relativ e to other classes The CBQ sc heduler
limits classes to their assigned bandwidth unless some
other classes are using less than their allo cated band
width in whic h case bandwidth b orro wing ma ybe
allo w ed
In eac h of these cases sim ulations pro vided a con
v enien t rep eatable en vironmen t in whic h to study
and compare algorithms F urthermore the existence
of a platform with whic h to study these new algo
rithms has facilitated a m uc h wider understanding of
their b eha vior thereb y lo w ering barriers to deplo y
men t in op erational net w orks
Finally it should b e noted that alternativ e queue
managemen t and sc heduling p olicies are useful for
other kinds of sim ulation studies as w ell F or ex
ample ns w as used in a study of the eects of service
priorit y on the p erformance exp erienced b y adaptiv e
audio applications A comparison of uniform and
priorit y dropping mec hanisms on the p erformance of

la y ered video also used ns extensiv ely In both
these cases while the sc heduling and dropping algo
rithms w ere themselv es not the fo cus of the study they w ere a critical part of the infrastructure needed
to carry out the sim ulation exp erimen ts
Multicast routing
More recen tly ns has b een used in the study of m ul
ticast routing proto cols Sev eral m ulticast routing
proto cols whic h establish distribution trees for de
liv ering datagrams from a single sender to all the
mem bers of a m ulticast group ha v e been prop osed
for the In ternet These proto cols can be classied
as either broadcastandprune or explicit join proto
cols In the former whic h include D VMRP and
PIMDM a m ulticast pac k et is transmitted to all
leaf subnet w orks in a distribution tree ro oted a the
source Leaf subnet w orks with no lo cal mem bers of
the group send prune messages to w ards the source of
the pac k et This prev en ts future pac k ets from b eing
transmitted to these subnet w orks and limits pac k et
distribution to those subnet w orks with group mem
bers In con trast in explicit join proto cols suc h as
CBT or PIMSM routers send hopb yhop
join messages for the groups for whic h they ha velo cal mem bers These con trol messages build forw ard
ing state in routers and are sen t upstream to w ards
the source to establish a distribution tree Ns cur
ren tly has implemen tations of both broadcastand
prune based on PIMDM and explicit join based
on PIMSM proto cols
These proto cols pro vide an example of the bene ts of the split programming mo del describ ed in Sec
tion Lo wlev el mec hanisms for forwar ding m ul
ticast pac k ets common to all m ulticast proto cols
are implemen ted in C
while the routing proto cols
themselv es are implemen ted in OTcl This allo ws
rapid design and exp erimen tation with the routing
proto col implemen tation with minimal p erformance
degradation as only con trol messages are sub jected
to the higher o v erhead of the in terpreted language
The ab o v emen tioned proto cols pro vide in ternet
w ork forw arding of m ulticast pac k ets Multicast is
also used in LAN en vironmen ts to exc hange rout
ing up dates and con trol messages or to b o otstrap
proto col mec hanisms Ns has b een extended
to supp ort m ultiaccess links connecting more than
t w o no des F acilities to supp ort pac k et tracing for
oline analysis and selectiv eloss ha vebeen imple
men ted Other than pac k et replication and forw ard
ing all the supp ort for LAN m ulticast w as done in
OTcl to pro vide exibilit y This facilit y has b een used
to study m ulticast routing proto col robustness in the
presence of con trolpac k et loss and no de failure In
particular this w ork iden tied sev eral pathological
cases in PIMSM and led to corresp onding xes to
the proto col sp ecication
The existence of m ulticast routing proto cols in ns
is not just of v alue to researc hers in terested in the de
sign and ev aluation of suc h proto cols Rather other
exp erimen ters dep end on m ulticast as necessary in
frastructure for their sim ulations W e next discuss
examples of suchsim ulations
Multicast T ransp ort
Asp ects of the man ytoman y comm unication
paradigm of m ulticast forw arding complicate the
design of transp ort proto cols These include the
anon ymit y of mem b ership mem bership dynamics
and the heterogeneit y of the mem bers Ns through
infrastructure consisting of man y dieren t link and
net w ork la y ers as w ell as unicast and m ulticast
routing proto cols pro vides a unique en vironmen t
to dev elop m ulticast transp ort proto cols In this
section w e describ e some of the w ork on reliable
transp ort congestion con trol and application
dev elopmentin whichsim ulation has b een used
Reliable T ransp ort Scalable Reliable Multicast
SRM w as designed originally for realtime
whiteb oard applications It uses a NA CKbased pro
to col to ac hiev e reliabilit y Receiv ers detecting a loss
m ulticast a negativeac kno wledgemen t to the group
These negativ e ac kno wledgemen ts are m ulticast in
order to suppress duplicates In addition receiv ers
dela y their negativeac kno wledgemen ts for a random
time to a v oid in undating the net w ork with sync hro
nized requests and retransmissions While the orig
inal sim ulations of SRM w ere done in a standalone
sim ulation to ol an SRM implemen tation has b een

added to ns This is v aluable to other researc hers
in v estigating reliable m ulticast transp ort F or ex
ample Routing P olicy Multicast RPM is an
applicationsp ecic SRMlik e proto col for the reli
able deliv ery of routing p olicy ob jects These ob jects
are large but there is less of a time constrainton de liv ery than with SRM Giv en the common mec hanism
shared b et w een SRM and RPM RPM w as easily im
plemen ted b y deriving a new class in the ns ob ject
framew ork This implemen tation w as used to ev al
uate rapidly dieren t timer mec hanisms and deter
mine the optimal parameter settings for RPM
The p erio dic session messages in SRM lead to
bandwidth o v erhead that impacts proto col scalabilit y
when the group mem b ership is large Scalable Ses
sion Messages SSM algorithms use hierarc h y
mec hanisms to reduce this o v erhead ns w as used to
in v estigate the dieren t mec hanisms b y whic h rep
resen tativ es are c hosen and to quan tify the scaling
b enets that can b e ac hiev ed and the impact on the
proto cols loss detection and reco v ery mec hanisms
As stated ab o v e the existence of a publicdomain
sim ulator with an extensiv e set of proto col mo dules
facilitates comparison of researc h results As an ex
ample Hanle used ns to compared the Multi
cast File T ransfer Proto col MFTP a proto
col sp ecically designed for bulk data transfer to
SRM under dieren tnet w ork conditions in a v ariet y
of dieren t top ologies By sub jecting the proto cols
to iden tical test conditions their b eha vior could be
compared across a range of conditions
Congestion Con trol Multicast congestion con trol
is an activ e area of in v estigation The c hallenge is to
manage the feedbac k from a large set of homogeneous
receiv ers in a timely and scalable manner Receiv er
driv en La y ered Multicast RLM is an example
of a m ulticast congestion con trol proto col for la y ered
video transmission Muc h of the design of RLM used
ns joinexp erimen t heuristics congestion measure
men ts and other features w ere protot yp ed through
OTcl scripting
DeLucia prop osed a represen tativ e based con
gestion con trol algorithm for m ulticast bulk data
transfer applications ns w as used to v erify correct
b eha vior in the face of congestion and ev aluate the
p erformance of the proto col in the presence of com
p eting trac in the net w ork W ork is no w fo cused on
applying the congestion con trol sc hemes to Multicast
Dissemination Proto col MDP This has b een
accomplished b y retrotting an existing implemen ta
tion of MDP in to ns enabling study of the in tegrated
proto col and congestion con trol sc heme in the sim u
lator
Application Realtime T ransp ort Proto col
R TP is designed for unreliable but timely
deliv ery of datagrams for realtime audio or video
or other m ultimedia or realtime applications ns
implemen ts the con trol asp ects of R TP Realtime
T ransp ort Con trol Proto colR TCP in OTcl The
implemen tation is useful for further exp erimen ta
tion in the dev elopmen t of other transp ort and
application proto cols
Reddy prop osed a m ulticastbased application
of a net w ork of dynamically adaptiv e measuremen t
serv ers to gather lo calized information ab out the net
w ork ns is b eing used in the design of the proto col
to determine the scop e of eac h measuremen t serv er
so that ev ery no de or link is monitored byaserv er
and to mak e the algorithm robust in the face of the
addition deletion and failure of serv ers
McCanne et al are usingnstoin v estigate the
minimal set of mec hanisms that a router could imple
men t to simplify the design and impro v e the p erfor
mance of m ulticast transp ort proto cols One exam
ple of this is subtree m ulticast or sub cast where
the retransmit of a lost data pac k et is only sen t to
the subtree that exp eriences the loss
Sidebar Dynamics
Ns supp orts the study of net w ork dynamics That is
ho w are proto cols aected b y links and no des that fail
and reco v er This supp ort includes dynamic routing
proto cols and mo dels of link failures Study of net
w ork dynamics is imp ortan t to c haracterize the be ha vior of endtoend proto cols in the con text of a v a
rietyofnet w ork anomalies including route apping
routing lo ops and net w ork partitions W e ha v e

used these approac hes to study TCP and m ulticast
transp ort proto cols One eect of top ology c hange is the reordering of
pac k ets in transit Suchin terlea ving of ac kno wledge
men ts or data pac k ets can ha v e harmful consequences
for a TCP session and the net w ork In particular
the sender can see a sudden and large increase in the
amountof ac kno wledged data The sender resp onds
b y op ening its send windo w and sends a large burst
of pac k ets bac k to bac k The resulting congestion
reduces the throughput for the session Weha v e ob
serv ed these eects through sim ulation studies using
ns itremainstofuturew ork to recreate the eect of
pac k et in terlea ving in an op erational net w ork
T op ology c hange has ev en greater impact on m ul
ticast transp ort proto cols Our analysis of the timer
mec hanisms in SRM rev ealed the need for accurate
and timely distance estimation b y the group mem
b ers collecting suc h estimates in the face of dynami
cally c hanging top ology is problematic W e also iden
tied some undesirable op erating regions of the pro
to col Suchc haracterizations are useful both in the
design and ev aluation of the proto col and in enhanc
ing its manageabilit y when deplo y ed in op erational
net w orks
The exp erience gained studying TCP and c harac
terizing the b eha vior of SRM during top ology c hange
has con tributed to an ev olving systematic metho dol
ogy for the study of a proto col b eha vior under a range
of net w ork conditions including pac k et loss routing
transien ts and no de and link failure
Em ulation
Ns supp orts an emulation facilit y allo wing sim ula
tionstoin teract with actual net w ork trac In com
bination with the sim ulators tracing and visualiza
tion facilities em ulation pro vides a p o w erful analysis
to ol for ev aluating the dynamic b eha vior of proto cols
and their implemen tations in end systems An em ula
tion scenario is constructed b y placing the sim ulator
as an in termediate no de or end no de along an end
toend net w ork path as illustrated in Figure The
sim ulator con tains a sim ulated net w ork and passes
liv e net w ork trac through the sim ulation sub ject
ing it to the dynamics of the sim ulated net w ork The
Packet Flows
Packet Capture and Generation Interface
Local Operating System
NS Simulator
(Emulation Mode)
Simulated Network
Figure Em ulation liv e net w ork trac passes
through sim ulated top ology and crosstrac
sim ulators sc heduler is sync hronized with realtime
allo wing the sim ulated net w ork to em ulate its real
w orld equiv alen t so long as the sim ulated net w ork
can k eep pace with the real w orld ev en ts
Em ulation is useful bey ond con v en tional sim ula
tion in ev aluating b oth end system and net w ork ele
mentbeha vior With em ulation end system proto col
implemen tations can b e sub jected to pac k et dynam
ics eg drops reordering dela ys that are dicult
to repro duce reliably in a livenet w ork F urthermore
b y capturing trac traces of liv e trac injected in to
the sim ulation en vironmen t visualization to ols ma y
b e emplo y ed to ev aluate the end systems dynamic re
sp onses In the con v erse situation net w ork elemen t
b eha vior eg a queueing or pac k et sc heduling disci
pline ma ybe ev aluated in relation to liv e trac gen
erated b y realw orld end stations Suc h sim ulations
are useful in predicting undesirable net w ork elemen t
b eha vior prior to deplo ymen t in livenet w orks
The ns em ulation facilit y is curren tly under de
v elopmen t but an exp erimen tal v ersion has already
pro v en useful in diagnosing errors in proto col imple
men tation F or example researc hers at UC Berk e
ley ha vedev elop ed a shared white b oard application
using a v ersion of the SRM proto col supp orted in
the MASH to olkit The sim ulator is placed b e
t w een groups of liv e end stations comm unicating us
ing SRM Multicast trac passing bet w een groups

m ust tra v erse the sim ulator and is sub ject to the
dynamics of its sim ulated net w ork Visualization of
traces tak en within the sim ulation en vironmen t re
v eals end station retransmissions triggered bypac k
ets dropp ed or dela y ed within the sim ulated net w ork
This helps to pinp oin t timedep enden t b eha viors of
adaptiv e proto cols whic h are v ery dicult to diag
nose otherwise
Other examples
Although w e ha v e fo cused primarily on the study
of transp ort and routerlev el net w ork issues ns has
been used for studies of other net w ork la y ers and
problems F or example F ab er uses ns to examine ac
tiv e approac hes to congestion con trol P ortions of
ns also ha v e b een used to study m ulticast address al
lo cation new rate based congestion con trol algo
rithms and selforganizing clustering algorithms
applied to net w ork monitoring and reliable m ulticast
session message aggregation
Scenario generation
In general a sim ulation scenario is dened b y the
comp onen ts and parameters that comprise the sim u
lation In ns scenarios are represen ted b y sim ulation
scripts whic h describ e the network top olo gy includ
ing the ph ysical in terconnects b et weennodes andthe
static c haracteristics of links and no des tr ac mo dels
for b oth unicast and m ulticast senders and test gen
er ationwhic h creates ev en ts suchas m ulticast group
distribution receiv ers joining and lea ving and net
w ork dynamics no de and link failures designed to
stress test an implemen tation
T op ology
The top ology for a sim ulation is sp ecied as part
of the sim ulation script F or small sim ulations the
user will usually sp ecify the top ology b y hand
F or larger top ologies generating top ologies man ually
is less practical Hence top ology generation to ols
whic h create top ologies according to a set of user
sp ecied parameters are often used Our goal here is
to pro vide users with a library of top ology generation
to ols to use with ns In addition w eare accum ulating
a library of sample top ologies for use in sim ulations
Rather than recreate previous w ork w eha vec hosen
to lev erage the w ork of others on automatic top ology
generation F or the t w o top ology generation to ols
describ ed b elo w w eha v e written and made a v ailable
con v ersion programs that allo w the generated top olo
gies to b e used in ns sim ulations
The Georgia T ec h In ternet w ork T op ology Mo dels
GTITM soft w are pac k age can create
at random net w orks using av ariet y of edge distri
bution mo dels including pure random exp onen tial
lo calit y sev eral v ariations of W axmans mo del and the DoarLeslie mo del Giv en the size of a
grid and the n um b er of no des desired GTITM ran
domly places no des on the grid and connects them
according to the probabilitygiv en b y the edge mo del
The GTITM soft w are pac k age can also create dif
feren t t yp es of hierarc hical net w orks It can create
am ultilev el hierarc hyb y rst creating the toplev el
net w ork using one of the at random mo dels and
then recursiv ely replacing eac h no de in the curren t
lev el with a connected graph Edges bet w een lev
els are resolv ed b y randomly selecting a no de within
eac h replacemen t graph GTITM can also create
transitstub hierarc hies lik e those found in the In ter
net P arameters con trol the a v erage n umberoftran sit domains the a v erage n um b er of stubs p er transit
and the a v erage size of transit and stub domains
Doar has written tiers to create three
lev el hierarc hical top ologies similar to the transit
stub GTITM top ologies In tiers the three lev
els corresp ond to widearea metrop olitanarea and
lo calarea net w orks A net w ork at an y lev el is cre
ated b y randomly placing no des within a grid and
then connecting them with a minim um spanning tree
Edges b et w een lev els are created b y attac hing a c hild
net w ork to a paren t net w ork for example b y attac h
ing a lo calarea net w ork to a metrop olitanarea net
w ork T op ology generation is con trolled b y sp ecifying
the n um ber and size of net w orks at eac h lev el plus
the connectivitybet w een lev els
The k ey c hallenge in top ology generation is com
ing up with top ologies that em b o dy relev antc harac
teristics of real net w orks Once this is done the ns

framew ork easily allo ws sim ulation of an y top ology
that is generated Weha v e dev elop ed an API to the
GTITM at random and transitstub top ologies Ns
sim ulation scripts are automatically created byasim ple format con v ersion program W e plan to extend
the API for other t yp es of top ologies and top ology
generation pac k ages Similarly actual maps of parts
of the In ternet top ology can also b e used as sim ula
tion input
T rac Mo dels
Ns pro vides a wide v ariet y of source mo dels all of
whic h are congurable in Tcl allo wing exible cre
ation and parameterization F or sim ulations of TCP b oth bulk data and in teractiv e sources are a v ailable
The former can mo del an FTP application while the
latter based in part on a mo del dev elop ed from traf
c traces mo dels T elnetlik e applications T o
sim ulate w eb trac a trac generator based on the
mo del describ ed in has also b een implemen ted in
ns Other source mo dels are a v ailable for nono w
con trolled applications These include a constan t bit
rate source ono sources using either exp onen tial
or P areto distribution the latter useful in generating
selfsimilar trac and a source that gener
ates trac from a trace le The comp osable frame
w ork of ns mak es adding new trac mo dels easy and
allo ws construction of comp ound mo dels out of the
individual ones F or example in sim ulations of RLM
am ultila y ered video source w as created bycom bin
ing sev eral CBR streams A similar approachw as
used to incorp orate correlations of burstiness across
la y ers in another study in v olving la y ered video Ns pro vides an extensiv e set of mo dels for indi
vidual trac sources Ho w ev er in creating a sim
ulation scenario a net w ork researc her is often more
in terested in bac kground trac with desired c harac
teristics ie aggregate bandwidth burstiness self
similarit y etc Recen tlyw eha v e dev elop ed an API
to help users easily create random bac kground traf
c with v arious c haracteristics unicast or m ulticast
and with v arious distributions W e are con tin uing
to dev elop this API
T est Generation
Cho osing an appropriate set of test conditions for a
sim ulation exp erimen t is often straigh tforw ard F or
example the p erformance of an algorithm suc h as
RED can be ev aluated b y using a reasonable set of
parameter v alues as inputs to the sim ulation Ev al
uating the correctness of a proto col on the other
can be a m uc h more daun ting task W e dev el
op ed a framew ork for S ystematic T esting of Proto col
R obustness b y E v aluation of S yn thesized S cenarios
STRESS in order to reduce the eort
needed to iden tify pathological cases of proto col b e
ha vior As the name implies this framew ork in te
grates systematic syn thesis of test scenarios with the
VINT sim ulation en vironmentof ns W e are in the
pro cess of dev eloping automatic test generation al
gorithms for m ulticast proto cols Curren tly w e are
comparing three metho ds for testing m ulticast rout
ing robustness in the presence of selectiv e message
loss on a LAN W e call these metho ds a heuristic
test generation HTG b faultindep enden t test gen
eration FITG and c faultorien ted test generation
F OTG
HTG uses domainsp ecic heuristics and top ologi
cal equiv alence relations to limit the n um b er of sim
ulated scenarios Sim ulation of these scenarios in ns
is then conducted to explore the b eha vior of the pro
to col in the presence of message loss This metho d
do es not require a formal mo del of the proto col
In con trast FITG and F OTG pro cess a nite state
mac hine mo del of the proto col Using a forw ard
searc h tec hnique the FITG metho d in v estigates an
equiv alen t subset of the proto col state space for a
giv en top ology to generate the scenario ev en ts lead
ing to erroneous beha vior F OTG starts from the
target message and syn thesizes a top ology necessary
to trigger and be aected b y this message Then
using a bac kw ard searc h tec hnique it generates sce
nario ev en ts leading to proto col error
These metho ds w ere applied to m ulticast routing
proto col studies in ns Sev eral design errors w ere dis
co v ered and corrected with the aid of STRESS the
detailed results are presen ted in F uture w ork in this area will consider the eect
of a wider range of of net w ork failures on m ulticast

routing W e will also in v estigate systematic metho ds
for p erformance ev aluation and sensitivit y analysis of
endtoend proto cols suc has m ulticast transp ort
In addition w e plan to use the em ulation in terface
in ns to conduct systematic conformance testing and
p erformance proling of actual proto col implemen ta
tions
Scaling and Sim ulation
Abstraction
Ns has sho wn its utilit y in a n um ber of researc h
pro jects some of whic hw ere describ ed in Section Extending this success to other researc h problems re
quires the sim ulator to scale to thousands of no des
and links far b ey ond nss original design goals Cus
tom sim ulators built for a giv en problem can tailor
sim ulation abstraction to the problem at hand but
oftenomitman y of the services presen t in a general
purp ose sim ulator lik e ns for example tracing and
debugging supp ort and existing loss trac mo dels
and dynamic top ologies More imp ortan tlyit is of ten dicult to v alidate these sim ulators against more
detailed sim ulations and realit y In ns w e are w ork
ing on approac hes whichallo w sim ulations to abstract
a w a y unnecessary details while still using general ser
vices By supp orting an adjustable lev el of ab
straction w e will allo w users to select abstract mo dels
when needed but also fo cus in on the details to in
v estigate in teresting phenomena
W e are taking t w o complemen tary approac hes to
scaling ns carefully tuning the implemen tation and
allo wing the user to abstract out details unnecessary
for some sim ulations The former impro v es the e
ciency of the sim ulator without c hanging the sim u
lation abstraction The latter c hanges asp ects of the
sim ulated mo del to ac hiev e p erformance gains As an
example implemen tation impro v ementweha v e made
c hanges to our binding mec hanism b et w een C and
Tcl to impro v e memory consumption when there are
large n um b ers of shared ob jects Use of cen tralized
route computation is an example abstraction Al
though it pro duces sligh tly dieren t transien t b eha v
ior for man y sim ulations these details are unimp or
tan t while sa vings in time and memory is These
approac hes often in terpla y for example our session
level sim ulator abstracts a w a y crosstrac net w ork
in terference and uses a v ery ligh t w eightnode and link
represen tation an implemen tation c hange
The cost of abstraction is sim ulation accuracy The
degree to whic h accuracy is sacriced and the im
pact of this sacrice on the v alidit y of the results
v aries greatly from mo del to mo del F or example
while the details of a particular medias approac hto
segmen tation and reassem bly are imp ortan t for LAN
sim ulations they can b e reected adequately in the
links pac k et loss loss rate for higherlev el W AN sim u
lations An analysis of ho w abstractions c hange sim
ulation results accompanies our dev elopmen t of new
abstractions An analysis of SRM p erformance across
detailed and sessionlev el sim ulations suggests that
while individual SRM ev en ts do v arya v erage aggre
gate b eha vior c hanges b y only a few p ercen t in the
cases w e examined
Abstraction tec hniques that impro v e nss abilit y
to run large sim ulations is an area of ongoing eort
P articularly promising among the tec hniques w e are
pursuing is t w ophase top ology generation in whic h
a large top ology is generated and p opulated with a
few agen ts and then replaced with a more abstract
but equiv alen t top ology through no de aggregation
W e are also dev eloping h ybrid abstractions where one
part of the sim ulator runs in detailed mo de and an
other uses sessionlev el mo de or some other appropri
ate abstraction
In addition to enabling large sim ulations data
analysis in large sim ulations is an imp ortan t issue
Hierarc hical visualization to ols are needed for large
net w orks
Sidebar What is Scaling
In the abstract scaling a sim ulator is running bigger
sim ulations But sim ulations ha veman y dimensions
of big large n um b ers of no des links senders and
receiv ers m ulticast groups m ulticast group mem
b ers and pac k etsintransit as w ell as a large amoun t
It should b e noted that parallel sim ulation on large m ul
tipro cessing hardw are can pro vide a complemen tary tec hnique
to abstraction when additional p erformance gains are needed

of data collected A sim ulator that supp orts si
m ultaneous TCP o ws mayw ork w ell with a Mb s
LANst yle sending rate but ma y consume m uc h more
memory when sim ulating Mb s crosscoun try traf
c since man y more pac k ets are concurren tly in
igh t
Pushing these dimensions of scaling can tax a com
puters memory CPU and I O resources Although
w e feared our use of Tcl migh t mak e CPU usage a
b ottlenec k in our sim ulations weha v e found memory
to be the most frequen t problem It is understand
ably dicult for a single mac hine to k eep trac k of
most of the net w ork state for thousands of sim ulated
hosts
Tomak e matters w orse man y researc hers are cur
ren tly in v estigating the prop erties of m ulticast pro
to cols b oth routing and reliable transp ort when
there are large n um b er of mem b ers p er group Mem
ory required for these proto cols gro ws O n
as the
n um b er of group participan ts increases so a detailed
sim ulation of v ery large groups is often imp ossible
Our sessionlev el abstraction for m ulticast abstracts
a w a y the details of crosstrac in terference allo w
ing a m uc h more compact link and no de represen
tation This allo w ed sim ulations of t wice as man y
group mem bers with the same amoun t of memory
Figure Visualization
Usually the output of ns is a trace le consisting of
pac k et ev en ts and parameters These data are di
cult to in terpret b y direct insp ection so graphs and
aggregate statistics are often emplo y ed for example
time sequence n um ber plots common in TCP stud
ies While these approac hes are helpful they can
miss in teresting details of the sim ulation
Visualization has b een found to b e extremely p o w
erful in helping understand c haracteristics of large
complex data sets The VINT pro ject is exploring
this tec hnique through nam the n et w ork an imator
Lik e ns nam is built with a C
core and supp ort
of OTcl Tk whic h allo ws nam to b e extensible and
easily customized Using trace les generated b yns nam pro vides animation insp ection and sync hro
0
20
40
60
80
100
120
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
memory usage (in MB)
number of group members
detailed
session
Figure Sessionlev el abstraction allo ws substan
tially larger n um b ers of m ulticast group mem bers in
the same amoun t of memory nized viewbased drilldo wn analysis of the traces
Animation The animation to ol displa ys the net
w ork top ology animates pac k et transmission across
links and displa ys no de and link dynamics The unit
of animation is the event whic h is read from the
trace le In general there are four t yp es of ev en ts
No des and links dene the top ology of the sim ula
tion Packetr elate d events denote pac k ets lea ving
or reac hing a no de en tering or lea ving a queue and
being dropp ed Pr oto c olr elate d events describ e the
p osition of proto col instances agents and their in
ternal states variables A nnotations serv e as indices
to the trace le for easier bro wsing
Nam denes dieren t animation mo dels for dier
en t ev en ts P ac k ets are displa y ed as lled colored
rectangles with a tap ered end to indicate the direc
tion they tra v el Dropp ed pac k ets are displa y ed as
rolling squares falling out of the windo w Queues are
sho wn as stac k ed pac k ets on a link No des and links
can b e sho wn with dieren t shap es and colors to high
ligh t sim ulation ev en ts F or example link color can
indicate its status up or do wn or no de color can
indicate mem b ership in a m ulticast group Agen ts
data senders and receiv ers on a no de are displa y ed

Figure Nam screen snapshot
as small lab eled rectangles attac hed to no des These
visual represen tations mak e it easy to insp ect data
see b elo w
Nam pro vides t wow a ys to bro wse the traces First
a time slider is pro vided to start animation from an y
poin t in the trace le Second annotations are pro
vided as indices to the trace les They are displa y ed
in the b ottom pane in nam main windo w Figure
Double clic king on an y of them brings nam to
the time indicated b y that annotation Annotations
ma y b e edited in teractiv ely during animation Nam
pro vides V CRlik e buttons eg Pla yF ast F orw ard
Rewind etc to con trol the animation
Sometimes it is useful to compare sev eral anima
tions sideb yside for example when one is studying
b eha viors of the SRM proto col with dieren tpa rameters in the same scenario In order to supp ort
this sev eral instances of nam running on the same
mac hine can con trol eac h others animations It is as
sumed that only one nam instance will b e in con trol
at one time therefore no concurrency con trol mec h
anism is pro vided
Insp ection The animation to ol only displa ys some
asp ects of the sim ulation traces m uc h of the infor
mation suc h as pac k et headers or proto col state v ari
ables in the traces cannot b e sho wn directly bythe
animation to ol F or large data sets it maybe b en
ecial to ha v e m ultiple views emphasizing dieren t
asp ects of the data The insp ection to ol and statis
tics to ol of nam are designed with this in mind
When an in teresting p oin t is reac hed during an an
imation it is often desirable to examine certain data
suc h as the source and destination of a pac k et state
v ariables in an agen t and link parameters in more
detail The insp ection to ol pro vides t wow a ys to do
this First clic king on an y of the displa y ed ob jects
eg pac k ets and agen ts will bring out a panel sho w
ing additional information ab out that ob ject Sec
ond monitors can be used to con tin uously displa y
state ab out a link or anode Monitors displayand
up date all a v ailable information ab out the ob ject All
monitors are displa y ed in a pane in nams main win
do w as sho wn in Figure Nam pro vides m ultiple views of the same top ology

the lo w er smaller windo w in Figure th us allo wing
sim ultaneous animation and insp ection of dieren t
parts of the top ology All views are zo omable for
insp ection of details They are useful for examining
in detail m ultiple ob jects in a large top ology Sync hronized viewbased drilldo wn analysis
Besides examining pac k et o ws and no de link dy
namics wew ould also lik e to ha v e highlev el repre
sen tations or o v erviews of the en tire trace F or
example when w e are studying TCPs b eha vior it is
useful to ha v e TCP sequence n um ber plots in addi
tion to the animations These views are sync hronized
in time time c hange in an y view eg b ymo ving a
time slider or byclic king on an in teresting ev en t will
c hange time in all views The views also enable the
user to disco v er in teresting and useful results byse lecting a subset of trace ev en ts con tin uously zo om
ing in for drilldo wn analysis Eac h trace ev en t is
displa y ed in the same w a y ie color shap e etc
across views to help the user co ordinate ev en ts across
views
Curren tly t w o t yp es of analysis views are sup
p orted The rst displa ys linksp ecic information
Sp ecically utilization and pac k et loss o v er time is
sho wn for the en tire sim ulation see the t w o bars
starting with an arro w in Figure The second kind
of view is a highlev el proto col analysis to ol Figure
sho ws a plotting and analysis to ol for TCP The
windo w in the upp er righ t corner sho ws the n um ber
of activ e TCP sessions in the sim ulation t wointhis
case The windo w b elowit sho ws b oth the sequence
n um b er and ac kno wledgementn um b er as a function
of time for the rst tcp session The lo w er left windo w
sho ws the same plots for the second tcp session The
lo w er rightwindowisazoomed in viewofaportion of
the rst tcp connection The v ertical line inside eac h
view denotes curren t time in the animation A mes
sage view can b e op ened for eac h plot for instance
to sho w the sequence n um b er of a particular pac k et
in the graph Note that the same color is used for
the view of the rst tcp session its zo omed view and
its pac k ets in the animation The tcp sequence plot
is mark ed as dot while the A CK plot is mark ed as a
square Curren tly proto col plotting and analysis is
pro vided only for TCP Similar to ols for other proto
cols suchas SRM are b eing dev elop ed and will be
in tegrated in to nam in the same manner
Net w ork La y out Before animating a sim ulation
trace a la y out for the net w ork top ology m ust b e sp ec
ied This can b e done man ually b y the user or auto
matically b y nam Muchw ork has b een done in auto
matic la y out for arbitrary net w orks Nam
adopts an algorithm based on a springem bedder
mo del for its simplicit y and eciency It assigns
attractiv e forces on all links and repulsiv e forces b e t w een all no des and tries to ac hiev e balance through
iteration Exp erience has sho wn that it w orks v ery
w ell for top ologies with less than no des Man ual
la y out can be sp ecied in a trace le or b y in terac
tiv ely editing an existing la y out
Related W ork
Net w ork Sim ulators Net w ork sim ulation has a
v ery long history Ns itself is deriv ed from REAL whic h is deriv ed from NEST Although wecan not list all relev an t net w ork sim ulators here this sec
tion describ es distinguishing features of net w ork sim
ulators and compares prominen t examples with ns
Sim ulators ha v e widely v arying fo cuses Man y tar
get a sp ecic area of researchin terest suc h as net
w ork t yp e A TM Sim ulator or proto col PIM
SIM Others including ns REAL OPNET INSANE target a wider range of proto cols
The most general pro vide a general sim ulation lan
guage with net w ork proto col libraries for example
Maisie V ery fo cused sim ulators mo del only the
details relev an t to the dev elop er The dierences
bet w een net w orktargeted and general sim ulators is
m uc h less clear
The core of ns and most net w ork sim ulators is a
discrete ev en t pro cessor Sev eral complemen tary ap
proac hes ha v e b een tak en to impro v e accuracy p er
formance or scaling Some sim ulators augmentev en t
pro cessing with analytic mo dels of trac o w or
queueing b eha vior for example OO and uid
net w ork appro ximations for b etter p erformance
or accuracy
P arallel and distributed sim ulation is a second
w a y to impro v e p erformance Sev eral sim ulators
supp ort m ultipro cessors or net w orks of w orkstata
tions Although ns is fo cused only on se
quen tial sim ulation the T eD eort has parallelized
some ns mo dules w e see parallel sim ulation as
complemen tary to abstraction
Abstraction is a nal common approac h to im
pro ving sim ulator p erformance All sim ulators adopt
some lev el of abstraction when c ho osing what to sim
ulate Flo wSim w as the rst net w ork sim ulator to
mak e this tradeo explicit As discussed in Sec
tion ns supp orts sev eral lev els of abstraction
An um b er of dieren tsim ulation in terfaces are p os
sible including programming in a highlev el scripting
language a more traditional systems language or
sometimes b oth Some systems fo cus on allo wing
the same co de to run in sim ulation and a liv e net
w ork for example xSim and Maisie Most
systems augmen t programming with a GUI shell of
some kind Ns pro
vides a splitlev el programming mo del see Section where pac k et pro cessing is done in a systems lan
guage while sim ulation setup is done in a scripting
language Nam pro vides visualization output and w e
plan to add some abilit y to edit top ology Net w ork Em ulation Previous w ork in net w ork
em ulation has included sp ecial purp ose standalone
net w ork em ulators supp orting pac k et dela y Y ans
Hitb ox While these systems mo dify an existing
pac k et stream more sophisticated em ulation systems
suchasSuns P ac k et Shell ha veallo w ed genera
tion of new streams t ypically for proto col testing
By linking a general purp ose sim ulator to liv e net
w ork traces ns promises to accomplish b oth of these
aims
Visualization and animation Likenet w ork sim
ulation visualization has a long history Bec k er
Eic k and Wilks describ ed the SeeNet system for
D visualization of net w ork data They displa y
net w ork trac o v erload and idle capacit y on geo
graphic maps A matrix displayof net w ork o v erload
is pro vided as an alternativ e Animation is supp orted
for insp ecting timev arying c haracteristics of data
Lamm et al uses D displa ytoshoww eb serv er loads
on a geographic map Dieren t trac t yp es are
colored co ded Scullin et al uses Scattercub e Ma
trix metaphor to displayw eb serv er p erformance in a
virtual realityen vironmen t The ab o v e systems
mainly fo cus on highlev el statistics of net w ork data
while nam curren tly fo cuses on pac k etlev el anima
tion and insp ection
Man y researc hers ha vetac kled the problem of visu
alization of complex data An tis et al s SeeData gen
erates D visualization of database structure It
pro vides m ultiple colored views eac h of whic h fo cuses
on a dieren t asp ect of database structure ranging
from abstract o v erview of the en tire structure to de
tailed asso ciations among relations CataCone fo
cuses on searc h and bro wsing of v ery large hierarc hi
cal data It uses ConeT ree for bro wsing and
o v erview of searc h structure and W ebBo ok to
displa y searc h results These systems share a com
mon principle ie m ultiple link ed views are essen
tial in visualizing complex data Nam adopts this
principle It organizes visualization around the main
top ology view from whic h a n um ber of sp ecialized
views maybe deriv ed
Conclusions
The goal of net w ork sim ulation is exploration of new
proto cols and of old proto cols in new en vironmen ts
Distribution of early v ersions of ns has resulted in
b etter understanding of TCP and router queueing
mec hanisms Curren tly ns is seeing widespread use
in the dev elopmen t of reliable m ulticast proto cols
Finallyw e describ ed our curren t and future eorts in
scenario generation and proto col testing sim ulation
scaling through the use of abstraction and impro v ed
visualization to ols By oering an expanding set of
net w ork proto cols suitable across a wider range of
scenarios and scales wehopeto mak e future proto col
dev elopmen t and comparison easier

References
JongSuk Ahn P B Danzig D Estrin and
B Timmerman Hybrid tec hnique for sim ulat
ing high bandwidth dela y computer net w orks In
Pr o c e e dings of the A CM SIGMETRICS pages
! San ta Clara CA USA Ma y A CM
JS Ahn P eter B Danzig Z Liu and L Y an
Ev aluation of TCP V egas Em ulation and exp er
imen t In Pr o c e e dings of the A CM SIGCOMM pages ! Cam bridge Massac h usetts Au
gust A CM
C Alaettinoglu A U Shank ar K Dussa
Zieger and I Matta Design and imple
men tation of mars A routing testb ed
Journal of Internetworking R ese ar ch
and Exp erienc e ! mar ftp ftpisiedu pub cengiz publications
MaRSDesignpsgz
JM An tis SG Eic k and JD Pyrce Visual
izing the structure of large relational databases
IEEE Softwar e ! Jan uary Ra jiv e L Bagro dia and W enT oh Liao Maisie
A language for the design of ecien t discrete
ev entsim ulations IEEE T r ansactions on Soft
war e Engine ering ! April Sandeep Ba ja j Lee Breslau and Scott Shenk er
Is service priorit y useful in net w orks In A CM
SIGMETRICS June Sandeep Ba ja j Lee Breslau and Scott Shenk er
Uniform v ersus priorit y dropping for la y ered
video Toapp e ar in A CM Sigc omm A J Ballardie P F F rancis and J Cro w croft
Core Based T rees In Pr o c e e dings of the A CM
SIGCOMM San F rancisco Ric hard A Bec k er Stephen G Eic k and Al
lan R Wilks Visualizing net w ork data IEEE
T r ansactions on Visualization and Computer
Gr aphics ! Marc h Bones h ttp wwwcadencecom alta
pro duces b onesdath tml
L Brakmo and L P eterson Exp eriences with
net w ork sim ulation In Pr o c e e dings of the A CM
SIGMETRICSA CM Ken Calv ert and Ellen Zegura Geor
gia tec h in ternet w ork top ology mo dels
h ttp wwwccgatec hedu fac
EllenZegura graphsh tml
Kenneth L Calv ert Matthew B Doar and
Ellen W Zegura Mo deling in ternet top ology IEEE Communic ations Magazine June Stuart K Card George G Rob ertson and
William Y ork The W ebBo ok and the Web F or
ager An information w orkspace for the W orld
Wide W eb In Pr o c e e dings of the A CM SIGCHI
Confer enc e on Human F actors in Computing
Systems pages ! V ancouv er Canada

Mic hael K Coleman and D Stott P ark er
Aestheticsbased graph la y out for h uman com
sumption Softwar e Pr actic e and Exp erienc e ! Decem ber W Dang and Joseph Mac k er The Multicast Dis
semination Proto col MDP F ramew ork Inter
net Dr aft NONE working gr oup June
W ork in Progress
P eter B Danzig Sugih Jamin Ram" on C" aceres
Dann y J Mitzel and Deb orah Estrin An em
pirical w orkload mo del for driving widearea
TCP IP net w ork sim ulations Journal of In
ternetworking R ese ar ch and Exp erienc e!
Marc h
D DeLucia and K Obraczk a Am ulticast con
gestion con trol mec hanism using represen tativ es
T ec hnical Rep ort USCCS TR Depart
men t of Computer Science Univ ersit y of South
ern California Ma y FH Desbrandes S Bertolotti and L Dunand
Opnet an en vironmen t for comm unication

net w ork mo deling and sim ulation In Pr o c Eu
r op e an Simulation Symp osium Octob er
Matthew Doar tiers
ftp ftpnexencom pub pap ers tierstargz
Matthew Doar A better mo del for generating
test net w orks In IEEE Glob al T ele c ommunic a
tions Confer enc e GLOBECOM No v em ber
Matthew Doar and Ian Leslie Ho w bad is naiv e
m ulticast routing In IEEE INF OCOM
A Dupuy J Sc h w artz Y Y emini and D Ba
con NEST A net w ork sim ulation and proto
t yping testb ed Communic ations of the A CM ! Octob er ER Gansner E Koutsoos SC North
A tec hnique for dra wing directed graphs
IEEE T r ansactions on Softwar e Engine ering ! Marc h D Estrin D F arinacci A Helm y V Ja
cobson and L W ei Proto col Indep enden t
Multicast Dense Mo de PIMDM Proto
col Sp ecication Pr op ose d Exp erimental RF C
URL httpnetwebusce dupimpimdmPIM
DM ftxtps ggz Septem ber D Estrin D F arinacci A Helm y D Thaler
S Deering M Handley V Jacobson C Liu
P Sharma and L W ei Proto col Indep enden t
Multicast Sparse Mo de PIMSM Motiv ation
and Arc hitecture Pr op ose d Exp erimental RF C
URL httpnetwebusce dupimpimsmPIM
A r ch ftxtps ggz Octob er D Estrin D F arinacci A Helm y D Thaler
S Deering M Handley V Jacobson C Liu
P Sharma and L W ei Proto col In
dep enden t Multicast Sparse Mo de PIM
SM Proto col Sp ecication RF C
URL httpnetwebusce dupimpimsmPIM
SMvExpRF C ftxtps ggz Marc h
D Estrin M Handley AHelm yP Huang and
D Thaler A Dynamic Bo otstrap Mec hanism for
Rendezv ousbased Multicast Routing Submit
te d to IEEEA CM T r ansactions on Networking
URL httpwwwusce dudeptcs
te chnic al r ep ortshtml Ma y Theo dore F ab er Optimizing throughput in a
w orkstationbased net w ork le system o v er a
high bandwidth lo cal area net w ork A CM Op
er ating Systems R eview ! Jan uary

Kevin F all and Sally Flo yd Sim ulationbased
comparisons of taho e reno and sac k tcp A CM
Computer Communic ation R eview July

S Flo yd V Jacobson S McCanne CG Liu
and L Zhang A reliable m ulticast framew ork for
ligh tw eigh t sessions and application lev el fram
ing A CMIEEE T r ansactions on Networking T o app ear
Sally Flo yd Tcp and explicit congestion noti
cation In A CM Computer Communic ation R e
viewA CM Octob er Sally Flo yd and V an Jacobson On trac phase
eects in pac k etswitc hed gatew a ys Journal of
Internetworking R ese ar ch and Exp erienc e
Septem b er
Sally Flo yd and V an Jacobson Random early
detection gatew a ys for congestion a v oidance In
A CMIEEE T r ansactions on NetworkingA CM
August Sally Flo yd and V an Jacobson Linksharing
and resource managemen t mo dels for pac k et net
w orks In A CMIEEE T r ansactions on Network
ingA CM August TMJ F ruc h terman and EM Reingold Graph
dra wing b y forcedirected placemen t Softwar e
Pr actic e and Exp erienc e !
No v em b er
Nada Golmie Alfred Ko enig and Da vid Su The
NIST A TM Network Simulator Op er ation and
Pr o gr amming V ersion US Departmen t of

Commerce T ec hnology Administration National
Insitute of Standards and T ec hnology Com
puter System Lab oratory Adv anced Systems
Division Gaithersburg MD aug ftp isdnncslnistgo v atmsim sim manpsZ
R Go vindan H Y u and D Estrin Scal
able nontransactional replication in the In ter
net submitted for publication
AudioVideo T ransp ort W orking Group
H Sc h ulzrinne S Casner R F rederic k and
V Jacobson R TP A T r ansp ort Pr oto c ol for
R e alTime Applic ations RF C edition
C H anle A comparison of arc hitecture and p er
formance bet w een reliable m ulticast proto cols
o v er the MBone Masters thesis Institute of
T elematics Univ ersit y of Karlsr uhe Marti Hearst and Chandu Karadi Catacone
An in teractiv e in terface for sp ecifying searc hes
and viewing retriev al results using a large cat
egory hierarc h y In Pr o c e e dings of the th
A nnual International A CMSIGIR Confer enc e pages ! Philadelphia P A July A Helm y and D Estrin Sim ulationbased
#STRESS T esting Case Study A Multicast
Routing Proto col Sixth International Sym
p osium on Mo deling A nalysis and Simulation
of Computer and T ele c ommunic ation Systems
MASCOTS July Ahmed Helm y Deb orah Estrin and Sandeep
Gupta F aultorien ted test generation for m ulti
cast routing proto col design F ormal Description
T e chniques F OR TE XI Pr oto c ol Sp e cic a
tion T esting and V eric ation PSTV XVIII
IFIP TCWG Join International Con
fer enc e Paris F r anc e No v em b er
Ahmed AG Helm y Systematic T esting of Mul
ticast Proto col Robustness PhD Dissertation
pr op osal Submitte das T e chnic al R ep ort to Com
puter Scienc e University of Southern Califor
nia Decem b er P olly Huang Deb orah Estrin and John Heide
mann Enabling largescale sim ulations selec
tiv e abstraction approac h to the study of m ul
ticast proto cols In Pr o c e e dings of the Inter
national Symp osium on Mo deling A nalysis and
Simulation of Computer and T ele c ommunic ation
SystemsMon treal Canada July IEEE to
app ear
Sriniv asan Kesha v REAL a net w ork sim ulator
T ec hnical Rep ort Univ ersit y of Califor
nia Berk eley Decem b er
George Kesidis and Jean W alrand Quic k sim ula
tion of atm buers with ono m ulticlass mark o v
uid sources A CM T r ansactions on Mo deling
and Computer Simulations ! July

Satish Kumar P a vlin Radosla v o v Da v e Thaler
Cengiz Alaettino$ glu Deb orah Estrin and Mark
Handley The MASC BGMP arc hitecture for
in terdomain m ulticast routing T o app ear SIG
COMM Septem ber Will E Leland Murad S T aqqu W alter Will
inger and Daniel V Wilson On the self
similar nature of Ethernet trac extended v er
sion A CMIEEE T r ansactions on Networking ! F ebruary P Luders R Ernst and S Stille An approac h
to automatic displayla y out using com binatorial
optimization algorithms Software Pr acticeand
Exp erienc e ! No v em ber B Mah An empirical mo del of h ttp net w ork
trac In Pr o c e e dings of the IEEE Info c om Kob e Japan April IEEE
Bruce A Mah INSANE Users Manual The
T enet Group Computer Science Division Uni
v ersit y of California Berk eley ma y h ttp HTTP CSBerk eleyEDU bmah
Soft w are Insane InsaneManps
Matt Mathis and Jamshid Mahda vi F orw ard
ac kno wledgemen t Rening tcp congestion con
trol In A CM SIGCOMMA CM August

Matt Mathis Jamshid Mahda vi Sally Flo yd
and Allyn Romano w Tcp selectiv eac kno wledge
men t options rfc In Internet R e quest F or
Comments Octob er S McCanne Router forw arding ser
vices for reliable m ulticast Note
IAA % mlkcsb erk eley edu
to the Reliable Multicast list
rm%mashcsb erk eley edu April S McCanne V Jacobson and M V etterli
Receiv erdriv en la y ered m ulticast In A CM SIG
COMM pages ! Stanford CA USA
August Stev en McCanne Eric Brew er Randy Katz
La wrence Ro w e Elan Amir Y atin Cha w athe
Alan Co op ersmith Ketan Ma y erP atel Suc hi
tra Raman Angela Sc h uett Da vid Simpson
Andrew Sw an T ec kLee T ung Da vid W u and
Brian Smith T o w ard a common infrastructure
for m ultimedianet w orking middlew are In Pr o
c e e dings of the thInternational Workshop on
Network and Op er ating Systems Supp ort for Dig
ital A udio and Vide o pages ! St Louis
Missouri Ma y IEEE
Armin R Mikler Johnn y S K W ong and
V asan t Hona v ar An ob ject orien ted approac h
to sim ualting large comm unication net w orks
Journal of Systems Softwar e ! h uang folder general sim ulator
Robb Mills Comnet iii Ob jectorien ted net
w ork p erformance prediction In GW Ev ans
M Mollaghasemi EC Russell and WE Biles
editors Pr o c e e dings of the Winter Simula
tion Confer enc e pages ! Decem ber John K Ousterhout Tcl and the Tk T o olkit AddisonW esley Reading MA SteveP ark er and Chris Sc hmec hel The pac k et
shell proto col testing to ol Soft w are distribution
at h ttp pla ygroundsuncom psh V P axson Endtoend routing b eha vior in the
in ternet In A CM SIGCOMM August V ern P axson and Sally Flo yd Widearea trac
the failure of P oisson mo deling In A CM SIG
COMM pages ! London United King
dom August A CM
K P erumalla R F ujimota and A Ogielski
T ed a language for mo deling telecomm uni
cation net w orks A CM SIGMETRICS Perfor
manc e Evaluation R eview Marc h A Reddy A self organizing monitoring arc hi
tecture Thesis Prop osal Av ailable from the au
thors Ma y
Reza Rejaie Mark Handely and Deb orah Es
trin Rap An endtoend ratebased congestion
con trol mec hanism for realtime streams in the
In ternet Submitted to ICNP a v ailable at
h ttp net w ebuscedu reza icnpps George G Rob ertson Stuart K Card and
Jo c k D MacKinla y Information visualization
using D in teractiv e animation Communic a
tions of the A CM ! April K Rob ertson K Miller M White and
A Tw eedly StarBurst Multicast File T ransfer
Proto col MFTP Sp ecication Internet Dr aft
NONE working gr oupF ebruary W ork
in progress
D W aitzman S Deering C P artridge Distance
V ector Multicast Routing Proto col No v em ber
RF C
Hussein Salama Mcrsim users man ual Ma y
h uang folder sp ecialized sim ulator
WH Scullin SE Lamm DA Reed Real
time geographic visualization of w orld wide w eb
trac Computer Networks and ISDN Systems
Fifth International World Wide Web Confer
enc e ! Ma y
P Sharma D Estrin S Flo yd and V Jacobson
Scalable timers for soft state proto cols In IEEE
Info c om
Kannan V aradhan Deb orah Estrin and Sally
Flo yd Impact of net w ork dynamics on endto
end pro o cols Case studies in TCP and reliable
m ulticast T ec hnical Rep ort USC CS TR Univ ersit y of Southern California Marc h Vikram Visw esw araiah and John Heidemann
Impro ving restart of idle TCP connections
T ec hnical Rep ort Univ ersit y of Southern
California No v em ber Bernard M W axman Routing of m ultip oin t
connections IEEE Journal on Sele cte d A r e as
in Communic ations Decem b er Liming W ei The design of the USC PIM sim u
lator pimsim T ec hnical Rep ort Uni
v ersit y of Southern California Computer Sci
ence Los Angeles CA aug h ttp catarinauscedu lw ei TR psgz
Da vid W etherall and Christopher J Lin blad
Extending Tcl for dynamic ob jectorien ted pro
gramming In Pr o c e e dings of the USENIX
TclTk Workshop page T oron to On tario
July USENIX
DA Reed WH Scullin TT Kw an Real
time visualization of w orld wide w eb trac In
Pr o c e e dings of ICASEL aR C Symp osium
on Visualizing TimeV arying Data pages !
Williamsburg V A USA Septem ber Ellen W Zegura Ken Calv ert and S Bhat
tac harjee Ho w to mo del an in ternet w ork In
IEEE INF OCOM

Linked assets

Computer Science Technical Report Archive

Conceptually similar

PDF

USC Computer Science Technical Reports, no. 702 (1999)

PDF

USC Computer Science Technical Reports, no. 693 (1999)

PDF

USC Computer Science Technical Reports, no. 679 (1998)

PDF

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

PDF

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

PDF

USC Computer Science Technical Reports, no. 605 (1995)

PDF

USC Computer Science Technical Reports, no. 708 (1999)

PDF

USC Computer Science Technical Reports, no. 709 (1999)

PDF

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

PDF

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

PDF

USC Computer Science Technical Reports, no. 670 (1998)

PDF

USC Computer Science Technical Reports, no. 631 (1996)

PDF

USC Computer Science Technical Reports, no. 640 (1996)

PDF

USC Computer Science Technical Reports, no. 731 (2000)

PDF

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

PDF

USC Computer Science Technical Reports, no. 606 (1995)

PDF

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

PDF

USC Computer Science Technical Reports, no. 696 (1999)

PDF

USC Computer Science Technical Reports, no. 686 (1998)

PDF

USC Computer Science Technical Reports, no. 603 (1995)

Description

Sandeep Bajaj, Lee Breslau, Deborah Estrin, Kevin Fall, Sally Floyd, Padma Haldar, Mark Handley, Ahmed Helmy, John Heidemann, Polly Huang, Satish Kumar, Steven McCanne, Reza Rejaie, Puneet Sharma, Scott Shenker, Kannan Varadhan, Haobo Yu, Ya Xu, Daniel Zappala,. "Virtual InterNetwork testbed: Status and research agenda." Computer Science Technical Reports (Los Angeles, California, USA: University of Southern California. Department of Computer Science) no. 678 (1998).

Asset Metadata

Creator Bajaj, Sandeep (author), Breslau, Lee (author), Estrin, Deborah (author), Fall, Kevin (author), Floyd, Sally (author), Haldar, Padma (author), Handley, Mark (author), Heidemann, John (author), Helmy, Ahmed (author), Huang, Polly (author), Kumar, Satish (author), McCanne, Steven (author), Rejaie, Reza (author), Sharma, Puneet (author), Shenker, Scott (author), Varadhan, Kannan (author), Xu, Ya (author), Yu, Haobo (author), Zappala, Daniel (author)

Core Title USC Computer Science Technical Reports, no. 678 (1998)

Alternative Title Virtual InterNetwork testbed: Status and research agenda (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 19 pages (extent), technical reports (aat)

Language English

Unique identifier UC16269325

Identifier 98-678 Virtual InterNetwork Testbed Status and Research Agenda (filename)

Legacy Identifier usc-cstr-98-678

Format 19 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)