W orkload Generation for ns Sim ulations of Wide Area Net - - PDF document

▶
w orkload generation for ns sim ulations of wide area net
SMART_READER_LITE
LIVE PREVIEW

W orkload Generation for ns Sim ulations of Wide Area Net - - PDF document

Nov 25, 2022 253 likes •410 views

W orkload Generation for ns Sim ulations of Wide Area Net w orks and the In ternet y z z y M Y uksel B Sikdar K S V astola and B Szymanski y Departmen t of Computer Science z

slide-1
SLIDE 1
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y W
  • rkload
Generation for ns Sim ulations
  • f
Wide Area Net w
  • rks
and the In ternet
  • M
Y uksel y
  • B
Sikdar z K S V astola z and B Szymanski y y Departmen t
  • f
Computer Science z Departmen t
  • f
Electrical Computer and System Engineering Rensselaer P
  • lytec
hnic Institute T ro y
  • NY
  • USA
fyuksembsikdarv astolag net w
  • rksecserpiedu
szymanskcsrpiedu Ph
  • This
w
  • rk
supp
  • rted
b y D ARP A under con tract n um b er FC
slide-2
SLIDE 2
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Outline
  • f
the talk
  • The
ns net w
  • rk
sim ulator
  • Sim
ulating wide area net w
  • rks
Issues
  • T
rac comp
  • sition
and proto col dierences
  • T
rac generation
  • T
  • p
  • logy
  • Sim
ulating wide area net w
  • rks
Solutions
  • T
rac comp
  • sition
  • T
rac generation
  • Sim
ulation results
  • Summary
and conclusions
slide-3
SLIDE 3
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulation platform ns
  • ns
Net w
  • rk
sim ulator dev elop ed b y UCB LBLN and the VINT pro ject
  • Op
en source co de and particularly easy to mo dify
  • Includes
libraries
  • f
top
  • logy
and trac generators and visualization to
  • ls
  • Wide
acceptance in the net w
  • rking
re searc h comm unit y
slide-4
SLIDE 4
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulating wide area net w
  • rks
Issues
  • T
rac comp
  • sition
and proto col dier ences
  • Sim
ulated trac should main tain the prop er comp
  • sition
  • Sources
select their destinations randomly
  • Dieren
t applications generate sessions with dieren t distributions
  • Num
b er
  • f
sessions in a net w
  • rk
v aries con tin uously
  • ns
is incapable
  • f
accoun ting for these eects
slide-5
SLIDE 5
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulating wide area net w
  • rks
Issues Con t
  • T
rac generation
  • T
race based trac generators Unable to ac coun t for c hanging net w
  • rk
conditions and feed bac k
  • Source
based trac generators
  • ns
do es not use application sp ecic sto c hastic mo dels to generate trac
  • ns
do es not ha v e generators for longrange de p enden t Selfsimilar net w
  • rk
trac
slide-6
SLIDE 6
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulating wide area net w
  • rks
Issues Con t
  • T
  • p
  • logy
issues
  • W
ANs and the In ternet can b e view ed as a div erse collecti
  • n
  • f
in terconnected domains
  • Eac
h domain has its
  • wn
in ternal top
  • logy
  • Large
net w
  • rks
also ha v e wide v ariations in their link bandwidths
  • ns
has libraries for top
  • logy
generation
slide-7
SLIDE 7
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulating wide area net w
  • rks
Solutions
  • T
rac comp
  • sition
  • W
e ha v e implemen ted mec hanisms to con trol the p ercen tage
  • f
TCP and UDP sessions in the trac
  • Our
implemen tation randomly c ho
  • ses
source destination pairs from the no des in the net w
  • rk
  • Capabilit
y to use an y suitable distribution to c haracterize w
  • rkload
division b et w een no des and use it to select the sourcedestination pairs
slide-8
SLIDE 8
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulating wide area net w
  • rks
Solutions Con t
  • Session
generation
  • W
e in tro duce application sp ecic trac gener ators in ns for T elnet WWW FTP and SMTP
  • W
e use empirical distributions to c haracter ize the in terarriv al times duration and data transferred b y eac h applicati
  • n
  • Sessions
generated b y eac h application c harac terized b y
  • mean
n um b er
  • f
sessions MNS
  • mean
in terarriv al time
  • f
sessions MIATS
  • mean
duration time
  • f
sessions MDTS
slide-9
SLIDE 9
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulating wide area net w
  • rks
Solutions Con t
  • Session
generation
  • Eac
h expiring session replaced b y random n um b er
  • f
sessions
  • Con
tin uous v ariation in the n um b er
  • f
activ e sessions giving a more realistic net w
  • rk
sce nario
slide-10
SLIDE 10
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulating wide area net w
  • rks
Solutions Con t
  • T
rac generation Selfsimilar trac gen erators
  • Longrange
dep endence in aggregated WWW Ethernet and W AN trac
  • W
e implemen ted t w
  • selfsimilar
trac gener ators in ns
  • ApplicationTraff
icSu pFRP Based
  • n
su p erp
  • sition
  • f
F ractal renew al pro cesses
  • ApplicationTraff
icSS
  • Based
  • n
sup er p
  • sition
  • f
Mark
  • v
mo dulated P
  • isson
pro cesses
slide-11
SLIDE 11
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulation results Mean duration times
  • f
sessions

SMTP FTP Telnet WWW 55 50 45 40 35 25 30 Mean Duration Time (Seconds) 2 4 6 8 10 12 16 14 Simulated Time (Hours)

  • Mean
n um b er
  • f
sessions for v arious application as a function
  • f
the sim ulation length
  • The
mean duration times
  • f
sessions con v erges to the desired v alue
  • sec
within a short sim ula tion time
slide-12
SLIDE 12
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulation results Mean n um b er
  • f
sessions

12 11 10 9 8 6 7 5 2 4 8 6 10 12 14 16 Mean Number of Sessions Telnet WWW FTP SMTP Simulated Time (Hours)

  • Mean
n um b er
  • f
sessions for v arious application as a function
  • f
the sim ulation length
  • The
con tin uous v ariation
  • f
the n um b er
  • f
ses sions results in a closer mo del
  • f
real net w
  • rks
slide-13
SLIDE 13
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulation results Selfsimil ar trac generators

0.0 Poisson 0.2 0.50 H = 0.75 H = 0.90 H = 1.0 0.8 0.6 0.4

  • 0.2

20 40 80 60 k (lag) Covariance (SupFRP)

  • V
ariancelag plot for the trac generated SupFRP compared with a P
  • isson
pro cess
  • W
e note that the correlations deca y extremely slo wly and follo w a p
  • w
er la w c haracteristic
  • f
selfsimilar pro cesses
slide-14
SLIDE 14
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Sim ulation results Selfsimil ar trac generators

0.0 0.2 Poisson 0.4 0.50 H = 0.75 H = 0.90 H = k (lag) 60 80 40 20 1.0 0.8 0.6

  • 0.2

Covariance (SS)

  • V
ariancelag plot for the trac generated SS com pared with a P
  • isson
pro cess
  • W
e note that the correlations deca y extremely slo wly and follo w a p
  • w
er la w c haracteristic
  • f
selfsimilar pro cesses
slide-15
SLIDE 15
  • Departmen
t
  • f
ECSE Rensselaer P
  • lytec
hnic Institute T ro y Summary and conclusions
  • W
e presen ted a metho dology for gener ating realistic w
  • rkloads
for W ANs using the net w
  • rk
sim ulator ns
  • Our
metho d captures the temp
  • ral
and spatial correlation in net w
  • rk
trac
  • W
e created to
  • ls
to generate trac sp e cic to applications lik e T elnet FTP
  • WWW
and SMTP
  • Implemen
ted t w
  • accurate
selfsimilar traf c generators in ns to sim ulate longrange dep enden t aggregate trac