MI STRAL Modeling of Comput er Syst ems and Telecommunicat ion Net - - PowerPoint PPT Presentation

mi stral
SMART_READER_LITE
LIVE PREVIEW

MI STRAL Modeling of Comput er Syst ems and Telecommunicat ion Net - - PowerPoint PPT Presentation

MI STRAL Modeling of Comput er Syst ems and Telecommunicat ion Net works Philippe Nain I NRI A Sophia Ant ipolis Evaluat ion of I NRI A 1B Theme J ouy-en-J osas, Oct ober 22, 2003 Scient if ic goals MI STRAL is devot ed t o building


slide-1
SLIDE 1

MI STRAL

Modeling of Comput er Syst ems and Telecommunicat ion Net works

Philippe Nain I NRI A Sophia Ant ipolis

Evaluat ion of I NRI A 1B Theme J ouy-en-J osas, Oct ober 22, 2003

slide-2
SLIDE 2

Scient if ic goals

MI STRAL is devot ed t o

building mat hemat ical models developing sof t wares, prot ot ypes

f or

evaluat ing

  • pt imizing

cont rolling

perf ormance of st ochast ic discret e-event dynamical syst ems (net works)

slide-3
SLIDE 3

Scient if ic goals (cont d.)

Cont ribut e t o f undament al research

(st ochast ic processes, game t heory, opt imal cont rol, … )

Cont ribut e t o net working (prot ocol analysis

and opt imizat ion, ressource allocat ion, … )

Achieve cross-f ert ilizat ion bet ween

f undament al research and net work engineering

slide-4
SLIDE 4

Composit ion of MI STRAL (Oct . 03)

I NRI A r esear cher s (4)

Eit an Alt man Kost ya Avr achenkov Alain J ean-Mar ie (since Oct . 1, 2003) Philippe Nain (scient . leader )

Post -doc: Ar zad-Alam Kher ani

slide-5
SLIDE 5

Composit ion of MI STRAL (cont d.)

PhD st udent s (6)

Ahmad Al Hanbali (Ad hoc net work analysis) Ur t zi Ayest a (TCP modeling) Flor ence Clevenot (CDN analysis) Robin Groenevelt (Peer-t o-peer analysis) Daniele Mior andi (Bluet oot h) Balakr ishna Pr abhu (TCP over wireless)

Engineer : David Sagnol Administ r at ive assist ant : Ephie Der iche

slide-6
SLIDE 6

Changes in Mist r al since last evaluat ion

Depart ures f rom Mist ral

  • F. Baccelli in Sep. 1998 (TREC)
  • A. J ean-Mar ie in Sep. 1998 (U. Mont pellier )
  • Z. Liu in Aug. 2000 (I BM, T.J . Wat son)

Arrivals in Mist ral

  • K. Avr achenkov in Aug. 2000 (CR2)
  • A. J ean-Mar ie in Oct . 2003 (DR2) -->

MAESTRO

  • S. Alouf in Feb. 2004 (CR2) -->

MAESTRO

slide-7
SLIDE 7

Main cont r ibut ion - 1

TCP modeling

Gener al model f or calculat ing TCP t hr oughput (ar bit r ar y er godic & st at ionar y packet loss pr ocess) Fixed-point appr oach f or dimensioning of TCP/ I P net wor ks wit h ar bit r ar y t opology Ear ly r et r ansmit (I nt er net dr af t ) [Sigcomm ’00, I nf ocom ’02, Sigmet r ics ’00, … ..] Mor e t omor r ow (Kost ya)

slide-8
SLIDE 8

Main cont ribut ion - 2

Wireless net works

Capacit y of mult iser vice UMTS net wor ks wit h best -ef f or t applicat ions Power cont r ol and t r ansmission r at e cont r ol in CDMA cont ext (r eal-t ime) TCP on ad hoc net wor ks

[Mobicom ’02, Winet ’02, I EEE ToN ’01, ...] More t omorrow (Eit an)

slide-9
SLIDE 9

Main cont ribut ion - 3

On-line est imat ion Est imat ion mult icast audience

Dif f usion pr ocess (Or nst ein-Uhlenbeck) Linear f ilt ering (Kalman, Wiener ) Validat ion against t r aces Applicat ions: I nt er net TV/ Radio br oadcast

[I nf ocom ’01, ’02, I EEE Trans Sign. Pr oc. …

]

More t omorrow (P N)

slide-10
SLIDE 10

Sof t war e development

WAGON

Web ser ver benchmar k and Web t r af f ic gener at or

ALLEGRO

Mult imedia (voice, audio) t r af f ic gener at or

ERS

Set of t ools f or analysis of DEDS (ht t p:/ / www-sop.inr ia.f r / mist r al/ sof t / er s.ht ml)

slide-11
SLIDE 11

I ndust r ial collabor at ions

ALCATEL SPACE

Traf f ic modeling f or sat ellit e net works Design and opt imizat ion of sat ellit e net works

Fr ance Telecom

Decision making in compet it ive environment Ressource allocat ion in UMTS net works (pat ent ) Perf ormance and cont rol of mult imedia t raf f ic Generat ion of Web t raf ic on VTHD plat f orm

CEGETEL, CS-Telecom

Modeling of mult imedia t raf f ic on I P/ MPLS net works

slide-12
SLIDE 12

Posit ioning of MI STRAL

At I NRI A: St rong collaborat ions wit h 1B

proj ect s + Met a2, Miaou, Oasis, Trio

I n France: Eurecom, ENST, LAAS, U.

Marseille

Out side France

Many research groups on PE in t he world (act ive area) Close cont act wit h several of t hem

slide-13
SLIDE 13

Nat ional academic collabor at ions

ARC TCP (Armor , Hipercom, Planet e, Rap,

Trec)

ARC P

RI XNET (Armor , U. Versailles, I BM)

ARC MADDES (Met a2 (now Met alau), Trio) ARC COMMOBI L (Algo, Hipercom, Sloop

(now Mascot t e)

ARC Epsilon (Fract ale, ENST) AS P

erf ormance Modeling (Apache, Armor, Trec, I MAG, ENST, U. Versailles, P aris 13)

slide-14
SLIDE 14

I nt er nat ional collabor at ions

Aust ralia (U. Sout h Aust ralia) China (LI AMA) I ndia (I I S Bangalore) I srael (Technion, Tel-Aviv Univ.) J apan (Tsukuba) Morocco (Rabat ) Poland (Univ. Zielona Gora) Russia (Lyapounov I nst it ut e) The Net her lands (CWI , U. Eindhoven, Leiden, Twent e, Vrij e Univ.) US (I BM, U. Arizona, Br own, Columbia, I llinois, Maryland,

Massachuset t s, Bost on U.)

slide-15
SLIDE 15

Obj ect ives in 1999-2002

TCP

and impact on net work perf ormance

Maint ain t heoret ical act ivit ies and t ransf er

  • f met hods/ result s t o indust ry

Keep st rong visibilit y of MI STRAL

  • )

: decide you

slide-16
SLIDE 16

Next 4 year s

MI STRAL t erminat es (1993-2003)

slide-17
SLIDE 17

Next 4 year s (cont d.)

MI STRAL t erminat es (1993-2003) Creat ion of MAESTRO

Locat ed in Mont pellier and Sophia Ant ipolis I NRI A, U. Mont pellier , CNRS proj ect

slide-18
SLIDE 18

Next 4 year s (cont d.)

MAESTRO research t hemes:

I P net works (short TCP f lows, HS TCP, …

)

Wireless net works (ad hoc, TCP, WLAN , …

)

Overlay net works (CDN, P 2P,…

)

Net working games (pricing, equilibrium, ...) Theory of DEDS (queueing, net wor k calculus, …

)

Cont rol of st ochast ic processes (MDP, singular pert urbat ion, … )

slide-19
SLIDE 19

Next 4 year s (cont d.)

Challenges

Find macroscopic models more amenable t o

analysis and t o operat ional use

Mast er t ools necessary f or handling t hese

models (dif f usions, random graphs, simulat ors, et c)

Cont ribut e t o indust rial product s and

st andards