! Energy!Efficient!Content! Distribu2on!in!an!ISP!network ! - - PowerPoint PPT Presentation

energy efficient content
SMART_READER_LITE
LIVE PREVIEW

! Energy!Efficient!Content! Distribu2on!in!an!ISP!network ! - - PowerPoint PPT Presentation

! Energy!Efficient!Content! Distribu2on!in!an!ISP!network ! Jean:Claude!BERMOND! ! Sophia!An*polis,!France! Energy!Efficient!Content!Distribu*on! Work!presented!at!Globecom!2013! Giroire,!Modrzejewski,! Coa*,!Sophia!An*polis,!France!


slide-1
SLIDE 1

! Energy!Efficient!Content! Distribu2on!in!an!ISP!network!

Jean:Claude!BERMOND! !

Sophia!An*polis,!France!

slide-2
SLIDE 2

Energy!Efficient!Content!Distribu*on!

  • Work!presented!at!Globecom!2013!

2! Giroire,!Modrzejewski,! Tahiri! Coa*,!Sophia!An*polis,!France! Chiaraviglio! CNIT!and!University!la!Sapienza,!Rome,!Italy.!! Le!Rouzic! Orange/France!Telecom,!Lannion,!France! BoneSo! Politecnico!di!Torino,!Italy! Musumeci! CNIT!and!Politecnico!di!Milano,!Italy! Gonzalez,!Guerrero! Universidad!Carlos!III,!Madrid,!Spain!

slide-3
SLIDE 3

Context!

Huge!increase!in!Traffic!

3!

2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Year 500 1000 1500 2000 2500 3000 Global Consumer Traffic [PB/day]

Video streaming File sharing Other

Year%to%year)peak)increases)around)40%)

!

slide-4
SLIDE 4

Context!

Limited!financial!resources!to!accomodate! growth!

4!

1998 2000 2002 2004 2006 2008 2010 2012 Year 200 400 600 800 1000 1200 Price [$/Mb]

Year%to%year)price)falls)around)35%)

!

slide-5
SLIDE 5

Context!

Environmental!pressure!

5!

2)to)10%)of)global)energy)consump>on)

!

slide-6
SLIDE 6

2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Year 500 1000 1500 2000 2500 3000 Global Consumer Traffic [PB/day]

Video streaming File sharing Other

Context!

Crucial!trend:!Traffic!is!video.!!

6!

Video)streaming)already)over)50%) [CISCO)Forecast])86%)by)2016)

!

slide-7
SLIDE 7

Context!

Different!models!to!distribute!(redondant)!content!

7!

slide-8
SLIDE 8

Internet Core Network Access Network Metro Network DSLAM Users Peering Link Content

Problem!

  • We!consider!a!network!operator.!!
  • The!network!operator!distributes!

– Traffic!of!content!providers!! – Its!own!traffic!(money!is!in!the!content,!! less!in!the!transport)!

!

  • 70%!of!traffic!comes!from!the!peering!link!

8!

slide-9
SLIDE 9

Internet Core Network Access Network Metro Network DSLAM Users Peering Link Content

Problem!

  • Problem:!How!to!distribute!this!

content!in!order!to!be!energy! efficient?!!

  • Ques2ons:!!

– Where!to!place!caching!facili2es?! – Where!to!place!storage!facili2es!for! the!operator’s!own!content?!

9!

slide-10
SLIDE 10

Contribu*ons!and!Related!Work!

  • Lots!of!work!on!distribu*ng!content!and!caching,!e.g.!:!!

[1]!V.!Jacobson,!D.!K.!SmeSers,!J.!D.!Thornton,!M.!F.!Plass,!N.!H.!Briggs!and!R.!L.! Branard.!Networking!named!content.!ACM!CoNEXT!2009.! [2]!D.!Perino,!M.!Varvello.!A!reality!check!for!content!centric!networking.!ACM! SIGCOMM!2011.! [3]I.!Baev,!R.!Rajaraman!and!C.!Swamy.!Approxima*on!algorithms!for!data! placement!problems.!SIAM!Journal!on!Compu*ng,!2008.!

  • Here,!!

– Different!ques*on:!How!much!can!we!save!energy!in!an!

  • p*mal!content!distribu*on!when!controling!both!

transport!and!storage?!! – Realis*c!scenario,!provided!by!Orange/France!Telecom! – Efficient!analy*cal!model!for!op*mal!dimensioning!

10!

slide-11
SLIDE 11

Our!Model!

11!

Diffusion)of)content)is)done)via)a)logical)tree)

slide-12
SLIDE 12

Our!Model!

12!

A)video)is)sent)to)one)user)

slide-13
SLIDE 13

Our!Model!

13!

A)video)is)sent)to)several)users:)not)efficient)use)of)energy)

slide-14
SLIDE 14

Our!Model!

14!

Placement)of)storage)facili>es)

slide-15
SLIDE 15

Our!Model!

15!

Placement)of)storage)facili>es:)energy)is)saved)

slide-16
SLIDE 16

Our!Model!

16!

Not)worth)for)non)popular)videos)

slide-17
SLIDE 17

Modeling!Popularity!

17!

slide-18
SLIDE 18

Modeling!Popularity!

18!

slide-19
SLIDE 19

Modeling!Popularity!

19!

slide-20
SLIDE 20

Modeling!Popularity!

20!

slide-21
SLIDE 21

Modeling!Popularity!

21!

slide-22
SLIDE 22

Modeling!Popularity!

22!

slide-23
SLIDE 23

Power!Models!

23!

slide-24
SLIDE 24

Power!Models!

24!

slide-25
SLIDE 25

Efficient!Analy*cal!Model!

  • We!assume:!

– Network!represented!by!a!leveljregular!tree! – Homogeneous!user!popula*on! – Linear!power!model,!known!devices!for!levels! – Zipfian!popularity!model!

25!

slide-26
SLIDE 26

Efficient!Analy*cal!Model!

  • For!one!class!of!videos:!find!the!best!level!to!cache!it!
  • For!efficiency:!Compute!the!interval!of!popularity!of!

videos!cached!at!each!level.!!

26!

slide-27
SLIDE 27

Results!

27!

slide-28
SLIDE 28

Results!j!Sen*vity!

28!

Small!changes!in!β!–!huge!changes!in!results!

!

slide-29
SLIDE 29

Results!j!Sen*vity!

29!

Changes!in!total!throughput,!small!changes!in!results!!

!

slide-30
SLIDE 30

Takejaways!

  • Around!10%!savings!
  • Importance!of!model!parameters:!!

– Knowing!popularity!distribu*on!is!crucial! – Op*mizing!cost!of!storage!more!important!! than!I/O!!

  • Op*mal!is!feasible:!!

– No!changes!in!core!networks! – DSLAM!augmented!by!a!small!flash!module! routers!by!single!servers!

30!

slide-31
SLIDE 31

Future!Work!

  • Study!other!architectures:!!!

– with!mul*ple!peering!points,!! – other!tree!degree!distribu*ons.!! – with!Operator’s!datacenter.!!

  • Here:!op*mal!gain.!Test!caching!on!real!traffic!

traces.!!

  • Economic!studies:!Content!providers!are!

reluctant!to!allow!caching!of!their!content! (contract,!incen*ve…)!

31!

slide-32
SLIDE 32

Future!Work!

  • Study!other!architectures:!!!

– with!mul*ple!peering!points,!! – other!tree!degree!distribu*ons.!! – with!Operator’s!datacenter.!!

  • Here:!op*mal!gain.!Test!caching!on!real!traffic!

traces.!!

  • Economic!studies:!Content!providers!are!

reluctant!to!allow!caching!of!their!content! (contract,!incen*ve…)!

32!

THANKS!!!!!