Oscillation Compensating Dynamic Adaptive Streaming over HTTP

15
Oscillation Compensating Dynamic Adaptive Streaming over HTTP Christopher Mueller, Stefan Lederer, Reinhard Grandl, and Christian Timmerer AlpenAdriaUniversität Klagenfurt (AAU) Faculty of Technical Sciences (TEWI) Department of Information Technology (ITEC) Multimedia Communication (MMC) Sensory Experience Lab (SELab) http://blog.timmerer.com http://selab.itec.aau.at/ http://dash.itec.aau.at [email protected] Chief Innovation Officer (CIO) at bitmovin GmbH http://www.bitmovin.com [email protected] Slides: http://www.slideshare.net/christian.timmerer IEEE ICME 2015, June 29 – July 3, 2015

Transcript of Oscillation Compensating Dynamic Adaptive Streaming over HTTP

Oscillation  CompensatingDynamic  Adaptive  Streaming  over  HTTP

Christopher  Mueller,  Stefan  Lederer,  Reinhard Grandl,  and  Christian  Timmerer

Alpen-­‐Adria-­‐Universität Klagenfurt  (AAU)  w Faculty  of  Technical  Sciences   (TEWI) w Department  of  Information  Technology  (ITEC)  wMultimedia  Communication   (MMC) w Sensory   Experience  Lab  (SELab)

http://blog.timmerer.com w http://selab.itec.aau.at/w http://dash.itec.aau.atw [email protected]

Chief  Innovation  Officer  (CIO)  at  bitmovin  GmbHhttp://www.bitmovin.com w [email protected]

Slides:  http://www.slideshare.net/christian.timmerer

IEEE  ICME  2015,  June  29  – July  3,  2015

July  2,  2015 IEEE  ICME  2015 2

Submission   deadline:  November  27,  2015http://www.mmsys.org/ |  http://mmsys2016.itec.aau.at/ |  @mmsys2015

Outline• Introduction,  Motivation,  Problem  Statement• Metrics  and  Tools• Buffer-­‐based  Adaptation  Algorithm  with  Oscillation  Detection  and  Compensation

• Experimental  Results• Conclusions  and  Future  Work

July  2,  2015 IEEE  ICME  2015 3

Over-­‐The-­‐Top  – Adaptive  Media  Streaming

• In  a  Nutshell  …Adaptation logic is within the

client, not normatively specified by the standard, subject to research and development

July  2,  2015 IEEE  ICME  2015 4

Why  do  we  do  that?• HTTP-­‐based  multimedia  streaming  

is  being  massively  deployed– Accounts  for  more  than  60%  of  

Internet  traffic in  peak  periods• Client-­‐centric  approach

– Adaptation  algorithm/logic– Client  behavior  subject  to  research– Throughput-­‐basedvs.  buffer-­‐based

• What  happens  when  multiple  clients  compete with  each  other?

July  2,  2015 IEEE  ICME  2015 5

Source:  Global   Internet   Phenomena   Report:  2H  2014  

What’s  the  problem?• Big  Buck  Bunny  with  different  

representations• Throughput-­‐based  adaptation• Common  test  setup w/  two  

clients  and  varying  bandwidth

July  2,  2015 IEEE  ICME  2015 6

!"#$"%&"'q("')*q+)&,*&"

(-.%

3

4

f

S

c!"#$"%&"'q("')*q+)&,*&"

+$//",

0"123'%

3

43

f3

S3

c3

0"123'%

3 o3 433 4o3 f33 fo3 S33

+$//",

0"123'%

3 o3 433 4o3 f33 fo3 S33

Without  cache

What’s  the  problem?• Big  Buck  Bunny  with  different  

representations• Throughput-­‐based  adaptation• Common  test  setup w/  two  

clients  and  varying  bandwidth

July  2,  2015 IEEE  ICME  2015 7

!"#$"%&"'q("')*q+)&,*&"

(-.%

3

4

f

S

c!"#$"%&"'q("')*q+)&,*&"

+$//",

0"123'%

3

43

f3

S3

c3

0"123'%

3 o3 433 4o3 f33 fo3 S33

+$//",

0"123'%

3 o3 433 4o3 f33 fo3 S33

With  cache

Our  metrics  and  tools!• Adaptation-­‐specific

– Quality  switching  variance:  change  of  representation– Oscillation  variance:  includes  direction  of  switching

• Buffer-­‐specific– Buffer  model  restricting  available  quality  levels

• Based  on  buffer  fill  state• Fitting  to  available  quality  levels  &  network  conditions• Different  behavior:  linear,  exponential,   logarithmic

– Worst  case  buffer:  minimum  buffer  fill  state  in  seconds  that  shall  be  available  prior  to  the  download  of  segment

July  2,  2015 IEEE  ICME  2015 8

Our  approach!  (1/2)• Buffer-­‐based  adaptation  algorithm  

including:– Oscillation  detection– Oscillation  compensation– Fully  client-­‐centric

• Oscillation  factor– Depends  on  quality  switching  

variance and  oscillation  variance– Increases  when  both  metrics  

become  different

July  2,  2015 IEEE  ICME  2015 9

Our  approach!  (2/2)• Buffer-­‐based  adaptation

– c  …  min.  buffer  level  (aka  steady  state)– b  …  fitting  based  on  a  given  c– a  …  max.  representation  bitrate

• Compensation  algorithm– Low  &  high comp.

July  2,  2015 IEEE  ICME  2015 10

!"#$%&'()&*g%&'%$'+g,+-'%&*'&"./$0&)1)gS2$&3$43+g%&'%$'+/&.&)1)gS2$&3$43+g%&'%$'+

S33"5

+6g/

+6&$g%&'%$'+g7/48-9

u

A

v

l

b

%1::+%g;&33g<'$'1-g7h9

u usv usb usd usp A

Our  results!  (1/2)

July  2,  2015 IEEE  ICME  2015 11

!"#$"%&"'q("')*q+)&,*&"

(-.%

3

4

f

S

c!"#$"%&"'q("')*q+)&,*&"

+$//",

0"123'%

3

43

f3

S3

c3+$//",

4%1)55*&)23q6*1&2,

3

3pf

3pc

3pn

3pl

4

0"123'%

3 o3 433 4o3 f33 fo3 S33

4%1)55*&)23q6*1&2,

0"123'%

3 o3 433 4o3 f33 fo3 S33

!"#$"%&"'q("')*q+)&,*&"

(-.%

3

4

f

S

c!"#$"%&"'q("')*q+)&,*&"

+$//",

0"123'%

3

43

f3

S3

c3+$//",

4%1)55*&)23q6*1&2,

3

3pf

3pc

3pn

3pl

4

0"123'%

3 o3 433 4o3 f33 fo3 S33

4%1)55*&)23q6*1&2,

0"123'%

3 o3 433 4o3 f33 fo3 S33

Without

cache

With

cache

Our  results!  (2/2)

July  2,  2015 IEEE  ICME  2015 12

!"#$%&'()&*g×,--.%g/"0.1g30$2'$'&"34(%",#(2,'g/.$5,%.).3'g30$2'$'&"3

6,$1&'7g8

9&'*(

&3#g

:$%

&$3*.g;/

<25=>

pAy

T?pAy

4&).g;8.*"305>

A QA pAA pQA nAA nQA TAA

Without  cache

!"#$%&'()&*g×,--.%g/"0.1g30$2'$'&"34(%",#(2,'g/.$5,%.).3'g30$2'$'&"3

6,$1&'7g8

9&'*(

&3#g

:$%

&$3*.g;/

<25=>

pAy

T?pAy

4&).g;8.*"305>

A QA pAA pQA nAA nQA TAA

With  cache

Our  conclusions!• We  highlighted  some  issues

– Throughput-­‐based   adaptation  logics– Clients  competing for  bandwidth

• In  this  paper– Buffer-­‐based  adaptation  models– Clients  metrics for  oscillation  detection– Oscillation  compensation algorithm– Increase  streaming  performance  – higher   throughput   &  less  quality  

switches• Important:  client-­‐centric  approach

– Enables  scalability,  maintains  advantages of  DASH,  and  is  deployed!• Future  work

– Large-­‐scale  evaluationsJuly  2,  2015 IEEE  ICME  2015 13

http://www.dash-­‐player.com/

Thank  you  for  your  attention

...  questions,  comments,  etc.  are  welcome  …

Priv.-­‐Doz.  Dipl.-­‐Ing.  Dr.  Christian  TimmererAssociate  Professor

Alpen-­‐Adria-­‐Universität Klagenfurt,  Department  of  Information  Technology  (ITEC)Universitätsstrasse 65-­‐67,  A-­‐9020  Klagenfurt,  AUSTRIA

[email protected]­‐klu.ac.athttp://research.timmerer.com/

Tel:  +43/463/2700  3621  Fax:  +43/463/2700  3699©  Copyright:  Christian  Timmerer 14July  2,  2015 IEEE  ICME  2015

July  2,  2015 IEEE  ICME  2015 15

Submission   deadline:  November  27,  2015http://www.mmsys.org/ |  http://mmsys2016.itec.aau.at/ |  @mmsys2015