Transcript
Page 1: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Quality  of  Experience  of  Web-­‐based  Adap8ve  HTTP  Streaming  Clients  in  Real-­‐World  Environments  using  Crowdsourcing  

Benjamin  Rainer  and  Chris8an  Timmerer    

Alpen-­‐Adria-­‐Universität  Klagenfurt  (AAU)  w  Faculty  of  Technical  Sciences  (TEWI)  w  Department  of  Informa8on  Technology  (ITEC)  w  Mul8media  Communica8on  (MMC)  w  Sensory  Experience  Lab  (SELab)  

h"p://blog.+mmerer.com  w  h"p://dash.itec.aau.at/  w  h"p://selab.itec.aau.at    mailto:[email protected]­‐klu.ac.at  

December  2,  2014  

Slides:  hVp://www.slideshare.net/chris8an.8mmerer  

Page 2: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Outline  

•  Introduc+on  •  How  to  evaluate  DASH  and  QoE  •  Methodology  •  Results  •  Conclusions  

December  2,  2014   VideoNext  2014,  Sydney   2  

Page 3: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Mul+media  is  Predominant  on  the  Internet  

•  Real-­‐+me  entertainment  –  Streaming  video  and  audio  – More  than  50%  of  Internet  traffic  at  peak  periods  

•  Popular  services  –  NeVlix  (34.9%),  YouTube  (14.0%),  Amazon  Video  (2.6%),  Hulu  (1.4%)  

–  All  delivered  over-­‐the-­‐top  (OTT)  

– MPEG  Dynamic  Adap+ve  Streaming  over  HTTP  

December  2,  2014   VideoNext  2014,  Sydney   3  

Global  Internet  Phenomena  Report:  2H  2014    

Page 4: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Over-­‐The-­‐Top  –  Adap+ve  Media  Streaming  

•  In  a  nutshell  …  

December  2,  2014   VideoNext  2014,  Sydney   4  

C.  Timmerer  and  A.  C.  Begen,  “Over-­‐the-­‐Top  Content  Delivery:  State  of  the  Art  and  Challenges  Ahead”,  In  Proceedings  of  the  ACM  interna+onal  conference  on  Mul+media  (MM  '14),  Orlando,  FL,  USA,  Nov.  2014.    h"p://www.slideshare.net/chris+an.+mmerer/over-­‐the-­‐top-­‐content-­‐delivery-­‐state-­‐of-­‐the-­‐art-­‐and-­‐challenges-­‐ahead  

Adapta8on  logic  is  within  the  client,  not  norma8vely  specified  

by  the  standard,  subject  to  research  and  development  

Page 5: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

MPEG  Dynamic  Adap+ve  Streaming  over  HTTP  

What  is  specified  –  and  what  is  not?  

December  2,  2014   VideoNext  2014,  Sydney   5  

Media  Presenta+on  on  HTTP  Server  

DASH-­‐enabled  Client  Media  Presenta8on  Descrip8on  

     .  .  .  

Segment  

…  

.  

.  

.  Segment  

…  

     .  .  .  

Segment  

…  

.  

.  

.  Segment  

…  

…  

Segments  located  by  HTTP-­‐URLs  

DASH  Control  Engine  

HTTP/1.1   HTTP  Client  

MPD  Parser  

Media  Engine  

On-­‐8me  HTTP  requests  to  segments  

Page 6: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

MPEG  Dynamic  Adap+ve  Streaming  over  HTTP  

What  is  specified  –  and  what  is  not?  

December  2,  2014   VideoNext  2014,  Sydney   6  

Media  Presenta+on  on  HTTP  Server  

DASH-­‐enabled  Client  Media  Presenta8on  Descrip8on  

     .  .  .  

Segment  

…  

.  

.  

.  Segment  

…  

     .  .  .  

Segment  

…  

.  

.  

.  Segment  

…  

…  

Segments  located  by  HTTP-­‐URLs  

DASH  Control  Engine  

HTTP/1.1   HTTP  Client  

MPD  Parser  

Media  Engine  

On-­‐8me  HTTP  requests  to  segments  

Page 7: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

DASH  Data  Model    

December  2,  2014   VideoNext  2014,  Sydney   7  

Segment Info

Initialization Segment http://bitmov.in/500/init.mp4

Media Presentation

Period, start=0s

Period, start=100s

Period, start=200s

Period start=100 baseURL=http://bitmov.in/

AdaptationSet 1 500-1500 kbit/s

AdaptationSet 2 1500-3000 kbit/s

Media Segment 1 start=100s http://bitmov.in/500/seg-1.m4s

Media Segment 2 start=102s http://bitmov.in/500/seg-2.m4s

Media Segment 3 start=104s http://bitmov.in/500/seg-3.m4s

Media Segment 50 start=198s http://bitmov.in/500/seg-50.m4s

AdaptationSet 1 width=640-1280 height=360-720 …

Representation 1 500 Kbit/s

Representation 2 1500 Kbit/s

Representation 2 bandwidth=1500 kbit/s width=960, height=540

Segment Info duration=2s

Template: 500/seg-$Number$.m4s

Initialization: 500/init.mp4

Page 8: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

December  2,  2014   VideoNext  2014,  Sydney   8  

Page 9: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

How  to  evaluate  DASH?  •  Methodology  

– Dataset,  tools  (see  backup  slides  for  details)  –  Common  evalua+on  setup  –  Bandwidth  traces  (real/synthe+c)  vs.  models  

•  Metrics  – Average  media  bitrate/throughput  at  the  client  – Number  of  representa+on/quality  switches  – Number  of  stalls  (in  seconds)  –  buffer  level  

December  2,  2014   VideoNext  2014,  Sydney   9  

C.  Mueller,  S.  Lederer,  C.  Timmerer,  “An  Evalua+on  of  Dynamic  Adap+ve  Streaming  over  HTTP  in  Vehicular  Environments”,  In  Proceedings  of  the  Fourth  Annual  ACM  SIGMM  Workshop  on  Mobile  Video  (MoVid12),  Chapel  Hill,  North  Carolina,  February  2012.  

Page 10: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Quality  of  Experience  •  Quality  of  Experience  

–  “…  is  the  degree  of  delight  or  annoyance  of  the  user  of  an  applica+on  or  service…”  

–  Factors  influencing  /  features  of  QoE  may  lead  to  applica+on-­‐specific  defini+ons  

•  Subjec+ve  quality  assessments  –  Laboratory  environment  [ITU-­‐T  B.500  /  P.910]  –  Crowdsourcing  with  special  plaVorms  or  social  networks  

•  QoE  of  DASH-­‐based  services  –  Startup  delay  (low)  –  Buffer  underrun  /  stalls  (zero)  –  Quality  switches  (low)  and  media  throughput  (high)  

December  2,  2014   VideoNext  2014,  Sydney   10  

P.  Le  Callet,  S.  Möller  and  A.  Perkis,  eds.,  “Qualinet  White  Paper  on  Defini+ons  of  Quality  of  Experience  (2012)”,  European  Network  on  Quality  of  Experience  in  Mul>media  Systems  and  Services  (COST  Ac>on  IC  1003),  Lausanne,  Switzerland,  Version  1.2,  March  2013."  

Page 11: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Methodology  •  Quality  of  Experience  …  

–  Mean  Opinion  Score  [0..100]  –  [other  objec+ve  metrics:  

start-­‐up  +me,  throughput,  number  of  stalls]  •  …  Web-­‐based  Adap+ve  HTTP  Streaming  Clients  …  

–  HTML5+MSE:  DASH-­‐JS  (dash.itec.aau.at),  dash.js  (DASH-­‐IF,  v1.1.2),  YouTube  •  …  Real-­‐World  Environments  …  

–  DASH-­‐JS,  dash.js  hosted  at  ITEC/AAU  (~  10Gbit/s)  –  YouTube  hosted  at  Google  data  centers  –  Content:  Tears  of  Steel  @  144p  (250  kbit/s),  240p  (380  kbit/s),  360p  (740  kbit/

s),  480p  (1308  kbit/s),  and  720p  (2300  kbit/s);  segment  size:  2s  –  Users  access  content  over  the  open  Internet  

•  …  Crowdsourcing  –  Campaign  at  Microworker  plaVorm  (others  also  possible:  Mechanical  Turk,  

social  networks)  limited  to  Europe,  USA/Canada,  India  –  Screening  Techniques:  Browser  fingerprin+ng,  s+mulus  presenta+on  +me,  

QoE  ra+ngs  and  pre-­‐ques+onnaire  

December  2,  2014   VideoNext  2014,  Sydney   11  

Page 12: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Results:  QoE  

December  2,  2014   VideoNext  2014,  Sydney   12  

Page 13: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Results:  Media  Throughput  

December  2,  2014   VideoNext  2014,  Sydney   13  

Page 14: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Results:  Start-­‐Up  Time  

December  2,  2014   VideoNext  2014,  Sydney   14  

Page 15: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Results:  Number  of  Switches  

December  2,  2014   VideoNext  2014,  Sydney   15  

Page 16: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Results:  Number  of  Stalls  

December  2,  2014   VideoNext  2014,  Sydney   16  

Page 17: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Results:  Summary  •  DASH-­‐JS  

–  High  start-­‐up  +me  –  Low  number  of  stalls  –  Good  throughput,  QoE  

•  dash.js  –  Low  start-­‐up  +me  –  High  #  stalls  –  Low  throughput  –  Low  QoE  

•  YouTube  –  Low  start-­‐up  +me  –  Low  number  of  stalls  –  Best  throughput,  QoE  

December  2,  2014   VideoNext  2014,  Sydney   17  

Page 18: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Conclusions  •  QoE  evalua+on  of  DASH-­‐like  systems  in  real-­‐world  environments  

using  crowdsourcing  –  Detailed  methodology  described  in  the  paper  –  Results  indicate  that  the  delivered  representa+on  bitrate  (media  

throughput)  and  the  number  of  stalls  are  the  main  influence  factors  on  the  QoE  

–  Results  confirmed  by  previous  evalua+ons  but  within  controlled  environments  

–  Evidence  about  QoE  aspects  of  DASH-­‐enabled  Web  clients  within  real-­‐world  environments  

–  Feasibility  of  using  crowdsourcing  for  subjec+ve  quality  assessments  •  Future  work  

–  Comprehensive  evalua+on  of  various  adapta+on  logics  (both  objec+ve  and  subjec+ve)  and  

–  the  impact  of  dedicated  delivery  infrastructures  aiming  to  improve  DASH-­‐based  services  

December  2,  2014   VideoNext  2014,  Sydney   18  

Page 19: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

Thank  you  for  your  a"en+on  

...  ques+ons,  comments,  etc.  are  welcome  …  

   

 Priv.-­‐Doz.  Dipl.-­‐Ing.  Dr.  Chris+an  Timmerer  Associate  Professor  

Klagenfurt  University,  Department  of  Informa+on  Technology  (ITEC)  Universitätsstrasse  65-­‐67,  A-­‐9020  Klagenfurt,  AUSTRIA  

[email protected]­‐klu.ac.at  h"p://research.+mmerer.com/  

Tel:  +43/463/2700  3621  Fax:  +43/463/2700  3699  ©  Copyright:  Chris>an  Timmerer  

19  December  2,  2014   VideoNext  2014,  Sydney  

Page 20: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

December  2,  2014   VideoNext  2014,  Sydney   20  

Page 21: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

December  2,  2014   VideoNext  2014,  Sydney   21  

Page 22: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

BACKUP  SLIDES  

December  2,  2014   VideoNext  2014,  Sydney   22  

Page 23: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

End-­‐to-­‐End  DASH  System  Aspects  •  (Distributed)  dataset  

–  Full  movie  length  in  high  quality  

–  Various  bitrate,  resolu+ons,  segment  lengths  (2-­‐15s),  (sub-­‐)segments  

–  Distributed:  ini+al  3  sites,  now  9  in  Europe,  USA,  Taiwan  

•  DASH  encoder  –  Encoding  +  Mul+plexing  +  

MPD  genera+on  –  Fully  configurable  using  a  

configura+on  file  –  Enables  batch  processing  –  x264/ffmpeg  +  GPAC  MP4Box  

December  2,  2014   VideoNext  2014,  Sydney   23  

S.  Lederer,  C.  Müller,  C.  Timmerer,  “Dynamic  Adap+ve  Streaming  over  HTTP  Dataset”,  In  Proceedings  of  the  ACM  Conference  on  Mul+media  Systems  2012,  Chapel  Hill,  North  Carolina,  February  2012.  //  S.  Lederer,  C.  Mueller,  C.  Timmerer,  C.  Concolato,  J.  Le  Feuvre,  K.  Fliegel,  “Distributed  DASH  Dataset”,  In  Proceedings  of  the  ACM  Conference  on  Mul+media  Systems  2013,  Oslo,  Norway,  2013.  

Page 24: Quality of Experience of Web-based Adaptive HTTP Streaming Clients in Real-World Environments using Crowdsourcing

End-­‐to-­‐End  DASH  System  Aspects  •  Playback  

–  VLC  plugin  (first  implementa+on)  –  DASH-­‐JS  (HTML5  +  MSE)  –  libdash  /  qtsampleplayer  

•  MPD  valida+on  –  XML  schema  valida+on  –  Xlink  resolver  &  processing  –  Addi+onal  valida+on  rules  

(Schematron)  •  Experimental  

–  DASH  over  Content-­‐Centric  Networks  (CCN)  

–  VLC  +  libdash  

December  2,  2014   VideoNext  2014,  Sydney   24  

C.  Müller  and  C.  Timmerer,  “A  VLC  Media  Player  Plugin  enabling  Dynamic  Adap+ve  Streaming  over  HTTP”,  In  Proceedings  of  the  ACM  Mul+media  2011,  Sco"sdale,  Arizona,  November  2011.  //  B.  Rainer,  S.  Lederer,  C.  Müller,  C.  Timmerer,  “A  Seamless  Web  Integra+on  of  Adap+ve  HTTP  Streaming”,  In  Proceedings  of  the  20th  European  Signal  Processing  Conference  2012,  Bucharest,  Romania,  August  2012.  

h"p://records.sigmm.ndlab.net/2013/04/open-­‐source-­‐column-­‐dynamic-­‐adap+ve-­‐streaming-­‐over-­‐h"p-­‐toolset/  


Top Related