28
Improving the user experience of mul4media streaming services in highly dynamic environments Frank Eliassen ([email protected]) Verdikt conference, 26th April 2012 4/26/2012 ROMUS project @ VERDIKT 1

Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Embed Size (px)

DESCRIPTION

VERDIKT conference 2012

Citation preview

Page 1: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Improving  the  user  experience  of  mul4media  streaming  services  in  highly  dynamic  environments

Frank Eliassen ([email protected]) Verdikt conference, 26th April 2012

4/26/2012   ROMUS  project  @  VERDIKT     1  

Page 2: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Many problems still hampering Internet live video streaming

•  Increased heterogeneity of networks and terminals

•  Variability in resource availability such as bandwidth, CPU, and joining and leaving of devices

•  Challenge: How  to  provide  each  consumer  with  the  best  possible  viewing  experience  when  considering  heterogeneity  and  variability,  while  maintaining  efficiency  and  scalability

Car computer

Home PC

GSM /UMTS

The Internet

BT/ WLAN

4/26/2012   ROMUS  project  @  VERDIKT     2  

Page 3: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

ROMUS  project:  main  objec4ve  •  To  inves4gate  and  provide  solu4ons  for  mul4media  (video)  streaming  services  to  provide  each  consumer  best  possible  experience  in  a  highly  dynamic  environment  – best  possible  quality  (image  quality,  con4nuous  playback)  

–  least  possible  visual  distor4on  

4/26/2012   ROMUS  project  @  VERDIKT     3  

Page 4: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

ROMUS  –  three  main  areas  of  results  •  Adapta4on  and  Robustness  in  Live  P2P  Streaming  

–  Chameleon:  a  novel  adap4ve  P2P  streaming  protocol  targe4ng  live  video  streaming  (video  “broadcasts”)  

–  S*r:  A  social  network  based  P2P  streaming  solu4on  to  be[er  handle  peer  dynamics  in  live  video  streaming    

•  Video  quality  assessment  –  Randomized  Pair  Comparison  (R/PC),  a  novel  test  method  for  

subjec4ve  video  quality  assessment  –  A  set  of  guidelines  to  reduce  visual  distor4on  in  scalable  video  

streaming  •  Mul4core  Processing  to  handle  mul4media  workloads  

–  Techniques  for  exploi4ng  mul4core  processing  and  graphical  processing  units  on  individual  peer  nodes  to  improve  video  quality  

–  P2G:  a  framework  for  distributed  processing  on  computer  nodes  in  a  cluster  suppor4ng  mul4media  workloads  with  so_  deadlines.  

4/26/2012   ROMUS  project  @  VERDIKT     4  

Page 5: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

ROMUS  team  •  Professor  Frank  Eliassen,  University  of  Oslo  (project  leader)  •  Professor  Carsten  Griwodz,  Simula  Research  Laboratory  •  Professor  Pål  Halvorsen,  Simula  Research  Laboratory  •  Dr.  Viktor  S.  Wold  Eide,  University  of  Oslo  (post  doc  for  18  

months)  •  Dr.  Eli  Gjørven,  University  of  Oslo  (post  doc  for  6  months)  •  Anh  Tuan  Nguyen,  University  of  Oslo  (PhD  scholar)  •  Pengpeng  Ni,  Simula  Research  Laboratory  (PhD  scholar)  •  Håkon  Stensland,  Simula  Research  Laboratory  (PhD  scholar)  

4/26/2012   ROMUS  project  @  VERDIKT     5  

Page 6: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Adapta4on  and  Robustness  in  Live  Peer-­‐to-­‐Peer  Streaming  

ROMUS  project  

4/26/2012   ROMUS  project  @  VERDIKT     6  

Page 7: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Mo4va4on  •  Limita4ons  of  tradi4onal  live  P2P  streaming  systems  – No  differen4ated  QoS:  users  must  receive  the  same  stream  regardless  of  their  bandwidth  (high  capacity  users  perceive  the  same  low  quality  as  average  users)  

– No  con4nuous  playback/black  block  images:  with  the  current  best-­‐effort  Internet  and  the  peer  dynamics,  the  streaming  quality  at  each  peer  is  easily  impaired  (when  the  available  bandwidth  at  a  peer  drops  below  the  streaming  rate,  it  may  suffer  playback  skips)  

4/26/2012   ROMUS  project  @  VERDIKT     7  

Page 8: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Main  hypotheses  

•  Adaptable  coding  techniques  (such  as  SVC)  can  bring  significant  benefits  in  terms  of  differen4ated  QoS  and  con4nuous  playback  to  live  P2P  streaming  

•  Network  coding  and  social  networking  can  improve  the  robustness  of  the  P2P  system  with  respect  to  network  fluctua4ons  and  peer  dynamics  

•  Quality-­‐aware  overlays  can  ensure  high  capacity  peers  will  receive  high  quality  video  

4/26/2012   ROMUS  project  @  VERDIKT     8  

Page 9: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Scalable  Video  Coding  (H.264  AVC/SVC)  

•  A  video  coding  technique:  encodes  a  video  into  layers  of  quality  

•  Standardized  in  July  2007  by  ITU-­‐T  (H.264)  

•  ~10%  bitrate  overhead  and  an  indis4nguishable  visual  quality  compared  to  H.264  AVC    

4/26/2012   ROMUS  project  @  VERDIKT     9  

Page 10: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Scalable  Video  Coding  (source:  h[p://www.hhi.fraunhofer.de)  

The three scalability dimensions “Any” sub-stream can be extracted

4/26/2012   ROMUS  project  @  VERDIKT     10  

Page 11: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Network  Coding    (Linear  network  coding)  

•  Instead  of  simply  forwarding  data,  intermediate  nodes  may  recombine  several  input  packets  into  one  or  several  output  packets  

•  Perfect  collabora4on  –  Poten4al  throughput  improvements  –  A  high  degree  of  robustness  

4/26/2012   ROMUS  project  @  VERDIKT     11  

Page 12: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Chameleon:  a  pull-­‐based  P2Pstreaming  protocol  

Chameleon’s architecture with key components

4/26/2012   ROMUS  project  @  VERDIKT     12  

Page 13: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Evalua4on:  Baseline  •  FABALAM:  Y.  Liu,  W.  Dou,  and  Z.  Liu,  “Layer  Alloca4on  

Algorithms  in  Layered  Peer-­‐to-­‐Peer  Streaming,”  in  Proc.  of  IFIP  interna*onal  conference  on  network  and  parallel  compu*ng  (NPC),  Oct.  2004,  pp.  167–174  

•  Commons  –  Pull-­‐based  P2P  streaming  protocol  –  Adaptability  

•  Differences  

 Chameleon   FABALAM  

H.264/SVC   Synthe4c  layered  data  

Network  coding              -­‐  A  layer  is  delivered  from  mul4ple  senders  

Approxima4on  algorithm            -­‐  A  layer  is  delivered  from  one  sender  

4/26/2012   ROMUS  project  @  VERDIKT     13  

Page 14: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Evalua4on:  on  scalability  

Chameleon is scalable and offers much lower skip rates and higher quality satisfaction

Chameleon vs. FABALAM: Skip rates Chameleon vs. FABALAM: Quality satisfaction

4/26/2012   ROMUS  project  @  VERDIKT     14  

Page 15: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Evalua4on:  on  peer  dynamics  

Using weibull(k,2) for generation of different levels of peer dynamic Skip rates

Quality satisfaction

Chameleon can adapt well to peer dynamics to maintain low skip rates and high quality satisfaction

4/26/2012   ROMUS  project  @  VERDIKT     15  

Page 16: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Video  Quality  Assessment    

Flicker  effects  in  Adap4ve  Video  Streaming  

4/26/2012   ROMUS  project  @  VERDIKT     16  

ROMUS  project  

Page 17: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Quality  adapta4on  mechanism    in  Chameleon  at  work    

What are acceptable limits of quality fluctuations for the user? Can we provide guidelines for how to adapt to reduce visual distortion?

4/26/2012   ROMUS  project  @  VERDIKT     17  

Page 18: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Media Performance Group

Visual perception of dynamically adaptive video (1)!Understanding and using limits of user perception and perceived quality"

Signal-­‐to-­‐noise  ra4o  (SNR)  scaling   Noise  flicker  

Resolu4on  scaling  Blur  flicker  

Frame  rate  scaling   Judder  

Three main types of visual artifacts"

Page 19: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Media Performance Group

Visual Perception of dynamically adaptive video (2)  

Encoding  Layers  

Frames  over  Time  

High  Frequency  

Low  Frequency  

Two main fluctuation factors"

Amplitude" Frequency"

Field study"•  mobile devices, free seating, resolution 480x320@30fps, no sunlight,

lounge chairs"Experiment design"•  repeated measures, single-stimulus, randomized block design"•  blocking by flicker type and amplitude level"•  baselines for highest and lowest quality without quality fluctuations"

Page 20: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Media Performance Group

Visual Perception of dynamically adaptive video (3)!

Three influential factors"

Amplitude"Most dominant effect"Flicker is almost undetectable at amplitudes < 8QP and almost always detectable for larger amplitudes"

Frequency"Major effect"Acceptance thresholds compared to constant low quality video:"worse when above 1 Hz,"often better when below 0.5 Hz""

Content"Minor effect"But: content can influence"flicker perception;"low interaction for noise flicker and stronger for blur flicker""

Page 21: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

P2G:  Parallel  Processing  Graphs      

A  Framework  for  Distributed  Real-­‐Time  Processing  of  Mul*media  Data  

4/26/2012   ROMUS  project  @  VERDIKT     21  

ROMUS  project  

Page 22: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Mo4va4on  •  The  poten4al  to  use  mul4core  processing  and  new  heterogeneous  technology  (e.g.,  GPUs)  to  handle  mul4media  workloads  on  individual  peer  nodes  to  meet  new  demands  

•  Challenge:  difficult  to  program  because  of  heterogeneous  architectures  (e.g.,  data  parallelism  vs.  thread  parallelism)  

•  Exis4ng  frameworks  have  all  some  short-­‐comings;  do  not  meet  all  requirements  of  mul4media  data  processing  

4/26/2012   ROMUS  project  @  VERDIKT     22  

Page 23: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

P2G  main  features  •  The P2G framework allows developers to

express continuous multimedia workloads with fine grained parallelism in a single language

•  The runtime decides itself how a program should be partitioned, and which execution node should execute them

•  Open source at http://p2gproject.org 4/26/2012   ROMUS  project  @  VERDIKT     23  

Page 24: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Mo4on  JPEG  Experiment  

•  Continuous workload, CIF resolution •  DCT consumes most of the encoding time.

4/26/2012   ROMUS  project  @  VERDIKT     24  

Page 25: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Mo4on  JPEG  Results  

21  

11  

8  7   7  

6   6  5  

0  

2  

4  

6  

8  

10  

12  

14  

16  

18  

20  

22  

1   2   3   4   5   6   7   8  

4-­‐way  Core  i7  30  

15  

10  

8  

6  5   5   5  

0  

2  

4  

6  

8  

10  

12  

14  

16  

18  

20  

22  

24  

26  

28  

30  

1   2   3   4   5   6   7   8  

8-­‐way  Opteron  

Threads Threads

Tim

e (s

)

Tim

e (s

)

4/26/2012   ROMUS  project  @  VERDIKT     25  

Page 26: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Conclusions  (1)  •  The development of a simulation model to simulate and

evaluate novel adaptive and robust live P2P video streaming solutions is essentially important for long-term research in the field

•  It is essential to find solutions to important limitations of live P2P streaming technologies. Our basic solutions and findings could be inherited by other initiatives to build a more practical protocol taking other network metrics into account.

•  Tools and guidelines for how to design real time video streaming infrastructure adopting SVC techniques is crucial for being able to provide the best possible user experience with minimal visual distortion.

4/26/2012   ROMUS  project  @  VERDIKT     26  

Page 27: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Conclusions  (2)  •  Our results indicate that quality adaptation can

outperform constant low quality, but frequency and amplitude, as well as switching patterns are relevant

•  Multicore processor scheduling to handle multimedia workloads on individual peer nodes will be important in the future and may improve the multimedia experience of the user even further.

•  Our results shows the feasibility of providing a programming framework for automatic parallel, real-time processing of multimedia workloads exploiting heterogeneous multicore processors.

4/26/2012   ROMUS  project  @  VERDIKT     27  

Page 28: Improving the user experience of multimedia streaming services in highly dynamic environments, Frank Eliassen, UiO

Thank  you!    

Q&A  

4/26/2012   ROMUS  project  @  VERDIKT     28