MINESTRONE: Mobile Infrastructure Enablers for Streaming Optimization

1
MINESTRONE Mobile Infrastructure Enablers for
Streaming Optimization New Services

• Danjue Li, Chen-Nee Chuah, Gene Cheung, S. J.
  Ben Yoo
• Dept. of ECE, University of California, Davis
• HP Lab, Japan, HP Japan Ltd.
• http//www.ece.ucdavis.edu/rubinet/minestrone.html
  

2
Motivation
Video server

• Challenges
• Losses
• Latency/Jitter
• End host network heterogeneity
• Mobility

3
MINESTRONE Architecture
Type I Enabler
D
Routing Link-state announcement
Routing Proxy
Type II Enabler
Streaming proxy
A
Access point
Router
Video server
B
B
Access point
Related publication G. Cheung, C. N. Chuah,
and D. Li, "Optimizing Video Streaming Against
Transient Failures and Routing Instability," IEEE
ICC, June 2004.
4
Hot spot Wireless LANs

• Wireless Internet access inside the
• hotel lobbies, conference rooms, etc.
• Wireless at the airport, café
• Wireless in the libraries, dorms
• Imagine
• Starbucks Latte Laptops the SuperBowl
• Smart classroom

5
Outline
Goal Reliable video streaming service over WLAN

• MUVIS Multi-Source Video Streaming Service over
  WLAN
• System architecture
• Problem Formulation Proposed Solution
• Multi-point-to-point (MP2P) packet scheduling
• Combinatorial rate-distortion packet scheduling
  (RaDiO)
• Caching strategies
• Performance Evaluation
• Summary

6
Streaming video over WLAN

• Challenges

• Our Approach
• Multiple-source streaming to leverage
• server/peers resources
• Rate-Distortion Optimization

• High bit error rate
• Contention
• End host mobility

• Related work
• Sender diversity Nguyen02
• Rate-distortion Optimization
• Rate-distortion optimized streaming of
  packetized media Chou01
• Hybrid receiver/sender driven streaming scheme
  Chakareski02
• Server diversity in rate-distortion optimized
  media streaming
• Chakareski03

7
Traditional video streaming over WLAN
11430, I need Star Wars!!
Query
Reply
Internet
11540, I need Star Wars!!
Access Point
Streaming server
11510, I need Star Wars!!
8
MUVIS MUlti-source VIdeo Streaming
Streaming Proxy
I Do!!!
R-D preamble
Reply
Query
Internet
11540, I need Star Wars!!
Streaming Server
Do you have Star Wars ?
I Do!!!
9
MUVIS- Content Discovery
Content table
4
Streaming Proxy
2
3
R-D preamble
SREQ(UID, VFN, SR)
1
2
Internet
3
3
Streaming Server
PACK(UID, CL, AB)
3
2
MREQ(UID, VFN, SR)
UID user ID VFN video file name SR streaming
rate CL content list AB allocated
bandwidth SSID streaming session ID
10
MUVIS- Sender Selection
Mobility tracking table
4
Streaming Proxy
6
PUPD(CL, UID, AB)
Internet
5
Streaming Server

• Peer selection criteria
• available contents
• bandwidth allocated for streaming
• mobility level

11
Outline
Goal Reliable video streaming service over WLAN

• MUVIS Multi-Source Video Streaming Service over
  WLAN
• System architecture
• Problem Formulation Proposed Solution
• Multi-point-to-point (MP2P) packet scheduling
• Combinatorial rate-distortion packet scheduling
  (RaDiO)
• Caching strategies
• Performance Evaluation
• Summary

12
MP2P Packet Scheduling
pre-encoded media units
server

transmission opportunities
time

• At each transmission opportunity, the proxy will
  decide
• Which data unit to request for (re)transmission?
• From which sender ?

13
Source model

• Directed Acyclic Graph (DAG) Chou01

(a)
I
P
P
P
P
P
I
P
P
P
P
P

(b)

• Each data unit i is labeled by
• ni size in RTP packet of data unit Dui
• Ti decoding deadline for DUi
• Di distortion reduction if DUi is decoded

14
Network Model Constraints

• Network model
• Independent time-invariant packet erasure channel
  with
• random delays.
• Packet loss emn
• Delay for both forward/backward channel

rate
link capacity

• Rate constraints

packet size
loss event rate
round-trip time
TCP retransmission timeout
15
Performance Modeling

• System model

Prob. that a request sent to sender j at time T
will result in the requested data unit
successful arrival at the client by time T.
16

Performance Modeling (cont)

• Probability of successfully receiving DUi
  before its
• deadline Ti

Transmission History
Transmission decision at time to
17
Rate-distortion optimized packet scheduling

• Objective functions

• Constraints
• Maximum sending rate between proxy and node j
• min (Oj, Wj)

From TFRC
decoupling
Bandwidth allocated for streaming by sender j
Step2 Data selection via P2P RaDiO
Step1 Sender Selection via Asynchronous Clocks
18
Sender Selection via Asynchronous Clocks

• Transmission token
• Required for transmitting
• data over one link

Second leg

• Clocks Timer with period

First leg
First leg proxy (A) -sender j
Second leg proxy (A)-client (C)

• Clock j wakes up Proxy gets transmission token
  for link Aj/AC

• Communication path proxy-sender j-proxy-client

• Transmission opportunity for j Trans. Token Aj
  Trans. Token AC

19
Data Selection via P2P RaDiO Framework

• Data unit selection algorithm Chou01
• Benefit of delivering DUi at the optimization
  instance

Expected reduction of distortion if DUi is
received correctly
Increase in likelihood of successfully
delivering DUi
where,

• Chou01 P. A. Chou and Z. Miao,
  Rate-distortion optimized streaming of
  packetized media, February 2001. Microsoft
  Research Technical Report MSR-TR-2001-35

20
Cache Strategies

• Why caching?
• Local retransmission.

Second leg

• Cache strategies
• Simple caching simple fetching (SCSF)
• Distortion minimized caching strategies (DMSC)

First leg

• Transmission token re-interpretation
• SCSF
• Same to the case when there is no
  cache in the system
• DMSC
• Transmission opportunity Trans.
  Token Aj/AC

21
Performance of Cache-Enabled MUVIS System

• Case one SCSF is implemented

• For DUi that has not been cached

Same as the case where there is no cache in the
system

• For DUi that has been cached

Transmission History
Transmission decision at time to
22
Performance of Cache-Enabled MUVIS System

• Case Two DMSC is implemented

• For DUi transmitted between the proxy and sender
  j

Transmission decision at time to, cij
Transmission History

• For DUi transmitted between the proxy and the
  client

23
Outline
Goal Reliable video streaming service over WLAN

• MUVIS Multi-Source Video Streaming Service over
  WLAN
• System architecture
• Problem Formulation Proposed Solution
• Multi-point-to-point (MP2P) packet scheduling
• Combinatorial rate-distortion packet scheduling
  (RaDiO)
• Caching strategies
• Performance Evaluation
• Summary

24
Experimental Setup

• Examined six streaming schemes

• Source
• Sequence Foreman
• H.263 , QCIF, 300 frames, 30fps, 120kbps,
  I-frame freq. 1/25
• Quality measurement Peak-Signal-Noise-Ratio
  (PSNR)
• Buffering delay 1 second

25
Experimental Setup (cont.)

• Simulator
• Network Simulator (NS) 2.27

• Network parameters
• Wired links

• Wireless connections

26
Experimental Setup (cont.)

• Simulated realistic peer mobility patterns
  Balazinska03

• Session duration Vj
• The uninterrupted amount of time that a user
  stays associated with an AP.
• Equivalent to the persistence metric, which
  follows a power law distribution

• Revisit interval Rj
• The amount of time before next visit of the
  mobile user.
• Prevalence metric the fraction of time that a
  user spends
• with a given AP

27
Simulation Experiment I
Set I Multi-source streaming No bottleneck
Figure 1. Instantaneous PSNR in one typical run
(Foreman)
28
Simulation Experiment I (cont.)
Figure 2 Cumulative Distribution Function (CDF)
of across-run average PSNR (Foreman)
29
Simulation Experiment I (cont.)
Table 1 Average PSNR (Foreman)

• Averaging method over run-time 50 runs with
  different
• random seeds.
• Multi-source streaming can perform better by
  leveraging
• peer resources.

30
Simulation Experiment II
Set II Multi-source streaming serverAP is
bottleneck
No bottleneck (SA 150kbps)
Server-AP is Bottleneck (SA 100kbps)
Figure 3 PSNR variation over multiple runs
(Foreman)
31
Simulation Experiment II (cont.)
Table 2 Average PSNR when the server-proxy
connection is the bottleneck (Foreman)

• Averaging method over run-time 50 runs with
  different
• random seeds
• Multi-source streaming can compensate the
  congestion-
• caused quality degradation by leveraging peer
  connections.

32
Simulation Experiment III
Set III Cache effect
Figure 4 Effect of different caching strategies
(Foreman)
33
Simulation Experiment IV
Set IV PLR sensitivity
Figure5 Average PSNR vs. Varying Wired Loss
(Foreman)
34
Summary

• Proposed a multi-source video streaming scheme to
  leverage
• both media server and mobile peers in WLANs.
  
• Formulated the streaming process as a
  combinatorial packet
• scheduling problem and solved it by
  introducing asynchronous
• clocks and a point-to-point rate-distortion
  framework.
• Evaluated the schemes via simulation studies
• Multi-source streaming offers better performance
  than the
• single sender.
• Having cache in the proxy will increase the
  performance.
• DMSC performs better than SCSF.

35
Future Work

• Optimize streaming performance for multiple
  wireless clients.
• Design a better rate control for streaming over
  WLAN.
• TFRC is not sufficient
• Related work Chen04Markopoulou04
• Design a better peer selection algorithm.
• Resilient to peer failure
• Peer goodness (content, mobility, channel
  quality, energy,)
• Prototype MINESTRONE.
• Study impact of mobility
• Energy issues

36
References

• Nguyen02 T. Nguyen and A. Zakhor, Distributed
  video streaming over the internet, in SPIE
  Multimedia Computer and Networking, Jan. 2002.
• Chakareski02 J. Chakareski, P. A. Chou, and B.
  Girod, Computing rate-distortion optimized
  policies for hybrid receiver/sender driven
  streaming of multimedia , in Asilomar Conference
  on Signals, Systems, and Computers, November
  2002.
• Chakareski03 J. Chakareski and B. Girod,
  Server diversity in rate-distortion optimized
  media streaming, in ICIP, September, 2003
• Floyd00 S. Floyd, M. Handley, J. Padhye, and J.
  Widmer, Equation based congestion control for
  unicast applications, SIGCOMM 2000
• Balazinska03 M. Balazinska and P. Castro,
  Characterizing mobility and network usage in a
  corporate wireless local-area network, in
  MobiSys, San Francisco, CA, May 2003
• Chen04 M. Chen and A. Zakhor, "Rate Control for
  Streaming Video over Wireless" in INFOCOM 2004
• Markopoulou04 A. Markopoulou, E. Setton, M.
  Kalman, J. Apostolopoulos, Wise Video Using
  In-band Wireless Loss Notification to Improve
  Rate-Controlled Video Streaming'', IEEE ICME,
  June 2004.

37
Questions
http//www.ece.ucdavis.edu/rubinet/minestrone.html