CoDiO Scheduling for Low Latency Streaming

1
CoDiO Scheduling for Low Latency Streaming

• Eric Setton, Xiaoqing Zhu and Bernd Girod
• Information Systems Laboratory
• Stanford University

2
Presentation Outline

• Congestion-distortion optimized (CoDiO)
  scheduling
• Motivation
• Description of the algorithm
• Low-complexity CoDiO scheduling
• Application of CoDiO scheduling
• Sender-receiver wired network
• Sender-receiver multihop wireless adhoc network

3
Related Work

• TCP-friendly rate-control
• Indicates average rate as a function of collected
  statistics
• Does not indicate any particular schedule
• Rate-Distortion Optimized Scheduling
• Formalization of the multimedia scheduling
  problem
• Adapted the framework to video streaming
• Large gains compared to sequential scheduler

Floyd et al., 1997
Chou and Miao, 2001
Chakareski and Girod, 2002-5
Kalman and Girod 2003-5
4
Parameterized Delay Distribution
pdf
C
delay
5
Congestion-Distortion Optimized Scheduling

• Congestion used as a new metric
• Adaptive to network conditions
• Reflects the impact of a sender
• Blows up close to capacity
• Principle of CoDiO

Decide which packets to send (and when) to
maximize picture quality while minimizing
network congestion
6
Predicting Video Distortion

• Probability of packet arrival may be estimated by
  a parameterized delay distribution
• Compute expected distortion of each frame

P
P
I
P
P
P
Kalman, Ramanathan and Girod, 2003
7
Estimating Self-Congestion

• Congestion defined as average end-to-end delay

R(t4)
R(t3)
decrease rate C, the capacity
R(t0)
R(t1)
Size of the bottleneck queue
average queue size
8
Finding the Best Transmission Order

• Optimal schedule is
• Why is this difficult ?
• Large search space
• Tight coupling between schedules
• How to solve this problem ?

9
Find a Good Schedule at Random!
P
I
B
P
P
B
B
10
Low Complexity CoDiO Scheduler

• Select iteratively most important video packet to
  transmit or retransmit
• Space transmissions to limit congestion over
  bottleneck
• Sufficiently simple to be run at each node

11
Wired Network Scenario
Random cross traffic
High bandwidth links
Low bandwidth uplink
Video traffic
C
Acknowledgments
12
CoDiO Scheduling Performance
Mother Daughter sequence
News sequence
30
25
Simulations over ns-2 Packet loss rate
2 Bandwidth 400 kb/s Propagation delay 50ms
13
CoDiO
ARQ
H.264 _at_250 kb/s Bandwidth 400 kb/s 2 packet
loss0.6 sec latency
14
CoDiO vs. RaDiO
40
Sequence Mother Daughter Packet loss rate
2 Bandwidth 400 kb/s Propagation delay 50ms
15
Wireless Network Scenario

• 15-node wireless network simulated in ns-2
• Mobile nodes, speed 2.5 m/s
• 802.11b MAC protocol, ad hoc mode
• DSR routing
• 1 sender-receiver pair
• Receiver sends ACKs
• Video streaming
• Foreman CIF sequence, encoded with H.264
• 4 encoding rates
• 915 kbps, 495 kbps, 275 kbps, 120 kbps
• I B B P coding structure, GOP length 15
• Short playout deadlines between 350 ms and 700 ms
• Previous frame concealment

16
Network simulation
17
Scheduling Schemes

• CoDiO light TFRC throughput estimator
• TFRC estimator determines how to select the rate
• CoDiO light scheduler controls transmissions and
  retransmissions
• Sequential scheduling TFRC throughput estimator
• Quality is determined similarly as in the other
  scheme
• Retransmission based on fixed timeout

18
Performance ComparisonLoss Rate
19
Performance ComparisonVideo Quality
20
450 ms playout deadline
Sequential 35 dB
CoDiO light 37 dB
21
Summary

• Presented the idea of CoDiO scheduling
• Extended the scheduling framework to
  throughput-limited network
• New metric Lagrangian cost of congestion and
  distortion
• Simple CoDiO-light scheduler
• Determines the best packets to send
• Adaptive to varying network conditions
• Outperforms sequential scheduler