Congestion Control for Streaming Media

1
Congestion Control forStreaming Media

• Jae Won Chung

Committee Prof. Mark Claypool,
WPI Prof. Robert Kinicki, WPI Prof. Craig Wills,
WPI Prof. Kevin Jeffay, UNC-Chapel Hill
Ph.D. Dissertation
2
Internet Congestion Control (CC)

• Little Support From The Router
• Packet Drop Implicit Congestion Signal
• TCP Congestion Avoidance
• Respond to Congestion Signal

3
Efficient Congestion Control Feedback
Active Queue Management (AQM)
Queue

• Active Queue Management (AQM)
• Low Delay High Utilization
• Reduce Packet Loss
• Reduce Queue Overflow
• Explicit Congestion Notification (ECN)
• Stability and Configuration Issue

Outbound Link
Router
Inbound Link
4
Bandwidth Usage Control

• Bandwidth Control Mechanism
• Protect network and fairness
• Extend AQM Feature
• Scalability Issue

5
Efficient Bandwidth Usage Control
TCP-Friendly Transport Protocol
Queue

• TCP-Friendly Transport Protocol
• Average throughput does not exceed
• that of conforming TCP flow under the
• same network condition
• Application-Friendly also?

Outbound Link
Router
Inbound Link
6
Outline

• Internet Congestion Control
• Problem Statement
• The Crimson Architecture
• Aggregate Rate Control
• Summary

7
Problem Statement

• The Internet does not provide a
  streaming-friendly transport protocol (TCP is
  streaming-unfriendly).
• TCP API hides network information.
• TCPs reliable in-order delivery service incurs
  extra delays.
• The Internet stability is vulnerable to
  misbehaving high-bandwidth UDP streams.
• Streaming media applications often use UDP
  without a proper congestion control mechanism.
• Internet video has potentially high demand for
  bandwidth.
• ISPs provide broadband Internet connections (? 3
  Mbps).
• The Internet does not guarantee low transmission
  delays required by streaming media applications.
• Large queuing delays at IP routers in congestion.

8
The Crimson Architecture
TCP
TCP
Active Queue Management (IP Router)
Best-Delay-Effort
Protection
TCP
TCP
Multimedia Transport Protocol
Multimedia Transport Protocol
Bandwidth Controller
Congestion Controller
In
Filtered
Out
MTP
MTP
UDP
UDP
SFG
ARC
UDP
UDP

• MTP Multimedia Transport Protocol
• SFG Stochastic Fairness Guardian
• ARC Aggregate Rate Controller

9
Contributions (1 of 2)

• Internet measurement study
• Compare commercial Internet TCP UDP video
  streams
• Characterize streaming transport protocol
  requirements.
• Chung, 2003 Packet Video Workshop (PV)
• Chung, 2004 Kluwer Multimedia Tools and
  Applications
• Multimedia Transport Protocol (MTP)
• Modify TCP (Reno in NS) not to retransmit.
• Add streaming-friendly API.
• Chung, 2000 SCS Euromedia Conference
• Goddard streaming media client and server
• Design and implement a realistic streaming
  application in Network Simulator (NS).
• Simulates bandwidth estimation, media scaling and
  playout.

10
Contributions (2 of 2)

• Stochastic Fairness guardian (SFG)
• Design a lightweight bandwidth controller
  (statistical packet filter) that limits
  misbehaving high-bandwidth UDP traffic.
• Chung, 2000 NOSSDAV
• Chung, 2000 ACM Multimedia
• Chung, 2002 IEEE Symposium on Computers and
  Comm.
• Aggregate Rate Controller (ARC)
• Design a congestion controller that minimizes
  queuing delay while achieving high link
  utilization.
• Provide complete and practical configuration
  guidelines.
• Chung, 2003 Network Computing and
  Applications
• Chung, 2004 ACM SIGCOMM, (Poster)
• Integration of the Crimson components
• Evaluate Goddard over MTP with the Crimson
  (SFGARC).

11
Outline

• Internet Congestion Control
• Problem Statement
• The Crimson Architecture
• Aggregate Rate Control
• Summary

12
Random Early Detection (RED)

• RED (Floyd, 1993) 1G AQM congestion controller
• Uses a low pass filter on the queue length to
  detect and compute congestion notification
  probability (p).
• RED configuration problems
• Lack of configuration guidelines ? Queue law
  (Firoiu, 2000 Chung, 2003)
• Stability margin is small (Hollot, 2001) ?
  Gentle extension, self-configuring RED (add-hoc
  approaches).
• Proportional Integral (PI) AQM Controllers Apply
  control engineering paradigm to design AQM
• Large stability margin and prompt response.
• AVQ (Kunniyur, 2001),
• PI (Hollot, 2001) and REM (Athuraliya, 2001)

13
Aggregate Rate Control (ARC)

• Problem with current PI-based congestion
  controllers
• Difficult to configure PI controller for a
  time-delay system.
• Incomplete stability analysis measurement epoch.
• Queue sample-based control information
  acquisition
• ? Induce control noise when link is not fully
  utilized.
• Aggregated Rate Controller (ARC)
• Parameter reduced PI controller for TCP System
• ? Ease the control parameter configuration.
• Complete stability analysis
• ? Practical configuration guidelines
  recommendations.
• Rate-based control information acquisition
• ? Noise reduction flexible configuration
• ? Minimized queuing delay.

14
Rate-Based Implementation of PI

• Every d seconds
• 2 b ? 0
• Every packet arrival
• 3 if (uniform (0,1) ? p)
• 4 if (mark (packet) false)
• 5 drop (packet)
• 6 return
• 7
• 8 b ? b sizeof (packet)
• 9 if (enqueue (packet) false) drop
  (packet)

• p notification probability
• q queue length
• b bytes received this epoch

• C link capacity
• target utilization (C0/C)
• q0 target queue length
• d measurement interval
• ? virtual queue control const.
• ? queue control const.

15
TCP-ARC Feedback Control Model
(Hollot, 2001)
16
TCP-ARC Stability Conditions
17
ARC Configuration Guidelines

• Configure ARC (? /d ) for your
  average case lower boundary ( )
  condition.
• Set the measurement interval ( d ) close to the
  maximum expected system RTT (
  ).
• Check to see if the chosen ? meets the
  minimum stability condition.

18
Evaluation of ARC

• Evaluate ARC with other PI-based AQM congestion
  controllers (AVQ and PI) and Drop-Tail
• Over a wide range of realistic traffic mixes and
  loads.
• Show two simulation study results in this
  presentation.
• AQM Configurations
• AVQ
• ? 0.98, ? 0.15
• PI
• q0 50, ? 1.822 ?10-5, ? 1.816?10-5, ?
  170
• ARC
• ? 0.98, q0 0, d 1 sec, ? 1.42?10-5

19
Web Flash Crowd Simulation

• C 10Mbps
• Q 500 Kbytes
• RTLD 60, 1000 ms
• Nftp_fw 25, Nftp_bw 50
• Nweb 300 (OL0.25) ? 1300 (OL1.10) ? 300
• Nweb 10 sessions/min (from 100 sec)
• ? Nweb ? 10 sessions/min (from 6100 sec)
• Flash Rate (FIFA World Cup 98 Data)
• ? Peak Flash Rate 2M ? 10M reqs/h in 2 hours
• Web session setting (H-Campos, 2003)
• ? Sizeavg 5KB, Shape 1.2, Tavg_think 7sec
  (expo)
• Simulation time 12100 sec

20
Web Flash Crowd Queue Dynamics
21
Web Flash Crowd Data Losses
22
Light Traffic Load Simulation

• Simulation Objectives
• Compare PI-based AQMs on everyday light traffic
  load.
• Simulate sudden increase in delay (due to routing
  change).
• C 10Mbps
• Q 500 Kbytes
• Nftp_fw 5, Nftp_bw 10
• Nweb 300 sessions
• RTLD 100, 500 ms ? 2200, 2600 ms
• Increase the congested link RTLD 300 ms every
  200 secs.
• Average RTLD 300 ? 600 ? ? 2100 ? 2400 (ms)

23
Light Traffic Load Queue Dynamics
24
Light Traffic Load Throughput
25
Summary of ARC

• Minimize queuing delay at IP routers.
• Provide best-delay-effort Internet service to
  support streaming media and other delay sensitive
  applications.
• Practical and complete configuration guidelines
  and recommendations.
• Ease the controller parameter configuration
  through the PI parameter reduction.
• Provide configuration guidelines and
  recommendations that works for a wide range of
  traffic condition
• Robust congestion control performance over wide
  range of traffic conditions.
• Rate-based control information acquisition.
• High (flash crowd) and low (everyday) traffic
  loads.

26
Outline

• Internet Congestion Control
• Problem Statement
• The Crimson Architecture
• Aggregate Rate Control
• Summary

27
Conclusions (1 of 2)

• Internet measurement study
• Compare Internet TCP and UDP media streams.
• Characterize commercial video stream behavors.
• Identify streaming unfriendly features of TCP.
• Multimedia Transport Protocol (MTP)
• TCP-friendly TCP modification not to retransmit.
  
• API Streaming-friendly transport protocol.
• MTP offers streaming performance comparable to
  that provided by UDP, while doing so under a
  TCP-Friendly rate.
• Goddard streaming media client and server
• Design and build a realistic streaming
  application in NS.
• Simulates bandwidth estimation, media scaling and
  playout.

28
Conclusions (2 of 2)

• Stochastic Fairness guardian (SFG)
• Lightweight bandwidth controller that filters
  misbehaving high-bandwidth UDP traffic without
  flow monitoring.
• SFG outperforms other statistical traffic
  filters, and performs as well as bandwidth
  controllers using per-flow information.
• Aggregate Rate Controller (ARC)
• Minimizes queuing delay with high link
  utilization.
• Complete and practical configuration guidelines.
• Robust performance over wide range of traffic
  conditions.
• Evaluation of the Crimson network (SFG ARC)
• Goddard over MTP achieves the best stream
  quality.
• SFG controls high-bandwidth UDP Goddard streams.
• ARC minimizes the queuing delay.

29
Questions?
Thank You
30
Congestion Control forStreaming Media

• Jae Won Chung

Committee Prof. Mark Claypool,
WPI Prof. Robert Kinicki, WPI Prof. Craig Wills,
WPI Prof. Kevin Jeffay, UNC-Chapel Hill
Ph.D. Dissertation
31
Outline

• Internet Congestion Control
• Problem Statement
• The Crimson Architecture
• Aggregate Rate Control
• Summary
• ARC Leftovers

32
Block diagram of a TCP connection
Congested queue
__
1
N
?
TCP load factor
Control law (e.g. AQM)
Time Delay t
TCP window control
(Misra, 2000)
33
PID Controller for AQM

• PI Controller
• AVQ (Kunniyur, 2001)
• PI (Hollot, 2001)
• REM (Athuraliya, 2001)
• Other Designs
• SFC (Gao, 2003) PQPL Controller

34
Web Flash Crowd Queue CDF
35
Web Flash Crowd Throughput
36
Web Flash Crowd Service Time
37
Streaming Media Applications

• Popular use of streaming media in the Internet.
• Archived Jukebox, Video on Demand (VoD)
• Live Internet Radio, Internet TV
• Interactive Voice over IP (VoIP), Video
  Conferencing
• Streaming media applications estimate available
  network bandwidth to control the stream quality.
• Media scaling Choose a media of which the
  encoded bitrate is less than the available
  bandwidth.
• Monitor network information (loss rate, delay,
  throughput).
• Streaming media applications are sensitive to
  delay.
• There exist media playout deadlines to meet.
• Interactive streaming has even tighter delay
  requirements.