An Adaptive Congestion Control Mechanism for Video Multicast

1
An Adaptive Congestion Control Mechanism for
Video Multicast

• Yi Pan
• Yuji Imai

2
Necessity of Congestion Control of Video Multicast

• Users are now getting more and more interested in
  video data transmission.
• Video data consumes a lot of network resource and
  users want to find a way to transmit satisfactory
  video stream without severely interfering current
  data streams.
• Adaptive congestion control of video multicast
  will help to achieve this goal.

3
Current Research Related to Video Multicast
Congestion Control

• TCP-friendly congestion control
  (SF00,SF97,SCH98)
• current usage is only under unicast connections
• proved fairness with traditional TCP traffic
• existing protocol can be used to implement the
  congestion control (RTP/RTCP)

4

• Multi-layer encoded multimedia (SAMM, BTW)
• adaptive to heterogeneity of bandwidth in video
  multicast.
• the promising solution to encode video for
  multi-receivers
• Multi-layer multicast congestion control
  (INFO98, RLM)
• under classical multicast model
• need the assistance of intermediate nodes

5
Fundamental Problems in Multilayer Video Multicast

• Multicast branch pruning problem
• Behind a bottleneck, leaving of the only receiver
  subscribing to a layer should immediately
  alleviate the congestion at the bottleneck while
  in Internet Standard Multicast (ISM), it will
  take long time.

6

• Receiver interfering problem
• Behind a bottleneck, a receiver who adds to a new
  layer may cause congestion at the bottleneck.
  That may cause other receivers react by lowering
  their subscribing level which is not desirable
  feature of the video multicast.

7

• Finding of potential bandwidth
• Behind a single bottleneck, when there are
  receivers subscribing to different layers of
  video stream, a receiver adding to a new layer
  may not need extra bandwidth at the bottleneck,
  which means there is potential bandwidth exists
  at the bottleneck.

8

• Multicast group managing problem
• the problem is with the number of the multicast
  group needed for multi-layer video multicast.
• Fixed number of multicast group may have some
  empty groups which waste network resource.
• Adaptive number of multicast group will take long
  time to establish/delete a new multicast group
  under ISM architecture.

9
Goals of Video Multicast Congestion Control

• Quick convergence of video congestion control
• Adaptiveness of video congestion control to
  receivers in various situations.
• Fairness of congestion control of video
  transmission competing with other traditional
  methods (TCP-like)
• Those issues are stated in SF00,INFO98,SCH98
  .

10
Our Proposal of Adaptive Congestion Control over
XCAST

• Combination of good techniques
• there exists unicast congestion control which
  fixes the too sensitive feature of TCP.
• multi-layer encoded video stream satisfies
  heterogeneous receiver connections
• XCAST solves the problem of joining/leaving
  overhead in intermediate routers
• Sender-driven multicast congestion control has
  benefits of fine-tuned, quick convergence,etc.

11
Our Solution

• Sender-driven congestion control for multi-layer
  encoded video stream.
• Receiver make estimation of its available
  bandwidth and send the feedback to sender.
• Sender adjusts the number of layers and the rate
  of each layer based on the feedback messages.
• Make the receivers in the same layer as an XCAST
  group.
• With XCAST, its now possible to use multiple
  dynamic flows to transmit multiple layers in
  sender-driven scheme.

12

• Integrate Traffic Control to congestion control
  for video transmission.
• Equation-based Congestion Control
• Receiver reports estimated Round Trip Time and
  estimated loss rate p.
• Sender calculates transmit rates based on a
  certain equation, group receivers in different
  layers, and send out video stream
• FEC congestion control
• using FEC encoding to let receivers recover the
  loss

13

• Employ XCAST to solve fundamental problems with
  multicast video transmission.
• XCAST groups are explicit and branch pruning can
  be done within a round trip time (as long as the
  quit message arrives the sender)
• We can employ intelligent aggregation algorithms
  to make optimal grouping of receivers in dynamic
  number of layers
• XCAST does not need the assistance of
  intermediate nodes

14
Simulation and Expected Results

• Simulation Scenario
• compare our scheme with the existing multicast
  congestion control in
• convergence time for congestion control
• fairness between different users
• adaptive to different connections
• overhead of congestion control
• average goodput for receivers
• fairness when competing with other flows using
  TCP-like congestion control

15

• Expected results are
• quick convergence time for congestion control
• adaptive to heterogeneous connections under
  limited number of receivers
• low overhead incurred in network
• higher average goodput for receivers
• good fairness competing with traditional TCP-like
  congestion control mechanisms

16
Technical Details

• Equation Based Congestion Control
• basically the equation based congestion control
  is based on the following equation
• So we propose the receiver send feedback to
  sender including the RTT and loss rate p it
  estimates so that the sender can calculate the
  estimated acceptable rate.

17

• Dynamic receiver grouping
• Transmission rate to each receiver is calculated
  by sender and the sender need to optimally
  partition receivers to groups with a specific
  transmission rate for each group.
• Algorithm to find the optimal partition of
  receivers should achieve the goals of reducing
  the number of groups and maximize total goodput.

18

• Receiver interference problem
• two approaches
• Adjusted equation based control adjust the
  equation let higher layer receiver to be more
  sensitive to loss and RTT.
• Employ deaf period for lower layer receivers
  after a new higher layer receiver being added,
  allow lower layer receivers to be deaf to loss
  and RTT for a while.

19

• Approach 1
• receivers make estimation of RTT and loss rate p
  based on different transmission rate
• Advantage no need for sender to be aware of
  where the bottleneck is.
• Disadvantage the lower layer receivers still
  suffer for a short time, though less influenced.

20

• Approach 2
• Notify affected receivers to be deaf for a
  period of time to let the newly added high layer
  receiver quiet down.
• Advantage Lower layer receivers will not be
  affected at all.
• Disadvantages
• Sender needs to be aware of where the bottleneck
  is and who are behind the bottleneck.
• Necessary adjustment of rate may be delayed by
  the deaf period.

21

• One additional notation of receiver interference
• Since we transmit different layer in different
  XCAST group, that means we can use different
  flows to transmit different layers.
• With the help of current DiffServ Architecture,
  we can easily isolate those flows so that
  different layers will have much less impact on
  each others.

22

• Potential bandwidth finding
• May need to build tractable list of receivers in
  each layer. (need to discuss)
• probing packets for each destination are sent out
  periodically.
• The branching records are kept in the probing
  packets and are sent back to sender.
• Sender uses those branching records to build up
  the sink tree.
• Only need max(RTT) of all the receivers to build
  that sink tree.

23

• As long as we get such information, it is easy to
  find potential bandwidth in bottleneck. So its
  safe to add this new receiver for high layer.
• This technique is kind of congestion avoidance
  technique, not congestion control.

24
Future Research

• Prove the effectiveness of the architecture in
  future Internet providing DiffServ services for
  QoS requirements
• Future extension
• explore impact of combining XCAST and resource
  allocation in this architecture.
• explore impact of QoS routing and its impact on
  this architecture.

25
Reference List

• SF00 S. Floyd, M. Handley, J. Padhye, J.
  Widmer, Equation-Based Congestion Control for
  Unicast Applications, SIGCOMM 2000
• SF97 J. Mahdavi, S. Floyd, TCP-Friendly
  Unicast Rate-Based Flow Control, Jan 1997
• SCH98 D. Sisalem, H. Schulzrinne, The
  Loss-Delay Based Adjustment Algorithm A
  TCP-Friendly Adaptation Scheme, Proc. of
  NOSSDAV'98, July, Cambridge, UK
• SAMM B. Vickers, C. Albuquerque, T.Suda,
  Source Adaptive Multi-Layered Multicast
  Algorithms for Real-time Video Distribution,
  IEEE ToN, 1999
• INFO98 L. Vicisano, J. Crowcroft, TCP-like
  Congestion Control for Layered Multicast Data
  Transfer Proc. of IEEE INFOCOM'98.

26

• BTW J.C. Bolot, T. Turletti, I.Wakeman,
  Scalable Feedback Control for Multicast Video
  Distribution in the Internet, Proc. of ACM
  SIGCOMM, pp58-67, Aug. 1994.
• RLM S. McCanne, V. Jacobson, M. Vetterli,
  Receiver-Driven Layered Multicast, Proc. of ACM
  SIGCOMM, pp117-130, Aug. 1996.