Author Luigi Rizzo

1
PGMCC A TCP-friendly Single-rate Multicast
Congestion ControlScheme

• Author - Luigi Rizzo
• CS 590F Seminar
• Presenter - Ameya Varde

2
Terminology

• Multicast - data transfer from a single sender to
  multiple receivers.
• Group - set of receivers.
• Acker - slowest receiver within the group.
• Single rate multicast -
• All receivers get the same data rate.
• Sender adapts to the slowest receiver.
• A single slow receiver can drag down the data
  rate for the whole group.
• Scheme is relatively easy and consumes smaller
  bandwidth.
• TCP friendly - sender transmits no faster than
  the TCP fair rate at the slowest receiver.

3
Objective of PGMCC

• Prevent congestion in the multicast session.
• To do so, we need to -
• Find the slowest receiver (acker).
• Ackers may change over time.
• Transmit no faster than the acker can accept.

4
Working of PGMCC

• Acker election and tracking.
• PGMCC makes use of NAK's sent by receivers to
  decide the acker.
• Congestion control.
• The acker then supplies ACK's to the sender as an
  indication of successful data packet transfer.
• A window-based congestion controller scheme
  similar to the one used in TCP is used between
  the sender and acker.

5
Packet Formats
6
Acker Election and Tracking

• Two parameters (present in the NAK) decide the
  acker.
• rxw_lead
• Sender computes RTT.
• RTT (most recent seq_no) rxw_lead
• rx_loss
• Receiver computes loss rate p.
• p W x pprev (1-W) x seq_no
• Throughput T ? 1 / (RTT x ?p)
• Acker is the receiver with the lowest T.
• Receiver i becomes the new acker if
• Ti lt c x Tacker c ? 0.6-0.8
• c is used to reduce the number of acker switches.

7
Congestion Control

• Sender maintains two state variables,
• window W and token count T.
• W is an estimate of the number of packets in
  transit.
• T regulates the generation of data packets.
• on session start, W1, T1.
• on transmit, TT-1 (i.e. consume one token).
• on ACK, WW 1/W, TT11/W.
• on NAK, WW/2, T unchanged for next W/2 ACK's.

8
PGMCC enabled routers

• NAK suppression.
• Filters repeated NAK's coming from receivers to
  sender.
• Stores state.
• Maintains rx_id, rxw_lead.
• Selective Forwarding.
• Forwards re-transmitted packets only to expecting
  receivers
• Problem
• Could result in sender not receiving a NAK from a
  potential acker
• Solution
• Maintain rxw_loss in router as well.
• For a repeated NAK, compare incoming rxw_loss
  with the stored one. If lesser, forward NAK to
  sender.

9
Experimental Results
10
Advantages of PGMCC

• RTT computed without the need for globally
  synchronized clocks.
• Sender is not hindered by acker switches.
• Works fine even for unreliable protocols.
• Works with or without router support.
• Responsive to uncorrelated losses .
• Sender uses loss rates (rx_loss) computed by
  receivers.
• Avoids "drop-to-zero" problem.
• Achieves intra protocol and inter protocol
  fairness.
• Scaleable.

11
Disadvantages and Future work

• Disadvantages
• Requires loss reports from receivers.
• Most calculations are approximate.
• Future work
• Use of alternate methods to compute loss rate at
  the receivers.
• Computing throughput more accurately.
• Use adaptive rather than a fixed value (6) for
  slow start threshold.

12
Conclusions

• PGMCC has been implemented by the author in the
  PGM protocol on BSD Unix.
• Achieves the desired results in the
  configurations that were tested.
• Author believes that PGMCC can be reliably used
  for multicast communication over the Internet on
  a small-medium scale.