TCP Throughput, TCPfriendly and ratebased protocols Announcements: Homework

1
TCP Throughput, TCP-friendly and rate-based
protocolsAnnouncements Homework 4, project 1
EECS 325/425, Fall 2005 October 17
2
TCP throughput

• RTT round-trip time
• RTO 2RTT
• p loss rate
• B packet size
• T throughput

3
Why TCP-friendly protocols?

• The Internet is experiencing increasing volumes
  of traffic that has delay and reliability
  requirements which are unsuited to TCP.
• Streaming media
• UDP with no effective congestion control
• TCP with large variations in throughput
• Telephony and Video-telephony.
• UDP with no effective congestion control
• TCP is a reliable protocol. To achieve
  reliability while performing congestion control
  means trading delay for reliability. Telephony
  and streaming media have limited delay budgets -
  they dont want total reliability.
• Our question is how best to support such
  applications in the current best-effort Internet?
• Others are looking at providing ways to reserve
  network resources to better support these
  applications.

4
Window-based versus rate-based

• Window-based
• Sending rate is determined by a window size
• Window size increases/decreases when network
  conditions change (need to find out)
• Clocked (window adjustment on ACKs and
  losses/timeouts)
• Rate-based
• Sending rate is calculated based on information
  from the network or the other end host.

5
Examples of TCP-friendly protocols

• TCP-friendliness Any alternative congestion
  control scheme needs to co-exist with TCP in FIFO
  queues in the best-effort Internet. It must
  impose the same long-term load on the network
  at the same time it must not cause TCP any
  unexpected problems on the short term.
• Window-based General AIMD and other non-linear
  controls
• How to provide TCP-friendliness? Change the
  parameters of the increase/decrease rules.
• Rate-based TCP-friendly rate control
• How to provide TCP-friendliness? No explicit
  increase/decrease rules, but need the TCP
  throughput equation to calculate the rate.

6
TCP friendly rate control RFC 3448

• Materials based on a technical publication TCP
  Friendly Congestion Control for Unicast
  Applications, by Sally Floyd etc
• More resources at
• http//www.aciri.org/tfrc

7
General ideas and problems

• Use a model of TCPs throughout as a function of
  the loss rate and RTT directly in a congestion
  control algorithm.
• If Tx rate is higher than given by the model,
  reduce the transmission rate to the models rate.
• Otherwise increase the transmission rate.
• Key problems
• How to measure the loss rate?
• How to measure RTT?
• How to damp these values appropriately?
• How to use them in a control loop?

8
Loss rate measurement

• Average loss interval (ALI) is the weighted
  average of the last N intervals between losses.
• Loss event probability 1/ALI

9
Loss rate measurement

• More details in loss rate
• Need to treat interval since most recent loss
  differently.
• Need to ignore more than one loss in an RTT. This
  is essentially what TCP does (unless it results
  in a timeout, which we model separately).

• How many intervals? What distribution of weights?
  
• N8, and weight (1.0, 1.0, 1.0, 1.0, 0.8, 0.6,
  0.4, 0.2)
• Clear tradeoff between responsiveness and
  smoothness.

10
Effects of TFRC
11
Potential problems of TFRC?

• Although there are still some debates, TFRC has
  been used by some multimedia applications
• Stability issue
• Unfairness problem (against TCP)