Network CC1 2

1
Network ?? ??(CC1 2)

• 2000.3.17
• ? ? ?
• Data Networks Lab.

2
TCP Slow Start, Congestion, Avoidance, Fast
Retransmit, and Fast Recovery Algorithms

• RFC 2001, January 1997
• W. Stevens
• TCP properties
• ACK clocking (or self clocking)
• window based rate control (cwnd)
• cumulative ACK
• Slow start
• Send 2 packets per an ACK
• exponential growth of cwnd
• Getting to equilibrium

3

• Congestion Avoidance
• Send (1 1/cwnd) packets per an ACK
• linear growth of cwnd
• additive increase / multiplicative decrease
  policy
• Adapting to the path
• Assumption packet loss is from congestion
• loss detection - timeout and duplicate ACKs

4

• Algorithm Slow start and Congestion Avoidance
• Initialization cwnd 1 pkt, ssthresh 65535
  bytes
• When sending, send min(cwnd, awnd) packets
• When new data is acknowledged
• if cwnd lt ssthresh cwnd cwnd 1
• if cwnd gt ssthresh cwnd cwnd 1 / cwnd
• When congestion
• by timeout ssthresh cwnd / 2, cwnd 1
• by duplicate ACKs ssthresh cwnd / 2

5
packets
ACKs
ssthresh
cwnd
6

• Fast Retransmit
• Retransmit what appears to be the missing packet
• based on the information of duplicate ACKs
• Avoid lengthy timeout occurring
• TCP may generate an immediate ACK (a duplicate
  ACK) when an out-of-order packet is received
• Fast Recovery
• After fast retransmit, Congestion Avoidance is
  performed
• Allow high performance under moderate congestion

7

• Algorithm Fast Retransmit and Fast Recovery
• When the third duplicate ACKs is received
• ssthresh cwnd / 2
• retransmit the missing packet (Fast Retransmit)
• cwnd ssthresh 3
• On each additional duplicate ACK
• cwnd cwnd 1
• transmit a packet if allowed by cwnd
• When the ACK that acknowledges new data arrives
• cwnd ssthresh
• end Fast Recovery and continue Congestion
  Avoidance

8
Fast Retransmit and Fast Recovery
9
Simulation-based Comparisons of TCP Tahoe, Reno,
and SACK TCP

• Computer Communication Review, July 1996
• Kevin Fall and Sally Floyd
• Tahoe TCP
• Slow Start, Congestion Avoidance, Fast Retransmit
• 4.3 BSD
• Reno TCP
• Fast Recovery
• 4.4 BSD

10
(No Transcript)
11
(No Transcript)
12

• New-Reno TCP
• Change to the Reno TCP
• enhance Fast Recovery having multiple packet
  losses
• Use the information of a partial ACK
• Many variants
• RFC 2582 The NewReno Modification to TCPs Fast
  Recovery Algorithm
• Internet Draft The Rate-Having Algorithm for TCP
  Congestion Control

13
(No Transcript)
14

• SACK TCP
• RFC 2018 TCP Selective Acknowledgment Options
• Add Selective Acknowledge information in TCP
  header as option field
• SACK block (maximum three blocks)
• Both a sender and a receiver should support SACK
• hard to be deployed

15
(No Transcript)
16
Random Early Detection Gateways for Congestion
Avoidance

• IEEE/ACM Transaction on Networking, August1993
• Sally Floyd and Van Jacobson
• Design Guideline(1) - main goal
• Avoid congestion by controlling the average queue
  size
• detect incipient congestion
• decide which connections to notify of congestion
  at the gateway

17

• Design Guideline(2)
• Avoid a bias against bursty traffic
• Avoid the global synchronization
• Control the average queue size even in absence of
  cooperating sources

18

• Algorithm
• For each packet arrival
• calculate the average queue size avg
• exponential weighted moving average
• avg (1-Wq) avg Wq q
• if avg lt minth ? accept it
• else if avg gt maxth ? drop/mark it
• else if minth lt avg lt maxth ?
• drop/mark it with calculated probability Pa
• Pa increase linearly in the rage of 0 lt Pa lt
  Pmax

19

• Setting parameters
• minth ? utilization
• maxth - minth ? global synchronization
• Pmax ? about one drop out of 1 / Pmax ptks
• Wq ?
• if too small, unable to detect incident
  congestion
• if too large, can not accept packet burst

20

• Evaluation
• Congestion Avoidance
• Appropriate time scales
• No global synchronization
• Simplicity
• Maximize global power
• Fairness
• Appropriate for a wide range of environments

21
Additional References

• Tahoe and Reno TCP
• Congestion Avoidance and Control, Van Jacobson
  and Michael J. Karel, Computer Communication
  Review, August 1988
• TCP/IP Illustrated, Vol. 1 The Protocols,
  Addison-Wesley, 1994
• TCP Congestion Control, M.Allman, V.Paxson, and
  W.Stevens, RFC2581
• NewReno TCP
• Improving the Start-up Behavior of a Congestion
  Control Scheme for TCP, J. Hoe, SIGCOMM, August
  1996
• SACK TCP
• Forward Acknowledgement Refining TCP Congestion
  Control, M. Mathis and J. Mahdavi, SIGCOMM,
  August 1996
• RED
• URL http//www.aciri.org/floyd/red.html