TCP Slow Start, Congestion Avoidance, Fast Retransmit and Fast Recovery Algorithm - RFC 2001 W. Stevens Jan 1997

1
TCP Slow Start, Congestion Avoidance, Fast
Retransmit and Fast Recovery Algorithm - RFC
2001W. Stevens Jan 1997

• ???
• hjsohn_at_netlab.snu.ac.kr
• 2001? 3? 28? ???

2
Agenda

• Transport layer Introduction
• Slow Start
• Congestion Avoidance
• Fast Retransmission
• Fast Recovery
• References

3
Transport layer Introduction

• The Transport Layer
• provides reliable, cost-effective data transport
• from the source machine to the destination
  machine independent of the physical network
  (end-to-end layer)
• provides services to the application layer (or
  session layer if one exists)

4
Elements of Transport Protocols

• Differences from the Data Link Protocols
• Environment at the data link layer two routers
  communicate via a physical channel, at the
  transport layer via the entire subnet
• explicit addressing is required in the transport
  layer
• initial connection establishment is more
  complicated in the transport layer

5

• potential existence of storage capacity in the
  subnet
• produces a probability that a packet may be
  stored for a number of seconds and then delivered
  later.
• requires the use of special protocols for
  transport layer
• presence of a large and varying number of
  connections in the transport layer

6
Flow control

• Trade-off source buffering vs. destination
  buffering
• depends on the type of traffic
• interactive terminal buffering at the sender
• file transfer buffering at the receiver
• Dynamic buffer allocation
• variable-sized window
• The sender requests buffers, then the receiver
  grants as many of these as it can afford.

7
The first stage TCP

• How to control a flow?
• Used the advertised window size from receiver
  only
• Problems
• Performance
• Didnt concern about the network congestion
• High speed network connected slower network
• Higher retransmission rate and lower throughput
  specially between heterogeneous network
• Need to revise the flow control in TCP

8
(No Transcript)
9
Slow Start

• Self clocking
• New packets Injected into the network the rate at
  which ACKs returned by the other end
• Flow controlled by sender
• Congestion window exponential growth
• When starting or restarting
• Cwnd 1
• On each ACK for new data
• Cwnd Cwnd 1
• When sending,
• MIN Rwnd (Advertised window by receiver), Cwnd

10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
Congestion Avoidance

• How to deal with congestion?
• Congestion ? Packet loss ? Timeout
• Important to have good shaped Timers
• Duplicated ACKs
• Two components
• Signaling about the congestion
• Timeout or Duplicated ACKs
• Changing utilization factor
• Slow down when timeout occurred by changing Cwnd
  size

15

• On any timeout,
• Cwnd half the current window size
• On each ACK for new data,
• Cwnd Cwnd 1/Cwnd
• When sending,
• Min Rwnd, Cwnd

16
(No Transcript)
17
(No Transcript)
18
Slow Start and Congestion Avoidance

• Initialize
• Cwnd 1 and ssthreash 65535 byte
• Send
• MIN Rwnd, Cwnd
• When congestion occurrs
• ssthreash half of MIN Rwnd, Cwnd
• at least 2
• On ACK of new data,
• Cwnd Cwnd 1 in slow start
• Cwnd Cwnd 1/Cwnd in congestion avoidance

19
(No Transcript)
20
Fast Retransmission

• Duplicated ACKs
• Above 3 duplicated ACKs
• Send immediately lost segment
• Why do we retransmit immediately?
• In old TCP, send all segment again from dup. ACK
• If a receiver already had other segments?
• So, if receiver saves normal seg.s, and sends
  dup. ACK to sender, sender can retransmit at
  least one unacknowledged. seg. without waiting
  timer expiration

21
Fast Recovery

• When Fast retransmission occurred,
• Do not use slow start
• Use congestion avoidance
• Fast retransmission and fast recovery are usually
  implemented together

22

• When 3rd dup. ACK received,
• ssthresh half of MIN Rwnd, Cwnd
• Restransmit missing segment
• cwnd ssthresh 3 segsize
• Another dup. ACK received,
• cwnd cwnd segsize
• Normal ACK received,
• cwnd ssthresh

23
Single packet dropped in a window in Reno TCP
24
Two packets dropped in a window in Reno TCP
25
Two packets dropped in a window in Reno
TCP -Partial Acknowledgement
26

• Fast retransmission algorithm
• First appeared in 4.3BSD Tahoe release
• Fast recovery algorithm
• First appeared in 4.3 BSD Reno release