TCP Congestion Control Mechanisms II

1
TCP Congestion Control Mechanisms II

• For CSE5346/4346 Computer Networks II
• Qingchun (Dustin) Jiang
• Spring 2005

2
Outline

• DCCP
• Limited Slow-Start
• NewReno
• A Conservative Selective Acknowledgment (SACK)
• High Speed TCP
• Project

3
PART I

• Datagram Congestion Control Protocol

4
What is DCCP

• A congestion-controlled, unreliable flow of
  datagrams
• UDP plus congestion control"
• Also a modern transport protocol
• Partial checksums, mobility, DoS resistance, fast
  connections, . . .

5
Targeted Applications

• Long-lived flows that prefer timeliness to
  reliability
• Streaming media, Internet telephony, on-line
  games, . . .
• UDP not congestion controlled, apps must
  implement CC
• Apps want
• Buffering control don't deliver old data
• Different congestion cotrol mechanisms (TCP vs.
  TFRC)
• Low per-packet byte overhead

6
DCCP Which Layer?
SCTP
DCCP
Figure 2-11 TCP/IP Protocol Suite, Behrouz A.
Forouzan
7
Streaming Media
Source http//streaming.wisconsin.edu/Accessible_
Tutorials/Tutorial1/p1-3.htm

• What streaming media needs?
• Timeliness of data.
• What streaming media doesnt need?
• Retransmissions of lost packets.

8
Problems with TCP

• Introduce arbitrary delay
• Retransmission does not helpful and introduce
  more traffic to networks

9
Streaming media with TCP
Arbitrary Delay
D9-D12
A12
D12-D16
A16
D17-D19
D20-D22
Data is not useful
A16
D17
10
Streaming Media with UDP
No Congestion Control in UDP flows. Harmful to
internet health
11
Problems with UDP

• From Networks point of view
• Congestion
• Unfairness
• From the applications point of view
• Firewalls and NATs do not always pass UDP traffic

12
Solution

• UDP with
• congestion control
• Handshaking during connection setup and teardown

13
DCCP Overview

• Connection setup teardown Reliable handshaking
  
• Unreliable packet flow
• Bi-directional
• Unicast connections
• Reliable feedback from receiver to sender
• Choice of congestion control
• Minimal overhead

14
DCCP connection setup
Client
Server
DCCP A
DCCP B
Similar to TCP connection setup
DCCP Request
DCCP Response
DCCP ACK
15
DCCP Data transfer phase
Client
Server
DCCP DATA
DCCP ACK
DCCP DATA
DCCP DATA ACK
16
DCCP connection termination
Client
Server
DCCP A
DCCP B
DCCP CloseReq
DCCP Close
DCCP Reset
17
DCCP data transfer Example
B
A

• Each packet carries a seq
• Seq incremented per packet
• Pure Acks also increment seq

DCCP-DATA(seq 1)
DCCP-DATA(seq 2)
DCCP-ACK(seq 10, ACK 2)
DCCP-DATA ACK(seq 3, ACK 10)
DCCP-ACK(seq 11, ACK 3)
18
Congestion Control Mechanisms in DCCP

• TCP Like Congestion Control CCID 2
• TFRC Congestion Control CCID 3

19
CCID2 TCP Like Congestion Control

• AIMD behavior exhibited as in TCP/IP
• Slow start
• Timeouts
• Congestion event -gt Halve congestion window
• Abrupt rate changes

20
CCID 2 TCP-like Congestion Control

• Applications using this
• Respond quickly to changes in available
  bandwidth.
• Must tolerate abrupt changes

Online Interactive Games prefer this kind of
congestion control
21
CCID 3 TFRC Congestion Control

• TFRC, RFC 3448
• Equation-based congestion control
• Single congestion event does not halve rate
• Minimizes abrupt changes in sending rate
• Maintains longer-term fairness with TCP

Streaming Media doesnt need responsiveness but
prefer steadier less bursty traffic as provided
by TFRC
22
PART II

• Limited Slow-Start for TCP with Large Congestion
  Windows

23
Motivation

• For TCP connections that are able to use
  congestion windows of thousands MSS
• Increase cwnt by thousands of MSS per RTT
• Result in loss of thousands of packets in one RTT
• consequently
• Counter-productive for the TCP flow itself
• Also hard on the rest of the traffic sharing the
  congested link

24
Solution Limited Slow Start

• Limited Slow-Start, limiting the number of
  segments by which the congestion window is
  increased for one window of data during slow-
  start phase.
• How?
• Using max_ssthresh

25
Algorithm for limited slow start

• cwnd lt max_ssthresh
• Normal increase one MSS per ACK
• max_ssthresh lt cwnd lt ssthresh
• the congestion window is increased during
  slow-start by at most max_ssthresh/2 MSS per
  round- trip time.
• ssthresh lt cwnd,
• End of slow start phase enter congestion
  avoidance phase.

26
Example

• with Limited Slow-Start with
• max_ssthresh 100 MSS, 836 round-trip times to
  reach a congestion window of 83,000 packets
• the largest transient queue during slow-start
  would be 100 packets
• Normal slow start
• 16 round-trip times (assuming no packet drops).
• the transient queue during Slow-Start would reach
  more than 32,000 packets.

27
Question?

• How to choose appropriate value for max_ssthresh

28
PART III

• The NewReno Modification to TCP's Fast Recovery
  Algorithm

29
Motivation

• Typical Fast Retransmit algorithm(RENO) only
  retransmits a packet after
• a retransmit timeout has occurred,
• after three duplicate acknowledgements have
  arrived
• if the SACK option is available, the TCP sender
  has the information to make intelligent decisions
  about which packets to retransmit and which
  packets not to retransmit during Fast Recovery.

30
Motivation

• If SACK option is not available, after
  retransmitting a lost packet.
• there had been a single packet drop
• the acknowledgement for this packet will
  acknowledge all of the packets transmitted before
  Fast Retransmit was entered
• there were multiple packet drops
• the acknowledgement for the retransmitted packet
  will acknowledge some but not all of the packets
  transmitted before the Fast Retransmit.
• We call this packet a partial acknowledgment

31
The Fast Retransmit and Fast Recovery algorithms

• When the third duplicate ACK is received and the
  sender is not already in the Fast Recovery
  procedure, set ssthresh max (FlightSize / 2,
  2MSS) and record the highest sequence number
  transmitted in the variable "recover".
• Retransmit the lost segment and set cwnd to
  ssthresh plus 3MSS
• For each additional duplicate ACK received,
  increment cwnd by MSS and transmit a segment if
  allows.

32
The Fast Retransmit and Fast Recovery algorithms

• 4. When an ACK arrives that acknowledges new
  data, this ACK could be
• the acknowledgment elicited by the retransmission
  of the first lost data
• a later retransmission
• If ACK acks all data before retransmitting the
  first lost packet. Change cwnt to ssthresh and
  exit fast recovery procedure
• If it is a partial ACK packet,
• retransmit the first unacknowledged segment
• Adjust cwnt and

33
Key idea

• How to respond to a partial ACK.
• Retransmit one segment
• There are a few other options to handle a partial
  ACK.
• In TCP-Reno, the first partial ACK will bring the
  sender out of the fast recovery phase
• Results in timeouts when there are multiple
  losses
• In TCP New-Reno, partial ACK is taken as an
  indication of another lost packet (which is
  immediately retransmitted).
• Sender comes out of fast recovery only after all
  outstanding packets (at the time of first loss)
  are ACKed

34
PART IV

• A Conservative Selective Acknowledgment
  (SACK)-based Loss Recovery Algorithm for TCP

35
Motivation

• While the TCP SACK RFC2018 is being steadily
  deployed in the Internet , there is evidence that
  hosts are not using the SACK information when
  making retransmission and congestion control
  decisions.
• Conservative SACK one straightforward method for
  TCP implementations to use SACK information to
  increase performance

36
Main Idea
Scoreboard to keep tracking of information from
ack or sack.
RecoveryPoint
Enter loss recovery phase Dupthresh number of dup
ack received For each ack or sack, Reset ssthresh
cwnd flightsize/2 Retransmit the first
dropped segment
37
Part V

• High Speed TCP

38
What is High Speed TCP?

• Changes the way TCP behaves at high speed (ie
  large cwnd)
• Standard TCP has two modes
• Slow start (not very slow)
• Congestion Avoidance
• Focuses on Congestion Avoidance Region ie when
  TCP knows (thinks it knows) how well the network
  behaves
• BUT only when we are at high speeds, else do what
  normal Standard TCP does
• Readily deployable 1st step towards Equation
  Based Congestion Control

39
What does it do?

• Standard TCP uses two parameters
• Increase parameter, a
• Decrease parameter, b
• i.e. AIMD( a,b )
• Standard TCP uses
• a1
• b0.5
• High Speed TCP introduces
• a-gta(cwnd)
• b-gtb(cwnd)
• i.e. The value of a and b depends on the current
  congestion window size
• If we increase a more with larger cwnd we can get
  back up to our optimal cwnd size for the
  network path
• If we decrease b less we dont lose as much
  bandwidth due to a small congestion window

40
What exactly does it do?

• Based on the TCP response function
• Relates loss and throughput
• Uses the TCP response function to investigate
  certain parameters
• High_Window, High_Loss largest cwnd needed for x
  throughput and the required loss for that
  throughput
• Low_Window, Low_Loss smallest cwnd when we
  actually switch from Standard TCP and the
  required loss rate for that cwnd size
• High_B the smallest decrease in b when we are at
  a large cwnd
• Equations to transform this information into a
  table for a(cwnd) and b(cwnd)

41
PART IV

• Question, Discussion, Project

42
Issues

• Disk space issue
• Put all your data except your code to /tmp
  directory ( you have to make a directory for
  yourself there by mkdir /tmp/your_id
• No data between sources and destinations
• Start traffic at different time (i.e., every 0.01
  seconds)

43
Create Agents

• Create TCP/SACK in ns2
• set tcp new Agent/TCP/Sack1
• set sink new Agent/TCPSink/Sack1
• Create TFRC agent in ns2
• set tcp new Agent/TFRC
• set sink new Agent/TFRCSink

44
Tips

• For loops (suppose you name ns nodes as
  node_s(0), node_s(1), ..)
• set MAX_NUMBER_CONN 10
• for set index_of_conn 0 index_of_conn lt
  MAX_NUMBER_AMID_CONN incr index_of_conn
• set tcp_(index_of_conn) new Agent/TCP/Sack1
• tcp_(index_of_conn) set packetSize_ 1000
• tcp_(index_of_conn) set increase_num_ a
• tcp_(index_of_conn) set decrease_num_ b
• ns attach-agent node_s( index_of_conn)
  tcp_(index_of_conn)
• set ftp _(index_of_conn) new Application/FTP
• ftp _(index_of_conn) attach-agent tcp
  _(index_of_conn)
• set TCPSink _(index_of_conn) new
  Agent/TCPSinnk/Sack1
• ns connect tcp_(index_of_conn) TCPSink
  _(index_of_conn)
• Passing arguments
• set a lindex argv 0
• set b lindex argv 1
• set queueType lindex argv 2
• set Bandwidth lindex argv 3
• ns your_tcl_file argv0 argv1 argv2 argv 3

45

• Questions?