---

TCP Tahoe: Implementation of the first Congestion Control Algorithm in TCP

---

• An overview of of techniques used in TCP Congestion control
  • We have seen a high level discription of the TCP Congestion control algorithm consisting of 2 different phase:

    In the slow start phase, TCP's transmission rate doubles after each epoch - thus increases exponentially in time. In the congestion avoidance phase, TCP's transmission rate increases linearly after each epoch - this increases linearly in time.

    ---

  • In this webpage, we will study how the increase in transmission speed is implemented

    ---

    ---

  • Techniques used to implement TCP congestion control algorithm (has very sexy sounding names):

    Slow Start Fast Retransmit Fast Recovery

    We will look at each mechanism separately and indicated when the mechanism is appropriate.

  ---

  ---

  The SLOW START Phase

  ---

• The Slow Start Mechanism
  • During the slow start phase, TCP uses the slow start mechanism for congestion control.

  • Information needed to implement the slow start mechanism:

    SSThresHold = The window size that TCP believes to be safe to use SSThresHold = AWS when TCP begins for the first time SSThresHold = CWND/2 when TCP detects a packet loss           CWND = The (current) Congestion window size In the slow start phase, the CWND will increase as follows: CWND(t+1) = 2 × CWND(t) when no congestion (packet drop) has been detected in an epoch Graphically:

    ---

  • Implementing the slow start mechanism:

    Initilization: (when TCP first starts transmitting) SSThresHold = AWS           Slow Start procedure: Set CWND = MSS (i.e., one packet) Whenever TCP receives an new ACK packet, then increases CWND by MSS Note: A new ACK message is an ACK that has a higher ACK number than the receive sequence number If TCP receives a duplicate ACK (i.e., an ACK message that has a ACK number that is less or equal to the receive sequence number) the CWND variable is unchanged !

    ---

    ---

  • Example of TCP operation in the slow start phase:

    Notes:

    Initially (at time 0), CWND = 1 At time RTT (round trip time), CWND = 2 At time 2 RTT, CWND = 4 At time 3 RTT (not in figure), CWND = 8 And so on... When you plot CWND over time, CWND will double after each RTT time (Therefore, the epoch that is RTT seconds !!!) When plotted out, the CWND increases exponentially with time

  ---

  ---

• When to start using the Slow Start procedure ?
  • The slow start procedure is used in the following 2 stituations:

    When a TCP connection is first establish (first start transmitting). In this case, SSThresHold is set to (equal to) AWS         When TCP has detected a packet loss In this case, SSThresHold will be set to CWND/2 (this is done by the congestion avoidance algorithm prior to entering in the slow start process)

  ---

  ---

• When to stop the Slow Start procedure ?
  • The slow start epoch can ended by 2 events:

    Ended normally when CWND becomes > SSThresh In this case, TCP will enter the congestion control phase: TCP is now in "uncharted" territory and will increase its congestion window slower when no congested is detected Ended abnormally if TCP detects a packet loss during the slow start phase In this case, TCP will: TCP first sets SSThresh = CNWD/2 (CWND is the transmit window size that TCP was using at the time that the packet loss was detected) Then TCP re-start the slow start procedure I.e., set CWND = 1 and increase CWND by 1 for every new ACK

  ---

  ---

• Retionale for the slow start procedure
  • Question:

    Why would TCP use a complicated "slow start" procedure to increase CWND from ONE all the way to SSThrehHold Why not just set CWND = SSThresh and be done with it ???

  • Answer:

    TCP does this to obtain an accurate estimate for the Round Trip Time (RTT) TCP uses timeouts to tell if it needs to retransmit packets The timeout value used depends on the RTT of packets By sending packets at a exponentially increasing rate instead of in a single burst, TCP can obtain a more accurate meaure for the RTT of packets

  ---

  ---

• How can you call the EXPONENTIAL increase of transmission rate in "Slow Start" SLOW ???
  • The name "slow start" is probably one of the worst misnomer in networking..

    How on earth can you call an exponential increase in window size in each epoch SLOW ???

  • It's slow compared to what TCP was doing before they changed it...

    ---

  • Prior to Jacobson's work, TCP operates as follows:

  • Notes:

    A new TCP connection first negotiate a advertised window size (AWS) The source immediately transmits an amount of data that is equal to the advertised window size (e.g., when a large file is transfered).

  • Now, compared to sending AWS bytes of data, the new way of start transmitting ONE packet first is indeeds slower...

  ---

  ---

  ---

  ---

  The Congestion Avoidance Phase

  ---

• TCP's Congestion Avoidance mode
  • TCP enters the congestion avoidance phase when the slow start phase terminates normally

    (i.e., CWND > SSThresHold)

    ---

  • Goal of the congestion avoidance phase:

    Goal:: TCP increases CWND by one packet (= MSS bytes) when no packet drops have been detected for RTT sec (one epoch)              Graphically: Problem: It difficult to constantly monitor packet drops for a given period of time It is easier by far to increase CWND each time you receive a new ACK message Notice that if TCP is transmitting maximum size packets, and the congestion window is CWND, then there are approximately CWND/MSS packets sent using the transmit window

    ---

  • Practical implementation of the linear increase method:

    We want to increase CWND by approximately MSS in RTT second Fact: The are approximately CWND/MSS packets in transition towards the receiver If no packets are dropped, the sender will receive approximately CWND/MSS new acknowledgements Suppose we increase CWND by x each time we receive a new ACK Then after RTT sec (or CWND/MSS acks), the new value of CWND will be: CWND + (CWND/MSS × x)             If we want CWND to increase by MSS in RTT sec (or CWND/MSS acks), then we must use: CWND + (CWND/MSS * x) = CWND + MSS <==> (CWND/MSS * x) = MSS MSS * MSS <==> x = --------------- CWND Conclusion: If we want to increase CWND by MSS in RTT time, we can approximately achives this by increasing CWND by MSS2/CWND each time TCP receives a new acknowledgement

    ---

    ---

    Example of TCP operation in the congestion avoidance phase: using CWND = 4 × MSS

    Notes:

    TCP has sent out 4 packets (each containing MSS bytes) to the receiver. If there is no congestion (no packet drops), the sender will receive 4 new ACK packets in approximately RTT time When the first ACK packet is received, TCP updates CWND as follows: CWND = CWND + MSS * MSS/CWND // CWND = 4 MSS = 4 MSS + MSS * MSS/(4 MSS) = 4 MSS + MSS * 1/4 = 4.25 MSS When the second ACK packet is received, TCP updates CWND as follows: CWND = CWND + MSS * MSS/CWND // CWND = 4.25 MSS = 4.25 MSS + MSS * MSS/(4.25 MSS) = 4.25 MSS + MSS * 1/4.25 = 4.485 MSS When the third ACK packet is received, TCP updates CWND as follows: CWND = CWND + MSS * MSS/CWND // CWND = 4.485 MSS = 4.485 MSS + MSS * MSS/(4.485 MSS) = 4.485 MSS + MSS * 1/4.485 = 4.708 MSS Finally, when the fourth (and final) ACK packet is received, TCP updates CWND as follows: CWND = CWND + MSS * MSS/CWND // CWND = 4.708 MSS = 4.708 MSS + MSS * MSS/(4.708 MSS) = 4.708 MSS + MSS * 1/4.708 = 4.92 MSS Conclusion: The value of CWND is increased by 0.92 × MSS (We want to increase CWND by exactly MSS - this is pretty good !!!)

    ---

    ---

  • NOTE: the actual implementation of TCP (see Stevens - Volume 2) increases CWND during congestion avoidance slightly faster than above using to the following formula:

    CWND = CWND + MSS * MSS/CWND + MSS/8

  ---

  ---

• When does TCP start using the Congestion Avoidance algorithm ?
  • TCP starts to use Congestion Avoidance:

    When TCP has successfully completed the Slow Start phase           I.e.: when CWND ≥ SSThresh

  ---

  ---

• When does TCP stop using the Congestion Avoidance algorithm ?
  • TCP stop to use Congestion Avoidance:

    When TCP detects a packet loss event            (TCP times out waiting for an ACK) A packet drop means congestion

    ---

  • When TCP times out:

    Set: SSThresh = CWND/2            Enter the slow start phase

    Example:

    Notes:

    The sender has CWND = 4 × MSS Suppose packet 14 is lost The ACK message for packet 14 will not be received and the sender will time out Result of the time out: TCP sets SSThresh = CWND/2 = 2 × MSS. (This is the new "safe" operation level)... Then TCP enters the slow start procedure: Set CWND = 1 x MSS (i.e., 1 packet worth of data) and increases CWND at an expontial rate towards SSThresh (the "safe" level")

  ---

  ---

  ---

  ---

• Detecting packet drop in standard Go-back-N
  • The standard Go-Back-N method uses timeout to trigger retransmission

    Example:

    Note:

    All packets upto packet 13 have been received and acknowledged When packet 14 is lossed, and packets #15, #16, #17 and #18 arrive out of order at the receiver The receiver will send back one ACK 13 for each received packet (to indicate that the last consecutive packet received was #13)

    ---

  • Packet loss detection:

    The packet loss in standard Go-back-N is through time out Therefore: When the sender times out and retransmits the packet, the sender will begin a slow start phase

  ---

  ---

• Fast Retransmit: detecting packet drop using duplicate ACKs
  • Notice that when a packet is dropped, the packets that follow the dropped packet will trigger duplicate ACK messages from the receiver:

    Note:

    We see 4 duplicate ACKs in the above example

    ---

  • Caveat: not every duplicate ACK is the result of packet loss

    Example: packets that arrive out of order can result in duplicate ACK can arrive:

    Avoid misinterpretating duplicate ACK as packet loss:

    To eliminate false loss indications from duplicate ACK, TCP will conclude packet loss only after receive 3 duplicate ACKs (so TCP received a total of 4 identical ACK packets),

    ---

    ---

  • Improved packet loss detection mechanism:

    Note:

    After receiving 3 duplicate ACKs, TCP concludes that a packet has been lossed TCP retransmits the lossed packet immediately without waiting for timeout !!! This retransmission technique is called fast retransmit (It retransmits a loss packet faster than using time out) Furthermore, TCP will enter the Slow Start phase (because a packet loss had occured)

  ---

  ---

  ---

  ---

• TCP Tahoe
  • The TCP algorithm that uses:

    Slow start Congestion avoidance Fast Retransmit

    is known as TCP Tahoe

  ---

  ---

• TCP Tahoe Demo
  • TCP Tahoe (the original version by Jacobson) incorporates the Slow Start and Congestion Avoidance mechanisms.

  • We will look at the operation of TCP Tahoe in this sample network:

    ---

  • Example Program: (Simulates the above network in NS)                        
    • Here is a NS2 source file to simulate a TCP Tahoe source: click here
      • Right click and save the file in your directory.
      • Run program with:

              /home/cs455000/bin/ns    Tahoe.tcl
        

      • You should see the Network Animator window when it finish running... click PLAY to see the simulation in action

      ---

    • You don't need to run the simulation to see the animation...

      I have saved a copy of the animation file generated by the simulation.

      The NAM (Network Animation) output file is here: click here

      To see the animation, save the NAM file in your directory and use this command:

          /home/cs455000/bin/nam   Tahoe.nam
      

      ---

    • The Congestion Window CWND plot data output file is here: click here
      To see the plot of the CWND of TCP, save the Congestion Window CWND plot file in your directory and run gnuplot

      In gnuplot, issue the command:

           plot "WinFile" using 1:2 title "Flow 1" with lines 1
      

      You should see this plot:

      You can see the operation of TCP Tahoe clearly from the above figure:

      1. At approximately time 0, TCP Tahoe starts and it is in the slow start mode: the congestion window size increases exponentially

      2. At approximately time 5, packet loss is detected.

         TCP marks SSThresh = 25 (approximately) and begins another slow start

      3. When it reaches CWND = 25 (approximately), the CWND increases linearly - here TCP Tahoe enters the congestion avoidance mode

      4. At approximately time 19, TCP Tahoe detects packet loss and begins a slow start.

         SSThresHold is approximately 22.

      5. TCP begins another slow start and so on...

  ---

  ---

  ---

  ---