Scribd
Upload
30 views

Source Based Congestion Control Mechanisms: Chapter 5 - Unit I

The document discusses various source-based and delay-based congestion control mechanisms for TCP. It describes traditional TCP variants like Tahoe, Reno, and NewReno and their use of congestion windows and self-clocking. It then summarizes modifications like Scalable TCP, HighSpeed TCP, BIC, and CUBIC that address TCP's performance bottlenecks at high bandwidths. It also covers delay-based approaches like TCP Vegas and FAST TCP that use queuing delay as a congestion signal to smoothly calibrate window sizes.

Uploaded by

BhavithraRamesh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
30 views

Source Based Congestion Control Mechanisms: Chapter 5 - Unit I

The document discusses various source-based and delay-based congestion control mechanisms for TCP. It describes traditional TCP variants like Tahoe, Reno, and NewReno and their use of congestion windows and self-clocking. It then summarizes modifications like Scalable TCP, HighSpeed TCP, BIC, and CUBIC that address TCP's performance bottlenecks at high bandwidths. It also covers delay-based approaches like TCP Vegas and FAST TCP that use queuing delay as a congestion signal to smoothly calibrate window sizes.

Uploaded by

BhavithraRamesh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 16

Source Based

Congestion Control
Mechanisms
Chapter 5 – Unit I
Contents
Traditional TCP
TCP Modifications for Networks
◦ Scalable TCP
◦ HighSpeed TCP
◦ BIC
◦ CUBIC
Delay Based Congestion Control
◦ TCP Vegas
◦ FAST TCP
Traditional TCP
Traditional Versions:
◦ TCP Tahoe, TCP Reno, TCP NewReno
Uses Self-Clocking based congestion control
◦ Uses Congestion Window (CW)
◦ Idea: linearly increasing its rate when positive ACK scenario;
and exponentially reducing its rate when congestion is
detected
 Two phases in TCP Tahoe
◦ Slow Start and Congestion Avoidance
 TCP NewReno
◦ Fast Retransmit and Fast Recovery
Ie.., During CA phase for each ack, window is
increased as:
TCP Modifications for Networks
TCP Reno outperforms and satisfies in sever congestion
states.
But, that as the bandwidth-delay product grows, TCP Reno
eventually becomes a performance bottleneck
3 reasons for the drawback:
◦ Linear increase by one packet per round trip time is too slow, and
multiplicative decrease per loss event is too drastic.
◦ Maintaining large average congestion windows requires an extremely
small loss probability in equilibrium
◦ TCP Reno uses a binary congestion signal (packet loss), which
introduces strong oscillations in the congestion window.
a) Scalable TCP (STCP)
Same Slow start phase
In Congestion Avoidance phase:
◦ CW is increases as:

◦ where a is a constant satisfying 0 < a < 1.


When congestion is detected,
◦ CW is decreased as:

◦ where 0 < b < 1 is a design constant.


◦ The use of a = 0.01 and b = 0.125
b) High Speed TCP (HSTCP)
A TCP modification that applies only when the congestion
window is large.
Upon receiving an ACK while in congestion avoidance mode
◦ During positive ACKs, CW is altered as:

◦ During loss event, CW is altered as:

◦ a(w) and b(w) are appropriate functions of w.


◦ Ideal values: a(w) =1 and b(w) = 1/2
c) BIC
BIC consists of two parts
◦ The binary search increase and
◦ The additive increase.
The binary search increase:
◦ Congestion control is treated as a searching problem
◦ The system generates binary feedback, whenever at packet
loss the current window is larger than the network capacity
◦ Maintains 2 window thresholds:
 Wmin - window size just after the fast recovery
 Wmax - window size just before the fast recovery
The binary search increase …
◦ Search is between Wmin and Wmax
◦ The algorithm repeatedly computes the midpoint between
Wmax and Wmin,
◦ Current CW = midpoint
◦ Now checks for feedback, in the form of packet losses
 If packet loss; new Wmax = midpoint
 Else; new Wmin = midpoint

Additive increase:
◦ Leads to faster convergence and RTT fairness
◦ Maintains two increment thresholds Smin and Smax.
◦ If Wmin - CW >> 0 increase CW by Smax
◦ If CW - Wmax << 0 increase CW by Smin
d) CUBIC
 Enhanced version of BIC
 The window growth function of CUBIC is a cubic function

 where C is a scaling factor, t is the elapsed time from the last


window reduction, Wmax is the window size just before the last
window reduction, and

 The window grows very fast upon a window reduction,


 As it gets closer to Wmax, its rate of increase reduces.
 Around Wmax, the window increment becomes almost zero
Delay Based Congestion Control
TCP Vegas
TCP Vegas utilizes the packet queuing delay as a
congestion signal to determine the rate at which to
send packets
Queuing delay (t) = RTT(t) – RTTmin
◦ RTT(t) = Current RTT
◦ RTTmin = Minimum RTT until current transmission
Uses 2 thresholds α and β (Positive values)
The congestion window updating rule of TCP Vegas
is given by
Advantages of TCP Vegas:
◦ Congestion window size is calibrated more smoothly
◦ CW remains same when the difference between the expected
and the actual rate is between the two thresholds α and β
◦ It’s a proactive algorithm
Disadvantages:
◦ TCP Vegas is inter-operated with other TCP versions like
Reno
 Degraded performance as Vegas reduces its sending rate before
Reno

◦ Inaccurate measurement of the actual round trip propagation


delay
FAST TCP
It is the high-speed version of Vegas
Delay based Congestion control mechanism
FAST operation is comprised by four modules:
◦ Data Control - determines which packets to transmit
◦ Window Control - determines how many packets to transmit
◦ Burstiness Control – determines when to transmit these packets
◦ Estimation - drives the other parts by providing necessary
information
Estimation
◦ It computes two pieces of feedback information for each data
packet sent
◦ From every ACK that arrives, it takes RTT sample and update:
 Average queueing delay
 RTTmin.

◦ For any negative ACK,


 it generates a loss indication for this data packet to the other
modules.

◦ Delay is computed as:


Window Control:
◦ Computes how much packet to send

where γ ϵ (0, 1] and a is a function of the current


window size and the average queuing delay

You might also like