Slow Start Congestion Avoidance Global Sychronization
I have come to the conclusion that, it is good to be able to explain the above terms -- if an engineer wants to be able to sell QoS (to a manager or customer). Specifically, how Global Synchronization happens. And as a result, be able to explain why QoS might be a better solution, over just buying a bigger pipe.
It is all a bit confusing with the constant mathematical equations etc. As of yet, I still do not think that I would be able to regurgitate any of the explanations for these topics in simple English to a manager.
But I did find apresentationthat has this graph: Illustrations of Congestion Control If this is accurate, it does a lot to clear the fog on the subject.
The basic steps below are what I am coming up with as a simple explanation.
1. Slow Start is exponential and keeps doubling the congestion window (CWND) until it hits the first SSTHRES (slow start threshold).
2. THEN, Congestion Avoidance takes over, and it becomes additive (versus slow-start's exponential).
3 a. If there are ever three DUPLICATE ACK'S -- during either process -- the window size drops to half of the current window size (CWND)
b. But, if there is a single TIMEOUT (for any acks) -- the process starts from the very beginning (slow-start process) - minimum of ONE (1) MSS (max segment size). The SSTHRES will be set at 1/2, and Congestion Avoidance will begin at this point.
Do I have it finally? Thanks in advance for any corrections.
--------------------------------------------------------------------
EDIT: Graph from the presentation link above Congestion Control in TCP
