What is the unit and the behaviour of the congestion window?

Question

I've learned that the congestion window is the maximum amount of packets that can be sent in one transmission round. Now TCP seems to acknowledge the packets in bytes, so wouldn't it make more sense to define the congestion window in bytes. So how can the acknowledgments and the congestion window work together?

I hoped to find more information in the according RFC 5681 but after reading I still don't have an answer.

The RFC seems to measure the congestion window based on the SMSS (Senders Maximum Segment Size) which would be bytes and not a package amount.

Besides that the RFC mentions that the congestion windows increases by max SMSS which wouldn't really allow a exponential growth.

So how does congestion control really work in slow start?

Edit:

1. Following part of the RFC confuses me:

During slow start, a TCP increments cwnd by at most SMSS bytes for each ACK received that cumulatively acknowledges new data. Slow start ends when cwnd exceeds ssthresh (or, optionally, when it reaches it, as noted above) or when congestion is observed. While traditionally TCP implementations have increased cwnd by precisely SMSS bytes upon receipt of an ACK covering new data, we RECOMMEND that TCP implementations increase cwnd, per:

  cwnd += min (N, SMSS)                      

It says that the congestion window increases by N which has been referred to as the number of previously unacknowledged bytes acknowledged in the incoming ACK. or SMSS depending on which one is smaller. This would imply the congestion window is in bytes? And how can it be exponential then?

Answer 1

The receive window is shared with the sender, and it is how much data can be sent before the sender must pause to wait for an acknowledgement.

The congestion window has nothing to do with acknowledgments (other than those informing it of a lost segment). The congestion window is only known by the sender, and it is how much a sender can send in a burst of sending (as long as the receive window is still open). The sender uses the congestion window to learn how much it can burst before there is a problem.

The sender slowly grows the congestion window until there is a problem, then it will drastically shrink the congestion window when it learns of a problem. It will then slowly grow the congestion window until there is again a problem, at which point it again drastically shrinks the congestion window.

The receiver has no knowledge of this, and it simply acknowledges segments it has received, and that can inform the sender of lost segments, causing the sender to realize there is a problem, at which point the sender will adjust the send window.

The congestion window and the receive window are separate. The receive window is maintained by the receiver and shared with the sender, but the congestion window is maintained by the sender, but it is not shared with the receiver.