Congestion control behaviour of TCP Reno

Question

Which behaviour of TCP Reno after 3 DUP ACK is correct:

• The one described in the Italian Wikipedia (https://it.wikipedia.org/wiki/Controllo_della_congestione_in_TCP#TCP_Reno). It says the new cwnd is: cwnd_new = cwnd_old/2 + 3*MSS
• The one described in the English Wikipedia (https://en.wikipedia.org/wiki/TCP_congestion_control#TCP_Tahoe_and_Reno). It says the new cwnd is: cwnd_new = cwnd_old/2

I guess, it's the Italian Wikipedia that is right, since also the FSM in the book "Computer Networking: A Top-Down Approach, Kurose & Ross" shows it like that (http://www.linyibin.cn/images/Technology-ComputerNetworking-Internet-TCP-CongestionControl.png).

Is there a (good) reason why many sources neglect the term "+3*MSS"?

Answer 1

When TCP Reno sender gets 3 duplicate ACKs, it enters a so called fast retransmit mode. The idea of this mode is that a) since ACKs are delivered it indicates that packets are leaving the network b) if a packet leaves a network, the sender can inject new packet in the network. Duplicate ACK indicates that a packet has left the network. Fast retransmit increases congestion window for each duplicate ACK received. This is called "inflating". The 3 MSS (3 segments) are added upon entering, because 3 duplicate ACKs were already received. Now, this increase persists only in fast retransmit. When the mode is exited, congestion window is set to sstresh (which is cwnd_old/2, i.e., congestion window before the first duplicate ACK is received/2).

So, on 3 dupacks the sender does:

• set sstresh = cwnd/2
• set cwnd = sstresh
• enter fast recovery (upon entering increase cwnd by 3 segments to account for already received dupacks cwnd=cwnd + 3 MSS)
• in fast recovery increase cwnd by 1 segment on each duplicate ack
• when first non-duplicate ack is received exit fast recovery; set cwnd = sstresh (without 3 extra segments!)

So, technically these two sources described the state at 2 different points in time.

(fast retransmit specification here: https://datatracker.ietf.org/doc/html/rfc5681#section-3.2)

Answer 2

From Computer Networking A Top-Down Approach 6th Edition by Kurose & Ross linked to in English Wikipedia:

The congestion window then climbs linearly until a triple duplicate- ACK event occurs, just after transmission round 8. Note that the congestion window is 12 • MSS when this loss event occurs. The value of ssthresh is then set to 0.5 • cwnd = 6 • MSS. Under TCP Reno, the congestion window is set to cwnd = 6 • MSS and then grows linearly. Under TCP Tahoe, the congestion window is set to 1 MSS and grows exponentially until it reaches the value of ssthresh, at which point it grows linearly.

Potentially, cwnd_new = cwnd_old / 2 + 3 * MSS is counter-productive for a very small window when the value actually grows.