Performance differences between Go-Back-N and Selective Repeat ARQ protocols?

Question Detail: 

This is what I know:

• GBN weakness is the fact that when the window size is too large, the number of packets in the pipeline grows and one packet error causes the retransmition of many packets unnecessarily.
• Selective Repeat solves this by acknowleding just the suspicious packets, which sightly makes performance better, but if a wrong window size is chosen, then the reciever doesn't know if a packet is being retransmitted or another packet is being sent by first time.

So, back to the question, what other important performance differences exists between these protocols?

Asked By : Alejandro Sazo

Best Answer from StackOverflow

Question Source : http://cs.stackexchange.com/questions/11715

Answered By : inphi

If GBN is used in tandem with AIMD for congestion avoidance, the network link will experience a large amount of latency when a packet is lost. This is because the AIMD CA mechanism will typically wait for a timeout before declaring the packet that hasn't been received as being dropped. Also, the timeout wait-time will occur every time because of the nature of GBN to send cumulative acknowledgments. Worse still, the pipeline will decrease drastically every time this occurs, causing the network to incorrectly gauge congestion.

Once you contrast the above scenario with the SR instead of GBN, it becomes apparent that the window size is no longer really an issue when compared to the latency due to the lost packets in the pipeline. However, SR handles the scenario well since it acknowledges every packet. Thus, the pipeline grows at a favorable rate (as dictated by the AIMD CA) without having unnecessary decreases in the length of the pipeline.

More differences come to light when a congestion avoidance / prevention mechanism is presented.