What is the difference in reliability of sending a pure ACK packet (without any data), as opposed to a data packet in TCP?

Question

I was going through the book TCP/IP Illustrated and in the zero windows section, I noticed the use of a TCP persist timer - which is used to avoid losing a window update message from a receiver after a zero window scenario has just ended. Apparently, losing the window update message is of higher probability since it's a 'pure ACK' packet, as opposed to a packet with actual data.

Can someone please tell me what the difference in delivery of a pure ack packet and a data packet is? Thanks!

Answer 1

Can someone please tell me what the difference in delivery of a pure ack packet and a data packet is?

The difference is just that the one contains data (application payload) while the other does not. If a packet contains data it needs to be acknowledged by the recipient. If it does not contain data it will not be acknowledged.

This means that a loss of the packet will be recognized by the sender only if the ACK is outstanding. This is not true for a pure window update.

Or in maybe more clear scenarios:

• A sends dataA to B. B sends ACK(dataA)+dataB to A.
  A loss of the latter packet will be detected by B since an ACK(dataB) from A is expected.
• A sends dataA to B. B sends just an ACK without data back: ACK(dataA).
  A loss of the latter packet will not be directly detected by B. But A will miss the ACK(dataA) and thus retransmit dataA, resulting in B sending ACK(dataA) again.
• B has already send ACK(dataA). It now send ACK(dataA) with just a window update and no data.
  A will not detect if this packet got lost because it got ACK(dataA) before. B will not detect if this packet got lost either because it does not expect A to reply to it.