[e2e] tcp retransmission packet size.

[Vikram Visweswaraiah]

On Mon, 26 Nov 2001, Tianbo Kuang wrote:
>Hi,
>
>I noticed in a captured trace that a TCP sender actually retransmitted a
>bigger packet than the orignal one (after the ACK was dropped in the
>network). My question is,
>(1) Is this allowed by specification?

Assuming your packets equate to TCP segments, you could see TCP retransmitting
more segments than necessary. This is normal as an acknowledgement is cumulative
and the sender normally does not know that segments sent after the one that got
lost were successfully received. If the receiver manages to send a cumulative
ACK for a higher sequence number before the sender times out, that would cover
the previously lost ACK (because in your case, an ACK was lost and not the
segment itself). SACK relieves few such redundant retransmits.

>(2) If so, that means that a TCP receiver will always have to use the
>duplicated packet as the useful data as it contains more information. Is
>that true in the implementation?

The receiver will discard duplicates. There really isn't "more information" - in
fact, there's more likely unnecessary data (as it was already received)

-- 
:=Vikram