From: John Heffner (jheffner@psc.edu)
Date: 03/26/04
Date: Fri, 26 Mar 2004 15:10:14 -0500 (EST) From: John Heffner <jheffner@psc.edu> Subject: Re: tcptrace Congestion Control in TCP Message-ID: <Pine.NEB.4.33.0403261457540.2239-100000@dexter.psc.edu>
On Fri, 26 Mar 2004, Mark Allman wrote:
>
> > I think what you're missing is that when the congestion occurs at the
> > local interface, the packet drop is explicitly communicated back up to
> > TCP, so it will stop then retransmit without you seeing any loss on
> > the wire.
>
> I disagree. If the loss is local to the attached link then the link may
> retransmit the packet without any help from the transport (e.g., an
> ethernet collision). But, I don't think it is generally the case that
> layers 2 and 4 communicate in the way sketched above. (Or, didn't I
> understand what you're trying to say.)
The TCP transmission routine hands off a packet buffer to IP, which then
tries to put it in the transmission queue for the outgoing device. If
this queuing operation fails, it is indicated in a return value and
propagated back through the calling stack to TCP. If the packet was not
successfully queued, TCP will stop transmitting and not advance snd_nxt,
causing it to "retransmit" the packet later.
-John
----------------------------------------------------------------------------
To unsubscribe, send a message with body containing "unsubscribe tcptrace" to
majordomo@tcptrace.org.
This archive was generated by hypermail 2.1.7 : 03/27/04 EST