Re: Odd sequence....

From: Andi Kleen (ak@suse.de)
Date: Sun Jun 04 2000 - 12:39:57 EST


On Sun, Jun 04, 2000 at 05:55:18PM +0200, Rogier Wolff wrote:
>
>
> Hi,
>
> I just caught a TCP dump that looks kind of funny to me:
>
> 1 16:58:19.895993 bwww.ssh > 14dyn184.61101: . 864:864(0) ack 816
> 2 16:58:19.909331 14dyn184.61101 > bwww.ssh: P 816:836(20) ack 784
> 3 16:58:19.915995 bwww.ssh > 14dyn184.61101: P 784:824(40) ack 816
> 4 16:58:19.929933 14dyn184.61101 > bwww.ssh: . 836:836(0) ack 864
> 5 16:58:20.090871 14dyn184.61101 > bwww.ssh: P 836:856(20) ack 864
> 6 16:58:20.136009 bwww.ssh > 14dyn184.61101: P 864:884(20) ack 836
> 7 16:58:20.149553 14dyn184.61101 > bwww.ssh: . 856:856(0) ack 884
> 8 16:58:20.155974 bwww.ssh > 14dyn184.61101: P 884:904(20) ack 836
> 9 16:58:20.169551 14dyn184.61101 > bwww.ssh: . 856:856(0) ack 904
> 10 16:58:20.315990 bwww.ssh > 14dyn184.61101: . 904:904(0) ack 856
> 11 16:58:20.316002 bwww.ssh > 14dyn184.61101: P 904:924(20) ack 856
> 12 16:58:20.329548 14dyn184.61101 > bwww.ssh: . 856:856(0) ack 924
> 13 16:58:40.477658 14dyn184.61101 > bwww.ssh: P 856:876(20) ack 924
> 14 16:58:40.744978 bwww.ssh > 14dyn184.61101: . 1696:1696(0) ack 876

Some packet got lost.

> 15 16:59:19.536500 14dyn184.61101 > bwww.ssh: P 876:896(20) ack 924

14dyn acks immediately, but piggy backs some data in the ack.

> 16 16:59:19.803107 bwww.ssh > 14dyn184.61101: . 1696:1696(0) ack 896

bwww acks it. A bit later a retransmit would occur (fast retransmit
only really works for bulk data where the full window triggers enough
out of order acks)

>
> (all packets had win 32120 (DF) [tos 0x10], which is not show above).
>
>
> If I'm not mistaken, packet 4, is the one that messes things up:
> it acknowledges 40 bytes it hasn't recieved.

But it has. See packet 1.
>
> After 20 seconds of patience (*), at 16:58:40 I type another "return"
> to try and get things moving again. At that moment bwww, sends a
> packet indicating that we've missed 864-1696. Rosie then doesn't react
> and acks whatever it HAS recieved, to try to tempt bwww to perform a
> fast(!)-retransmit.

I see nothing wrong.

-Andi
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.rutgers.edu



This archive was generated by hypermail 2b29 : Wed Jun 07 2000 - 21:00:33 EST