From: "Eugene B. Berdnikov" <berd@elf.ihep.su>
To: kuznet@ms2.inr.ac.ru
Cc: linux-kernel@vger.kernel.org
Subject: Re: Bug report: tcp staled when send-q != 0, timers == 0.
Date: Sat, 21 Apr 2001 19:45:03 +0400 [thread overview]
Message-ID: <20010421194503.H23490@elf.ihep.su> (raw)
In-Reply-To: <20010413125437.D25085@elf.ihep.su> <200104181928.XAA04912@ms2.inr.ac.ru>
In-Reply-To: <200104181928.XAA04912@ms2.inr.ac.ru>; from kuznet@ms2.inr.ac.ru on Wed, Apr 18, 2001 at 11:28:43PM +0400
Hello.
On Wed, Apr 18, 2001 at 11:28:43PM +0400, kuznet@ms2.inr.ac.ru wrote:
> > However, this is _not_ a staled state. When I resume ssh on 194.190.166.31,
> > buffer gets empty and connection behaves as normal. I made this experiment
> > waiting for keepalive packets from both sides, as well as resuming ssh
> > before keepalives. In both cases connection did not become stale.
>
> Yes, I have said that it is practically impossible to reproduce this.
> My guess was that it is due to inaccurate counting of sacks when path
> mtu discovery happens or when segments are fragmented due to SWS avoidance
> override.
Im my case P-MTU discovery and fragmentation should be ruled out, but
sacks are really frequent: my hosts are connected via poor leased line.
> Actually, the most dubious place is your statement that this connection
> was not idle for 2 hours. It is _necessary_ condition
> for my scenario to work...
I only wrote that it was active when got stuck. It may be idle before -
I do not remember, but have a habit to keep connections for weeks. :)
As my experiments show, any connection, entering keepalive once,
have lose its ability to send zero probes - forever.
> > [I hope we will continue this discussion later.]
>
> I am ready.
OK. Let us return to the "mss/mtu bug". The most mystifying thing for
me is the dependance of the MTU threshold on the kernel version, etc.
I also wrote that it depends on keepalive flag. It seems, it was a mistake.
My additional experiments show that there is no distinct threshold of MTU:
trying the same value many times, I observed loss of acks in some cases,
and in some did not. So, the MTU boundary is not strict. Well, let it be.
But the question is what the minimum "reliable" MTU. There are lots of
situations when data comes rapidly in small packets (say, monitoring logs).
Is there a danger to lose such connections on a heavily loaded host?
--
Eugene Berdnikov
next prev parent reply other threads:[~2001-04-21 15:45 UTC|newest]
Thread overview: 16+ messages / expand[flat|nested] mbox.gz Atom feed top
2001-04-09 14:43 Bug report: tcp staled when send-q != 0, timers == 0 Eugene B. Berdnikov
2001-04-10 17:38 ` kuznet
2001-04-10 21:19 ` Eugene B. Berdnikov
2001-04-11 10:16 ` Eugene B. Berdnikov
2001-04-11 16:56 ` kuznet
2001-04-11 18:35 ` Eugene B. Berdnikov
2001-04-11 19:04 ` kuznet
2001-04-11 19:28 ` Eugene B. Berdnikov
2001-04-11 19:37 ` kuznet
2001-04-11 16:35 ` kuznet
2001-04-11 18:50 ` Eugene B. Berdnikov
2001-04-11 19:09 ` kuznet
2001-04-11 19:18 ` Eugene B. Berdnikov
2001-04-13 8:54 ` Eugene B. Berdnikov
[not found] ` <200104181928.XAA04912@ms2.inr.ac.ru>
2001-04-21 15:45 ` Eugene B. Berdnikov [this message]
2001-04-21 17:02 ` kuznet
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20010421194503.H23490@elf.ihep.su \
--to=berd@elf.ihep.su \
--cc=kuznet@ms2.inr.ac.ru \
--cc=linux-kernel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.