From mboxrd@z Thu Jan  1 00:00:00 1970
From: Andy Furniss <andy.furniss@dsl.pipex.com>
Date: Sat, 09 Apr 2005 19:59:41 +0000
Subject: Re: [LARTC] About sockets in "CLOSING" state
Message-Id: <4258342D.8010908@dsl.pipex.com>
List-Id: <lartc.vger.kernel.org>
References: <20050408144937.GA3694@dsl-202-173-141-208.sa.westnet.com.au>
In-Reply-To: <20050408144937.GA3694@dsl-202-173-141-208.sa.westnet.com.au>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: lartc@vger.kernel.org

Denis Potapov wrote:
> Hi,
> 
> I have met the problem: when I use the shaping discipline 
> 
> tc qdisc add dev ppp0 parent 1:2 tbf latency 50ms burst 1450 rate 50kbit
> 
> one of my application (namely, "aMule") starts leaving sockets in
> "CLOSING" state. These sockets accumulate and do not disappear.
> Eventually I have so many of these dead sockets that the kernel warns
> "Out of socket memory" in syslog (That happens when I have more than
> 5000 dead sockets). 
> 
> When the discipline above canceled the problem disappears. Now the
> questions. 
> 
> 1. Is it problem of my application (=aMule) or QoS.
> 
> 2. Is there any way to eliminate these "dead" sockets out of kernel
>    memory, like "kill"?
> 
> Please, share if you have any ideas.

Using burst 1450 on tbf will drop all packets > 1449 bytes if your MTU 
is > 1449 then this could be the problem.

Andy.
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc