[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20120215025200.6e6f02a2.billfink@mindspring.com>
Date: Wed, 15 Feb 2012 02:52:00 -0500
From: Bill Fink <billfink@...dspring.com>
To: Eric Dumazet <eric.dumazet@...il.com>
Cc: Neal Cardwell <ncardwell@...gle.com>, alekcejk@...glemail.com,
netdev@...r.kernel.org
Subject: Re: limited network bandwidth with 3.2.x kernels
On Wed, 15 Feb 2012, Eric Dumazet wrote:
> Le mercredi 15 février 2012 à 01:03 -0500, Neal Cardwell a écrit :
> > There are some potentially interesting differences in sysctl values:
> >
> > > diff -u /tmp/sysctl-a-3.1.txt /tmp/sysctl-a-3.2.txt | grep mem_def
> > -net.core.wmem_default = 114688
> > -net.core.rmem_default = 114688
> > +net.core.wmem_default = 163840
> > +net.core.rmem_default = 163840
> >
> > Can you please try running this as root on the 3.2 kernel and then see
> > what sort of performance you get:
> >
> > sysctl net.core.rmem_default=114688
> >
> > This should make the 3.2 config closer to the 3.1 config. Should be
> > interesting to see if the behavior is more similar when that happens.
> >
> > neal
>
> These changes are normal, since skb truesize now include shared_info.
>
> By the way, these params are not used for tcp streams.
>
> This is why I asked netstat -s values, to check if we had drops because
> of truesize limits.
>
> Its not the case here since skb are not too fat (as some NICS can do,
> using a full PAGE to hold a single 1500 bytes frams)
Would it be useful to check if there was any difference in
TCP parameters between the 3.1 and 3.2 kernels by performing
the following on each:
grep . /sys/module/tcp_*/parameters/*
-Bill
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists