Subject: Re: CVS commit: src/sys/arch/i386/conf
To: YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp>
From: Manuel Bouyer <bouyer@antioche.eu.org>
List: source-changes
Date: 10/03/2005 16:37:15
On Mon, Oct 03, 2005 at 06:45:10PM +0900, YAMAMOTO Takashi wrote:
> > > > Log Message:
> > > > Add
> > > > options         MCLSHIFT=12
> > > > in order to avoid a copy when a mbuf cluster has to be sent to a domU
> > > > (which is the common case when the packet comes from a physical network
> > > > interface).
> > > 
> > > i doubt if it's a reasonable value for this "GENERIC" kernel.
> > > it makes mclsize dominant about socket buffer calculation, doesn't it?
> > 
> > Can you give more details ? I don't understand what you mean here.
> 
> sbreserve() sets sb_mbmax to (sb_hiwat * 2).
> given mtu=1500 and mclsize=4096, as mclsize > (mtu * 2),
> now sbspace() likely takes care of sb_mbmax rather than sb_hiwat
> even for bulk transfer.

Sorry but I don't know what sb_mbmax and sb_hiwat are used for. What would
be the consequence of this ?

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--