Date: Tue, 11 Mar 1997 16:27:30 +0900 (JST) From: Michael Hancock <michaelh@cet.co.jp> To: Tom Samplonius <tom@sdf.com> Cc: Brian Tao <taob@vex.net>, "matthew c. mead" <mmead@goof.com>, isp@freebsd.org, hackers@freebsd.org Subject: Maxcontig for news server (was Re: freebsd as a news server?) Message-ID: <Pine.SV4.3.95.970311160947.7676A-100000@parkplace.cet.co.jp> In-Reply-To: <Pine.NEB.3.94.970310194240.27200A-100000@misery.sdf.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 10 Mar 1997, Tom Samplonius wrote:
> > > On Tue, 11 Mar 1997, Michael Hancock wrote:
> > >
> > > ...
> > > > I've been using -i 3072 -b 4096 -f 1024 -a 8 for non-binary newsgroups and
> > > > the default for binary newsgroups.
> > > >
> > > > This way news articles on average will fit in a block.
> > >
> > > You always want maxcontig to beigger than 8. It has nothing to do with
> > > file allocation, only have many sectors are transfered at once to the
> > > driver.
> >
> > And the magic number is ____.
>
> As big as possible. All the disk drivers should be able to handle at
> least 128. I don't really know how big it can be. 256? 1024?
> A too small maxcontig will hamper scatter/gather. I noticed that by
> looking at the sps and tps values from iostat that it was never
> transfering more than 16 sectors per transfer. I bumped up maxcontig, and
> noticed a nice performance increase (at least for my application).
Was your application INND?
> The default maxcontig setting for 2.1.x is just way too small.
The default is 8 assuming a bsize of 8K, the man pages say it's 1.
if (maxcontig == 0)
maxcontig = MAX(1, MAXPHYS / bsize - 1);
MAXPHYS is 64K.
Regards,
Mike
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.SV4.3.95.970311160947.7676A-100000>