[Top] [All Lists]

Re: benchmarks

To: Steve Lord <lord@xxxxxxx>
Subject: Re: benchmarks
From: Ragnar Kjørstad <xfs@xxxxxxxxxxxxxxxxxxx>
Date: Mon, 16 Jul 2001 23:56:03 +0200
Cc: Mike Gigante <mg@xxxxxxxxxxxxxxxxxxxxxxx>, Simon Matter <simon.matter@xxxxxxxxxxxxxxxx>, "P.Dixon" <P.Dixon@xxxxxxxxx>, linux-xfs@xxxxxxxxxxx
In-reply-to: <200107162111.f6GLBPQ24732@jen.americas.sgi.com>; from Steve Lord on Mon, Jul 16, 2001 at 04:11:25PM -0500
References: <xfs@ragnark.vestdata.no> <200107162111.f6GLBPQ24732@jen.americas.sgi.com>
Sender: owner-linux-xfs@xxxxxxxxxxx
On Mon, Jul 16, 2001 at 04:11:25PM -0500, Steve Lord wrote:
> > > The mongo benchmarks use relatively small files. XFS scales very well to
> > > large files/directories, but that benchmark sticks to small scale stuff.
> > > It *seems* to show ReiserFS in the best possible light and the XFS results
> > > don't seem to reflect anything like real world experience (except that it
> > > correctly highlights the relatively slow delete performance of XFS!)
> > 
> > I partly disagree. mongo tests filesystem performace - benchmarks that
> > test writing really big files is mostly testing your hardware. The
> > filesystem still matters, but not so much as on small files. I think
> > the difference between ext2/xfs/reiserfs is less than 10%. (ext3 with
> > datalogging may be a different story though).
> mongo tests how the filesystem behaves under the endcase of continuous
> metadata load from an application which is doing nothing with the data
> except reading it and writing it. It is a rare application which behaves
> in this way, but it is a very common benchmark, the filesystem is there
> to provide service to applications, not to consume all the resources
> of the machine.

Yes, and reiserfs is more cpu-intensive than other filesystems, so
benchmarks (wich are generally not very cpu-intensive) may indicate
better performance on reiserfs than what is the case for real
applications (wich may or may not be cpu-intesive).

> Note that I am not arguing for one filesystem or another, just commenting
> on the nature of filesystem benchmarks, reiserfs is good for some things,
> XFS is good for others.

No argument there.

Ragnar Kjorstad
Big Storage

<Prev in Thread] Current Thread [Next in Thread>