Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 19 Oct 2004 23:31:14 +0200
From:      "Poul-Henning Kamp" <phk@phk.freebsd.dk>
To:        Julian Elischer <julian@elischer.org>
Cc:        freebsd-current@freebsd.org
Subject:   Re: FreeBSD 5.3b7and poor ata performance 
Message-ID:  <94275.1098221474@critter.freebsd.dk>
In-Reply-To: Your message of "Tue, 19 Oct 2004 14:12:58 PDT." <4175835A.2050609@elischer.org> 

next in thread | previous in thread | raw e-mail | index | archive | help
In message <4175835A.2050609@elischer.org>, Julian Elischer writes:

>How do you disable GEOM? I believe it is no longer optional. Everythi g 
>runs through geom. you may however
>be not installing some modules..

GEOM cannot be disabled any more, it is our disk infrastructure now.

I havn't followed this thread, this email just happened to pass through
my "interesting vs. not-interesting" spam filter instance, so let me
just say some bits about what I guess this is all about.

Shitty benchmark design.

>>>Why is a SCSI raid-10 system slower than a plain IDE disk? Something is
>>>wrong here.

It has been proven that apples and oranges can indeed be compared,
(for instance spectroscopically, see ISBN 0-7167-3094-4 page 93),
but in general the results from such comparisons leave a lot to be
desired in the interpretation and applicability phases of investigation.

RAID10 performance characteristics are very different from a single
plain disks performance.

If you want to compare RAID10 to plain disks, do so on the same disks.

If you want to compare SCSI disks to IDE disks, then do that, but
keep RAID10 out of it, or use RAID10 on both kinds of disks.

And are the SCSI disks you are testing same age and capacity as the
IDE disks ?

Two year old SCSI technology does not compare fairly to state of
the art IDE technology (or vice versa).

A 9GB speed optimized SCSI disks do not compare fairly to a 300GB
space optimized IDE disk (even though it may not be unfair the way
you think !)

A sequential speed test like the one done here is pointless for
99.999% of all disks being used in the world.

Is your application really writing and reading data only sequentially
to raw disks ?

Please do a real and realistic test instead.  One which includes
seek times, rotational delay and read/write mixes.

Or at least realize that the quantity you are measuring is very
special case that not many people care about in practice and is in
no way indicative of a general concept like "poor ata performance".

There are tools in the ports collection for diskbenchmarking, please
use them rather than come up with some half-assed home-brew stuff.

If all you want is a simple micro benchmark, then run:

	diskinfo -t /dev/$disk ...

Thanks.


-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?94275.1098221474>