Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 4 Apr 2003 15:41:35 +0930
From:      Greg 'groggy' Lehey <grog@FreeBSD.org>
To:        Lukas Ertl <l.ertl@univie.ac.at>
Cc:        freebsd-stable@FreeBSD.org
Subject:   Re: vinum performance
Message-ID:  <20030404061135.GO60909@wantadilla.lemis.com>
In-Reply-To: <20030402102225.Y19644@pcle2.cc.univie.ac.at>
References:  <20030330125138.K23911@leelou.in.tern> <20030330231000.GB1861@wantadilla.lemis.com> <20030401162820.E19644@pcle2.cc.univie.ac.at> <20030402001826.GU34617@wantadilla.lemis.com> <20030402102225.Y19644@pcle2.cc.univie.ac.at>

next in thread | previous in thread | raw e-mail | index | archive | help

--B3NBd8mrXZtPJEYR
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline

On Wednesday,  2 April 2003 at 10:24:39 +0200, Lukas Ertl wrote:
> On Wed, 2 Apr 2003, Greg 'groggy' Lehey wrote:
>
>>> I've done some rawio benchmarks with different process counts (1, 2, 4, 8,
>>> 16, 32 and 64). The results are at
>>> <http://mailbox.univie.ac.at/~le/rawio.html>.
>>
>> How does this compare to the raw disk?
>
> Oops, sorry, I've forgotten to include them on the page. They are on there
> now.
>
> Raw disk performance is indeed worse than RAID0/RAID5.

Yes, those figures look much more like what I would expect.  Those
RAID-0 figures look nice, don't they?

> Also interesting is that the sequential read performance drops
> significantly in all cases when going from 8 to 16 processes.

Yes, that's strange.  I wonder what causes it.

> Probably the bus is filled up then.

No, the various busses don't know how many processes are active.
Looking at those figures, it's clear that there's some cache activity
then.  There's no way you can transfer 52 MB/s from a single disk
platter, so it must be in cache.  Somewhere between 8 and 16
concurrent transfers from different parts of the disk must confuse the
cache.  You'll notice that after that point, the read speeds pick up
again.

Another thing of interest is that there's a slight drop-off in some
cases when going from 256 kB to 512 kB stripes.  I've seen that
elsewhere as well, but I still don't have a good explanation.

I'd be interested to see how RAIDFrame compares.

Greg
--
See complete headers for address and phone numbers

--B3NBd8mrXZtPJEYR
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.0 (FreeBSD)

iD8DBQE+jSIXIubykFB6QiMRAovBAJ9SJD3+NxBgXoaaAX+D94NC7R6lwwCfZy7J
55WZ8VMJSdLh5LjRhubdM2Y=
=U8Kg
-----END PGP SIGNATURE-----

--B3NBd8mrXZtPJEYR--



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030404061135.GO60909>