Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 25 Feb 96 14:39:36 +0100
From:      cracauer@wavehh.hanse.de (Martin Cracauer)
To:        freebsd-hackers@freebsd.org
Cc:        se@ZPR.Uni-Koeln.DE
Subject:   Re: Disk perf. with different HDs/Adapt.
Message-ID:  <9602251339.AA24996@wavehh.hanse.de>
References:  <rashid@rk.ios.com> <199602232004.AA07830@Sysiphos>

next in thread | previous in thread | raw e-mail | index | archive | help
se@ZPR.Uni-Koeln.DE (Stefan Esser) wrote:

>On Feb 23, 14:26, Rashid  Karimov wrote:
>} Subject: Disk perf. with different HDs/Adapt.

>} 	13107200 bytes transferred in 2 secs (6553600 bytes/sec)


>If you report 'dd' numbers, then **please** add at least 'time' info ...
>E.g.:

># time dd if=/dev/rsd0a of=/dev/null count=200 bs=64k
>13107200 bytes transferred in 2 secs (6553600 bytes/sec)
>        1.92 real         0.01 user         0.05 sys

I think /dev/null should not be used for benchmarks. I've seen systems
where it makes a substantial difference whether the data was read and
just not used or where the data was sent to /dev/null because the
benchmark program couldn't just discard it.

Additionally, as Stefan points of, the performance numbers of dd are
quite useless anyway.

I suggest using the 'bonnie' benchmark, which has a nice
multi-processes seek benchmark as well and does precise timing.

bonnie works on filesystems only, no raw devices. But I think the
filesystem numbers are more useful anyway.

>} 	P6-200 ASUS( hate it!) with RAID array ( all in HW no
>} 	special drivers reqd) - only 3Mb/sec ! 

>Well, RAID doesn't seem the way to go, if you are looking
>for top performance ... Were the drives synchronized and 
>was a reasonable write buffer in the controller ?

That's my impression, too. I didn't see any real fast *and*
system-independet RAID solutions so far.

>How does CCD compare ?
>It was quite good according to the last values I saw ...

I tested on NetBSD with two NCR Controllers and four Quantum 540 MB
drives. 

The ccd driver turned those old junk hardisks into one 2 GB drive with
max 8 MB/sec write and 7.5 MB read out of a standard BSD filesystem
(but I didn't find one single configuration where both of these top
numbers were possible...). One individual drive had 3.5 MB/sec. The
benchmark was bonnie, that means the test is reading one large file
sequentially in chunks of 8 KB. Surprisingly, it wasn't too difficult
to find good value for the disk geometry. Good performance even
occured when using a 64 Head/32 sectors lazy-admin scheme.

These software stringing solutions with standard harddisks/controller
seem to be a good way. I've seens amazing numbers from HP 7000
workstation as well (years ago) and you don't need any special
hardware. My impression is that most RAID solutions are not fast and
will never be because most users don't run any benchmarks on them...

Martin
-- 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Martin Cracauer <cracauer@wavehh.hanse.de>  -  Fax +49 40 522 85 36
 BSD User Group Hamburg, Germany   -   No NeXTMail anymore, please.
 Copyright 1995. Redistribution via Microsoft Network is prohibited



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?9602251339.AA24996>