[3760] in linux-scsi channel archive
Re: LINUX source code
daemon@ATHENA.MIT.EDU (Larry M. Augustin)
Fri Apr 24 18:23:21 1998
Date: Fri, 24 Apr 1998 11:01:03 -0700
From: "Larry M. Augustin" <lma@varesearch.com>
To: Zlatko.Calusic@CARNet.hr
Cc: Ian Stirling <root@mauve.demon.co.uk>, linux-scsi@vger.rutgers.edu
In-Reply-To: <87n2db1lr8.fsf@atlas.infra.CARNet.hr>
There's a lot more to getting good performance out of a drive than
just the seek time and rotational speed. Consider:
rpm buffer avg seek t-t seek internal transfer
Fujitsu M2954ESP 7200 506K 8.0ms 0.8ms 9.9-14.0 MB/sec
Seagate Cheetah 10,000 2M 8.2ms 0.8ms 15.25-22.1 MB/sec
The Fujitsu web page doesn't say if their internal transfer rate is
formatted or unformatted. But either way it is on the low side. The
7200 Atlas II has internal transfer rates up to 22.5MB/sec. The 506K
buffer seems like a strange size. I would think you would want at
least a full sector buffered. The Cheetahs are available with up to
2MB buffer, and the Atlas III has 1MB. I usually take about 70% of
the average internal transfer speed and expect that to be close to the
bonnie/iozone numbers. For the Fujitsu that's 8365. For the Cheetah
that's 13000, which is about what we see in benchmarks.
The hdparm results you get for the Fujitsu seem about right.
Bottom line: I'd expect your Fujitsu drive to get about 8MB/s
sequential block input numbers, and you're seeing close to that with
hdparm. Maybe there's something else wrong in your setup (I'd like to
try the Fujitsu disk on a fast PII with a real SCSI controller), but
the Fujitsu M2954ESP doesn't look like its going to be a performance
leader.
Larry
Larry Augustin, President Tel: +1.650.934.3666 x124
VA Research, Inc. Fax: +1.650.964.7668
1235 Pear Ave., Suite 109 Email: lma@varesearch.com
Mountain View, CA 94043 Web: http://www.varesearch.com
Zlatko Calusic writes:
> Ian Stirling <root@mauve.demon.co.uk> writes:
>
> > <snip>
> > > I found that WDC AC22000L IDE UDMA 2GB drive (5400rpm) is slightly
> > > _faster_ than Fujitsu M2954ESP UltraSCSI 4GB drive (7200!rpm). How can
> > > it be???
> > >
> > > My setup: Linux v2.1.97, glibc, P166MMX, 32MB RAM, Adaptec 2940UW.
> > >
> > > SCSI:
> > > -------Sequential Output-------- ---Sequential Input-- --Random--
> > > -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
> > > Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
> > > 100 1614 76.3 5607 27.7 2440 23.4 1638 71.2 6583 25.3 92.4 3.9
> > > ^^^^ ^^^^
> >
> > Ahh, but look at the seeks ^
> >
> > > UDMA:
> > > -------Sequential Output-------- ---Sequential Input-- --Random--
> > > -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
> > > Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
> > > 100 1564 74.2 7443 42.2 2815 28.8 1766 76.5 8092 35.2 69.8 2.6
> > > ^^^^ ^^^^
> > >
> > ^
> >
> > The rotational speed has little to do with the transfer speed, unless
> > all other things are exactly equal.
> > What are you using to measure that?
> > I get 10.5Mb/sec using quantum bigfoot 8Gb, (hdparm -t) and my brother reports
> > 15Mbish with a fuji drive.
> > Both TX motherboards.
> >
> >
>
> Well, yes, SCSI drive is seeking faster, but shouldn't higher
> rotational speed also read data faster from the media?
>
> What bothers me is that many people report speeds of >= 10MB/s with
> their drives (even cheaper IDE ones) and it seems I can't get more
> than around 7MB/s with so called Enterprise UW SCSI drive with big
> capacity and high rotational speed???
>
> I made above tests with "bonnie" program. Altavista is your friend. :)
>
> Testing with hdparm -t:
>
> {atlas} [~]# hdparm -t /dev/sda
>
> /dev/sda:
> Timing buffered disk reads: 32 MB in 3.71 seconds = 8.63 MB/sec
> {atlas} [~]# hdparm -t /dev/hda
>
> /dev/hda:
> Timing buffered disk reads: 32 MB in 3.70 seconds = 8.65 MB/sec
> {atlas} [~]# hdparm -t /dev/sda
>
> ??????
>
> I probably don't need to mention that the SCSI drive costs much more
> than the IDE one. SCSI controller is another story. :)
>
> Bye!
> --
> Posted by Zlatko Calusic E-mail: <Zlatko.Calusic@CARNet.hr>
> ---------------------------------------------------------------------
> Reputation: what others are not thinking about you.
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
> the body of a message to majordomo@vger.rutgers.edu
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.rutgers.edu
