Date: Wed, 14 Jan 1998 11:15:23 +0100 From: sthaug@nethelp.no To: Amancio Hasty <hasty@rah.star-gate.com> Cc: freebsd-hackers@FreeBSD.ORG Subject: Re: Any good comparison of Linux/FreeBSD/NetBSD/OpenBSD/Hurd? Message-ID: <6373.884772923@verdi.nethelp.no>
next in thread | raw e-mail | index | archive | help
>From <34BC2FA5.E334B191@star-gate.com>:
> > this one mystifies me, since the *BSD world seems to inherit the
> > "SCSI is God's bus" attitude of the old-fashioned academic world.
> > let me just say that on the dozens of machines I run, most are IDE,
> > rock solid and deliver optimal performance from their disks.
> > and it's been this way for years.
>
> Please don't be mystified, on my FreeBSD 3.0 :
> {root} dd if=/dev/wd2s1 of=/dev/null count=100 bs=8192
> 100+0 records in
> 100+0 records out
> 819200 bytes transferred in 1.081979 secs (757131 bytes/sec)
> {root} dd if=/dev/wd2s1 of=/dev/null count=100 bs=16384
> 100+0 records in
> 100+0 records out
> 1638400 bytes transferred in 1.506444 secs (1087594 bytes/sec)
> {root}
>
> The drive is of course an IDE.
I'm not sure what you're trying to prove here, but I took this to email
instead of News - I don't have much hope for a discussion that is cross-
posted this widely (comp.unix.bsd.freebsd.misc,comp.unix.bsd.netbsd.misc,
comp.unix.bsd.openbsd.misc,comp.os.linux.misc).
I have a P-166 with 3.0-971225-SNAP and two IBM 4 GB disks: DCAS-34330W
(Wide SCSI), and DHEA-34330 (EIDE). As far as I know, this is really the
same disk, and it's nice and fast whether SCSI or EIDE.
Reading from the raw device, I get ("records in/out" removed):
# dd if=/dev/rwd0 of=/dev/null count=5000 bs=32768
163840000 bytes transferred in 16.041941 secs (10213228 bytes/sec)
# dd if=/dev/rsd0 of=/dev/null count=5000 bs=32768
163840000 bytes transferred in 20.113770 secs (8145663 bytes/sec)
Reading the same size file with block size 8k instead gives essentially
the same result. Pretty good.
(Looking at the idle time with vmstat seems to indicate that the EIDE
drive is *not* using bus mastering - I thought
ide_pci0: <Intel PIIX3 Bus-master IDE controller> rev 0x00 on pci0.7.1
indicated that bus mastering *would* be used, but evidently not so.)
Reading through the file system isn't nearly as good:
# dd if=/dev/wd0s3 of=/dev/null count=5000 bs=32768
163840000 bytes transferred in 50.196384 secs (3263980 bytes/sec)
# dd if=/dev/sd0s3 of=/dev/null count=5000 bs=32768
163840000 bytes transferred in 71.835646 secs (2280762 bytes/sec)
and with 8k block size, I get:
# dd if=/dev/wd0s3 of=/dev/null count=20000 bs=8192
163840000 bytes transferred in 175.664112 secs (932689 bytes/sec)
# dd if=/dev/sd0s3 of=/dev/null count=20000 bs=8192
163840000 bytes transferred in 72.153464 secs (2270716 bytes/sec)
So when reading through the file system the performance depends *much*
more on block size for the EIDE drive - I don't know why. But given
the right block size, *this* particular EIDE disk seems to be always
somewhat faster than the corresponding SCSI disk.
Steinar Haug, Nethelp consulting, sthaug@nethelp.no
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?6373.884772923>