Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 29 Aug 2000 11:01:49 +0930
From:      Greg Lehey <grog@lemis.com>
To:        Garance A Drosihn <drosih@rpi.edu>
Cc:        efinley@efinley.com, freebsd-questions@FreeBSD.ORG
Subject:   Re: softupdates VS. SCSI (Greg Lehey)
Message-ID:  <20000829110149.C11422@wantadilla.lemis.com>
In-Reply-To: <v04210101b5d04c1de476@[128.113.24.47]>; from drosih@rpi.edu on Mon, Aug 28, 2000 at 01:13:47PM -0400
References:  <39aa81fb.27249289@mail.afnetinc.com> <v04210101b5d04c1de476@[128.113.24.47]>

next in thread | previous in thread | raw e-mail | index | archive | help
On Monday, 28 August 2000 at 13:13:47 -0400, Garance A Drosihn wrote:
> At 3:27 PM +0000 8/28/00, Elliot Finley wrote:
>> Hello,
>>     I was thinking about softupdates/IDE VS. softupdates/SCSI.
>> It doesn't seem like the SCSI has much advantage as far as
>> writing to the disk.  But it _does_ seem like the SCSI would
>> have a _big_ advantage when reading from the disk because it
>> is able to reorder the reads to be more efficient.
>>
>>     So the question to all you drive experts (Greg are you
>> listening?) is:  does this sound right?
>
> Not that I am a disk expert, but I think the main question is
> whether this machine has multiple disks.  A second important
> question is WHAT KIND of IDE (ata33, ata66, ata100) vs WHAT
> KIND of SCSI (regular, wide, ultra-wide, etc).  What are the
> spindle speeds of the drives?
>
> If you have a single-disk system, then chances are you can
> get fine performance from the latest IDE drives.  I think
> it is only when you're getting into multiple-disk setups
> that you would be likely to see a big advantage from SCSI.

There's a certain truth in this.  IDE controllers can only handle one
request at a time.  This means you send a request to the disk.  The
disk spends about 8 ms positioning, during which time the bus is idle,
and then a few hundred microseconds transferring data.  A good SCSI
host adaptor handles multiple concurrent transfers.

Obviously this only makes any difference if you have more than one
disk on the IDE controller.  It's not a problem with multiple IDE
controllers, and that's the way the industry is going.  All
motherboards in current production have at least two controllers, and
some have more.  You can only put a single ATA66 or ATA100 device on a
controller.

Where SCSI also wins is that the drives themselves are typically
faster.  But there's not much in it.  Recently the German magazine c't
published a comparison of 331 different disk drives.  They showed an
"average" speed, which for IDE drives was between 4 and 21 MB/s, and
for SCSI between 3 and 27 MB/s.  But the two fastest IDE drives (IBM
DLTA 307030 and DLTA 307075, with 20.7 and 20.6 MB/s) were faster than
all except one SCSI drive (Quantum Atlas 10K II, 26.5 MB/s).  The next
fastest SCSI drive (Seagate ST318451LC) only did 19.5 MB/s, slower
than yet another IBM IDE drive.

FWIW, I've been quite happy with IBM drives, and over the years I've
noticed that they've been some of the fastest IDE drives.  They're
also probably less than half the cost per MB of a SCSI drive.

The MB/s are only part of the story, though.  Looking at these same
drives, the IDE drives have random access times between 6 and 11ms.
The first value is for the first 504 MB, the second for the whole
disk.  Guess which one the manufacturers use :-)

By contrast, the SCSI drives have 4.7 and 8.1 ms (Quantum) and 3,2 and
5.2 ms (Seagate).  Obviously these would be better choices if you were
doing lots of really random accesses.

Finally, a word of warning on these speeds.  Most disk benchmarks cut
corners, and they give you the impression that the disks are really
fast.  Even the ultra-fast Seagate, with 5 ms positioning time, can
only do 200 transfers a second with random access.  An average
transfer is about 6 kB, so that automatically limits its speed to
about 1.2 MB/s.

Greg
--
When replying to this message, please copy the original recipients.
For more information, see http://www.lemis.com/questions.html
Finger grog@lemis.com for PGP public key
See complete headers for address and phone numbers


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-questions" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20000829110149.C11422>