SCSI vs IDE on wimpy Win2K server (IMAP)

I was considering switching from IDE to SCSI to improve disk
performance on a small Windows box running IMAP.

Would switching from a Maxtor IDE (DiamondMAX) to a wide SCSI disk @
10K on Utra160 drastically improve the time IMAP takes to fetch say,
2000 messages? (present IDE filesystem is NTFS + 4KB clusters).

I now have a 30-second wait, over which Perfmon shows high disk I/O
from my IMAP server process (Mercury/32), and a disk queue of only
1.0, low CPU %, and low I/O by all other related processes. (I'm also
running Apache, Horde MySQL, and IMP -- a PHP-based Webmail server).

The system is a 1GHz P3 w/ 256MB, with a Maxtor D740X IDE (rated for
44MB/sec at OD & 24MB/sec at ID) which (via Dr. Hardware) shows a
respectable ~35MB/sec read and ~26MB/sec write, ~5.8ms access time.

Is that speed decent or pathetic?

Note that I have a 40-pin cable, and I'm not sure how to interpret Dr.
Harware's details -- it says Single DMA '-', Multi DMA '0 1 2' UDMA/33
'0/33 1/33 2/33', UDMA/66-133 '3/66 4/66 5/100(a) 6/133', (E)PIO '3 4'
Do i need to switch to an 80-pin cable to enable DMA?

Thanks. Jaz
(Please excuse the 'burp' when replying)

NTFS is poor under load pattern of "updates to lots of tiny files". If your
IMAP software is such - then either change this software or change the OS or
change the FS to FAT32.

On the other hand, MySQL is also poor, and is used on UNIXen only due to
lack of more decent and free alternatives. On Windows, you have the free Jet
database engine (.MDB file) which is better then MySQL.

--
Maxim Shatskih, Windows DDK MVP
StorageCraft Corporation

http://www.storagecraft.com

"Jaz" wrote in message
...


I was considering switching from IDE to SCSI to improve disk
performance on a small Windows box running IMAP.

Would switching from a Maxtor IDE (DiamondMAX) to a wide SCSI disk @
10K on Utra160 drastically improve the time IMAP takes to fetch say,
2000 messages? (present IDE filesystem is NTFS + 4KB clusters).

I now have a 30-second wait, over which Perfmon shows high disk I/O
from my IMAP server process (Mercury/32), and a disk queue of only
1.0, low CPU %, and low I/O by all other related processes. (I'm also
running Apache, Horde MySQL, and IMP -- a PHP-based Webmail server).

The system is a 1GHz P3 w/ 256MB, with a Maxtor D740X IDE (rated for
44MB/sec at OD & 24MB/sec at ID) which (via Dr. Hardware) shows a
respectable ~35MB/sec read and ~26MB/sec write, ~5.8ms access time.

Is that speed decent or pathetic?

Note that I have a 40-pin cable, and I'm not sure how to interpret Dr.
Harware's details -- it says Single DMA '-', Multi DMA '0 1 2' UDMA/33
'0/33 1/33 2/33', UDMA/66-133 '3/66 4/66 5/100(a) 6/133', (E)PIO '3 4'
Do i need to switch to an 80-pin cable to enable DMA?

Thanks. Jaz
(Please excuse the 'burp' when replying)

I was considering switching from IDE to SCSI to improve disk
performance on a small Windows box running IMAP.

highly debatable motivation.

Would switching from a Maxtor IDE (DiamondMAX) to a wide SCSI disk @
10K on Utra160 drastically improve the time IMAP takes to fetch say,
2000 messages? (present IDE filesystem is NTFS + 4KB clusters).

the scsi probably has faster seeks, but since it sounds more than
a couple years old, its bandwidth is probably poor relative to the IDE
(again, depends mainly on the generation/density/vintage.)

I now have a 30-second wait, over which Perfmon shows high disk I/O
from my IMAP server process (Mercury/32), and a disk queue of only
1.0, low CPU %, and low I/O by all other related processes. (I'm also
running Apache, Horde MySQL, and IMP -- a PHP-based Webmail server).

sounds like lots of seeks. do you get any sort of breakdown regarding
the size distribution of your IOs?

The system is a 1GHz P3 w/ 256MB, with a Maxtor D740X IDE (rated for
44MB/sec at OD & 24MB/sec at ID) which (via Dr. Hardware) shows a
respectable ~35MB/sec read and ~26MB/sec write, ~5.8ms access time.

ouch. you should put more ram in the machine first. and no, those
are not respectable performance numbers, at least by current-gen standards.

it's also true that switching the machine to Linux may well give a
performance boost (at application, pagecache, filesystem and low-level
IO levels.)

Note that I have a 40-pin cable, and I'm not sure how to interpret Dr.
Harware's details -- it says Single DMA '-', Multi DMA '0 1 2' UDMA/33
'0/33 1/33 2/33', UDMA/66-133 '3/66 4/66 5/100(a) 6/133', (E)PIO '3 4'
Do i need to switch to an 80-pin cable to enable DMA?

first, you ALWAYS want a UDMA mode, since without it, the integrity of
transfers is not checked (indeed, pre-ultra scsi also has no checksum.)
and you always want an 80-wire cable, since it never hurts and may help.
IIRC, UDMA-33 does not actually require 80-wire.

regards, mark hahn.

it's also true that switching the machine to Linux may well give a
performance boost (at application, pagecache, filesystem and low-level
IO levels.)

Dunno on Linux, but Samba on FreeBSD is 70% slower then native Windows
fileserver, at least under load pattern of transferring ~100MB files. And
FreeBSD was always considered to be "serious" OS (compared to "toy" Linux).

The known weak point of Windows is bad NTFS performance about updates to lots
of tiny files. As about low-level IO - sorry, Linux is a new to zero-copy IO,
while Windows used it since 1993.

--
Maxim Shatskih, Windows DDK MVP
StorageCraft Corporation

http://www.storagecraft.com

On Wed, 04 May 2005 15:22:31 GMT, Jaz wrote:

The system is a 1GHz P3 w/ 256MB, with a Maxtor D740X IDE (rated for
44MB/sec at OD & 24MB/sec at ID) which (via Dr. Hardware) shows a
respectable ~35MB/sec read and ~26MB/sec write, ~5.8ms access time.

Changing this drive to anything else could be worth a try. I hope you
have a very good backup solution, since this drive is prone to
failure.

Nick