Recently there was a classic thread here on this list: scsi vs ide.
I've got both SCSI and an IDE systems. Since I've been beginning to
doubt myself whether SCSI is really worth the (hugely extra) cost, I
did some quick informal testing with bonnie.
What follows are some bonnnie results from which you might be able to
draw some conclusions. My tests certainly weren't scientific or
rigorous, but they do give a general idea about this topic.
Both tests used the same self-compiled bonnie version 1.02a. I didn't
go to the trouble to completely unload both systems though. They were
both running [EMAIL PROTECTED] (at nice level 20). Both could have been doing
some other random proccessing as well (e.g. NAT serving, reading
email, etc), but nothing too heavy (other than seti).
So first is sewage, my workstation. It's stats are as follows:
- Abit KT7 (via kt133 chipset) motherboard
- AMD Thunderbird 1 GHz CPU
- 512 MB RAM
- Tekram DC-390u2w PCI SCSI controller
- Quantum Atlas II 10k RPM 18 GB hard drive
The bonnie commandline was as follows:
sewage:/data1/bonnie# bonnie++ -d /data1/bonnie/ -s \
1500 -n 8 -u garman:garman
Results (note: these results use exactly 80 columns. Make sure you're
screen is at least 80 columns wide. You'll also need to use a
fixed-width font):
Version 1.02a ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
sewage 1500M 5741 66 27494 17 14210 12 5131 58 30689 16 208.0 2
------Sequential Create------ --------Random Create--------
-Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
8 3937 74 +++++ +++ +++++ +++ 3823 74 +++++ +++ 9792 76
sewage,1500M,5741,66,27494,17,14210,12,5131,58,30689,16,208.0,2,8,3937,74,+++++,+++,+++++,+++,3823,74,+++++,+++,9792,76
The second computer, septictank, is my diald server, NAT box,
firewall, etc. Relevant stats are as follows:
- Gigabyte ga-7ixe4 (amd 751 chipset) motherboard
- AMD Duron 750 MHz CPU
- 256 MB RAM
- IBM Deskstar 60GXP 20 GB hard drive (7200 rpm I think)
Commandline (same as above except directory):
septictank:/wrk/bonnie# bonnie++ -d /wrk/bonnie/ \
-s 1500 -n 8 -u garman:garman
Results:
Version 1.02a ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
septictank 1500M 2863 40 5044 22 2433 24 2748 56 4849 44 87.4 1
------Sequential Create------ --------Random Create--------
-Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
8 1401 83 +++++ +++ +++++ +++ 1532 84 +++++ +++ 3096 85
septictank,1500M,2863,40,5044,22,2433,24,2748,56,4849,44,87.4,1,8,1401,83,+++++,+++,+++++,+++,1532,84,+++++,+++,3096,85
That IBM hard disk in septictank is an ATA/100 disk I believe. The
Gigabyte motherboard supports ATA/66, but I'm not sure if the Linux
kernel supports it.
So I went ahead and bought a Promise TX2 Ultra ATA/100 PCI card. I
recompiled my kernel, same as before but with support for the Promise
card. I installed the kernel, powered down, hooked up the hardware,
booted and re-ran bonnie. Everything was roughly the same as before
(including [EMAIL PROTECTED]), except for the Promise card. I used the same
bonnie commandline.
Results were as follows:
Version 1.02a ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
septictank 1500M 6004 80 23009 12 7690 6 6489 87 24244 14 107.0 0
------Sequential Create------ --------Random Create--------
-Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
8 1674 94 +++++ +++ +++++ +++ 1545 86 +++++ +++ 3587 98
septictank,1500M,6004,80,23009,12,7690,6,6489,87,24244,14,107.0,0,8,1674,94,+++++,+++,+++++,+++,1545,86,+++++,+++,3587,98
These (completely un-rigorous) tests seem to suggest that the
performance margin between SCSI and IDE isn't as great as the price
margin. Depending on the feedback I get, I might do some more
testing. Does anyone have any ideas/suggestions/thoughts on how I can
achieve some more data? E.g., other methods, other programs,
...anything that doesn't involve buying stuff :)
Hope somebody finds this interesting!
Matt
--
Matt Garman, [EMAIL PROTECTED]
``I ain't never seen no whiskey, the blues made my sloppy drunk!''
-- Sleepy John Estes, ``Leaving Trunk''
