On Fri, May 21, 1999 at 02:00:22PM -0500, Soukup, Kevin wrote:
>
> Hello linux-raid!
>
> I'm a newbie, so please forgive any stupidity. I've searched and read
> diligently and feel I need to ask this more specifically. I have:
>
>
> Dell 6300
> Dual 400 mHz Xeon w/512K cache each
> 1GB RAM
> Adaptec SCSI controller
> Dell PERC 16MB (AMI MegaRAID)
> Redhat 6.0 vanilla (2.2.5-15)
> 6-18GB Quantum 7200 RPM in
> RAID 5
Nice box :)~
>
> I've been benchmarking Oracle on this server against an identical
> Oracle instance on an IDENTICALLY EQUIPPED NT server. The Linux box
> KICKS the NT's butt, drastically, on all but the simplest, least
> loaded queries. I've got statistics if anyone is interested.
I am, very much indeed. Send them my way please, or, send them
to the list for the general amusement value.
> Given the Linux box is so much faster than the NT server, and given
> I'm using a (reputedly) slow RAID controller, I'd like to verify
> something before proceeding. IF this RAID controller is so slow,
> I can imagine the performance margin would increase if I could
> increase the RAID performance.
You could try software-RAID instead of the hardware-RAID controller.
That would be _very_ interesting, seeing HW vs. SW RAID on the same
hardware.
> All the reading I've been doing would suggest that the MagaRAID
> controller is the slowest RAID solution I could choose. Based on
> the Bonnie stats below, I THINK I recognize what is meant. The
> final two Bonnie runs on files of size 2047GB look like the IO is
> a serious bottleneck.
Never use a filesize smaller than twice the amount of RAM in your
machine. (Ofcourse, you're hitting a ceiling with 1G of memory and
2G of filesize in ext2fs).
Bonnie stats on files smaller than your amount of RAM tells you
nothing. It's large numbers alright, but cat /dev/random can do
that as well.
> But I need help and validation to continue.
>
> 1) Am I correct in reading the Bonnie numbers to say the greatest
> throughput (using a file of size 2047GB) is 9.655MB during the
> Sequential Block Input, and does that mean "write" or "read"?
Yes, that sucks.
Seq. block input is read, output is write.
> 2) Regarding the hdparm output, is it saying the best disk read
> is 12.08 MB/sec?
Yes. Four of those disks in a raid-0 should give you beyond 40 MB/s
and not 9.6 MB/s.
If you put those disks of yours on controllers than can handle them
(eg. not eight cheetas on a slow-narrow bus), you should definitely
see much much better performance with software raid.
This is four IBM UltraStar 9ES 4.5 G disks running RAID-0:
-------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
1024 6453 98.0 27118 91.3 12232 79.2 8428 98.0 33920 92.5 199.8 6.3
Machine: dual PPro-150, 256 MB RAM
Disks: 3 IBM UltraStar 9ES U2W
on Adaptec 2940U2W
and: 1 IBM UltraStar 9ES UW
on Adaptec 2940UW
One disk does around 12 MB/s, and I guess the reason this setup doesn't give me
above 40 MB/s is because the motherboard is old, and therefore the memory bandwidth
kinda sucks.
> 3) And does this reinforce the Bonnie stats?
It tells you that something smells.
You'd probably see 12 MB/s too, if you did a bonnie on one single disk.
> 4) Isn't this all pretty slow?
Indeed :)
> 5) if so, is there anything I can do about it?
Read the software RAID howto, and tell us how your disks perform after having
configured that up :)
(http://ostenfeld.dk/~jakob/Software-RAID.HOWTO/)
................................................................
: [EMAIL PROTECTED] : And I see the elder races, :
:.........................: putrid forms of man :
: Jakob Østergaard : See him rise and claim the earth, :
: OZ9ABN : his downfall is at hand. :
:.........................:............{Konkhra}...............: