Re: Raid suggestions

From: John-Paul Stewart (jpstewart_at_binaryfoundry.ca)
Date: 06/16/05 

Date: Thu, 16 Jun 2005 17:52:26 -0400

Jason L. Woodruff wrote:
> Rich Walker wrote:
>
>>Måns Rullgård <mru@inprovide.com> writes:
>>
>>
>>
>>>Rich Walker <rw@shadow.org.uk> writes:
>>
>>
>>>>Yes, in the sense that the host CPU does all the work. For Raid-0, all
>>>>that is necessary is that the block driver layer converts "request for
>>>>block NN on md0" to either "request for block MM on /dev/hda2" or
>>>>"request for block MM on /dev/hdb2".
>>>
>>>A RAID of (partitions on) hda and hdb is bad idea. These drives are
>>>on the same cable, so the performance will suffer compared to using
>>>drives on separate IDE channels. The software overhead is still the
>>>same (low), of course.
>>
>>
>>Well spotted. Mind you, (physical drive throughput)*2 < ultra-ata-133
>>throughput. Anyone fancy doing the test?
>>
>>cheers, Rich.
>>
>
> So what is the definitive answer? Current hardware solutions versus
> software raid not an ounce of difference? I must admit that my ignorance
> previously lead me to believe hardware raid PCI cards like the Promise
> card that I used was a significance performance gain when used. I did

Cheap RAID cards (e.g., Promise, HighPoint) don't really do the RAID in
hardware---they do it in software (in the driver). It's somewhat
misleading to call them "hardware RAID". Expensive RAID cards (e.g.,
3Ware) do handle the RAID in on-board hardware.

However, for RAID-0, the RAID overhead is very minimal (as another
poster already pointed out), so even with the true hardware RAID, there
is *very* little overhead to offload. So there's little (if any)
performance advantage to hardware RAID-0, ever. (I've even seen results
from $1000+ IBM ServeRAID adapters that were slower than software RAID.)

> use hdparm one time to see what kind of read and write speeds I was
> getting. The only problem was, I didn't have nothing to compare it to at
> the time. That is against the same test bed without raid or software
> raid. Maybe just going with one single 180GB drive would be the thing to
> do.

A single drive *will* be slower than any RAID-0. The bottleneck is
physical disk speed, and the only way around that is to use more disks.
  In theory, a RAID-0 of two identical disks can be up to twice as fast
as a single disk.

Here's an example from my system (albeit with 15000 RPM Ultra320 SCSI
disks---don't expect the same speeds from 7200 RPM IDE disks):

Single disk:

# hdparm -t /dev/sda6

/dev/sda6:
  Timing buffered disk reads: 212 MB in 3.03 seconds = 70.00 MB/sec

Software RAID-0 of two such partitions:

# hdparm -t /dev/md0

/dev/md0:
  Timing buffered disk reads: 420 MB in 3.00 seconds = 139.97 MB/sec

That's effectively scaling linearly. (The difference is so small it is
insignificant.) There is literally no room for a hardware RAID-0
controller to improve on those disk speeds since they're already at the
theoretical limit.

The only thing a hardware RAID controller *might* do is lower CPU load
during RAID operations. But again, there's very little of that for
RAID-0. Considering the cost of real hardware RAID, it's probably not
worth it for simple RAID-0.

Relevant Pages

  • Re: Implementing a RAID 1 Array
    ... For your RAID 1 drive is hardware RAID 1, you will not see the mirrored ... drive in the Disk Management console. ... The Volume Shadow Copy Service provides the backup infrastructure for the ...
    (microsoft.public.windows.server.sbs)
  • Re: SBS 2003 - Hal.dll Getting Currupt
    ... Consistenly getting that error would mean, in my mind, a strong possibility of disk corruption - especially with a raid set. ... If you have hardware raid, you should be able to enter the raid diagnostics at boot time, and run some tests on the controller, disks, and raid set from there. ... "Windows could not start because the following file is missing or corrupt: ...
    (microsoft.public.windows.server.sbs)
  • Re: Disk Utility to align tracks on a hardware RAID1 system
    ... That goal normally complete by hardware RAID 1. ... Microsoft CSS Online Newsgroup Support ... Disk Utility to align tracks on a hardware RAID1 system ...
    (microsoft.public.windows.server.sbs)
  • Re: Dynamic to basic disk?
    ... But I have used software RAID, so that option might be out. ... Hardware RAIDs are worth their weight in the small extra cost they will ... Just writing the "new" partition information to the disk ... What is it you are hoping to do with a software-based mirror RAID? ...
    (microsoft.public.windowsxp.setup_deployment)
  • Re: Will WoW auto adjust to new video card?
    ... >> One would think that RAID-0 would help... ... >RAID stripping won't help you cut down your head relocation times. ... >will only load the data faster if it is continual on the disk. ...
    (alt.games.warcraft)