Basic IDE - disk per channel performance
Blake Barnett
shadoi at nanovoid.com
Mon Nov 13 17:55:22 MST 2006
On Nov 13, 2006, at 4:27 PM, Michael Halcrow wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: RIPEMD160
>
> On Mon, Nov 13, 2006 at 05:15:38PM -0700, Andrew Jorgensen wrote:
>> On 11/13/06, Kenneth Burgener <kenneth at mail1.ttak.org> wrote:
>>> Is there a benefit of putting a software raid array on separate
>>> channels?
>>
>>> From The Linux Software-RAID HOWTO:
>>
>> It is very important, that you only use one IDE disk per IDE bus.
>
> That's why I have three chipsets to service 6 disks in my RAID. Don't
> put the OS disk and a RAID disk on the same channel. And don't try to
> use two SIL0680 chips in the same box (the Linux driver probably can't
> handle two cards); get a SIL0680 and a High Point or a Promise. Or,
> better yet, get a 4-channel IDE card.
>
> Keep in mind that trying to manage 6+ ribbon cables in any box can be
> a pain in the neck. The IDE connections seem to be a bit flimsy in my
> box too, leading to phantom drive failures (OS-reported failures that
> aren't really drive failures; some cable just needed to be
> reseated). The next time I build a software RAID, I am definitely
> going all SATA.
In my experience (albeit focused purely on server hardware) there are
so many downsides to doing IDE RAID that it doesn't even make sense
to use it. I suppose if you're trying to find the absolute cheapest
solution then it's viable, but then you'd be forced to ask yourself;
"How much do I value this data?" and "Why not just use software RAID
and/or rsync-type backups?". Chances are, unless it's something
you've purchased you don't really require RAID at all, and if you do,
then the cost of a really good SATA controller (w/battery backup!!)
shouldn't be an issue. I prefer the 3ware (AMCC now) 95xx series of
cards.
-Blake
More information about the PLUG
mailing list