Oleg Goldshmidt wrote:

Muli Ben-Yehuda <[EMAIL PROTECTED]> writes:



The last discussion I've heard on the subject on lkml said that
software raid is a much better idea, since (a) hardware raid is just
software implemented on the card itself



But off-loaded from the main CPU(s), right? What is the related CPU overhead for a busy server?

I am not saying it's wrong, just that (a) above is incomplete.



To add to that - some RAID controllers have battery backup. They promise you that information in their cache will reach the hard disks, even if power fails suddenly.

Also, if you count raid buses, the hardware can sometimes achieve better performance by relying on the battery above to defer operations that would otherwise slow you down.

But, as far as I'm concerned, the main reason to use hardware raid is the fact that reassembly on new disk insertations happen immediately. I wrote a rant here a while back on why the current partionable raid in 2.6 is not enough to achieve that, just yet.

<>and (b) that way you are not tied to a specific vendor's metadata
format, with everything such a dependency entails.

Yes, I agree. Hardware raid has it's drawbacks too. As an example, I know how to migrate a RAID-1 running on two disks to a RAID-5 running on three. I'm not sure how to do that with hardware raid, if it's at all possible.

--
Shachar Shemesh
Lingnu Open Source Consulting ltd.
http://www.lingnu.com/


================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]



Reply via email to