On Sat, 2004-11-06 at 09:19 -0500, AWC Maillists wrote: > installed onto lower-end motherboards. By what you are saying it sounds > like these "RAID" controllers are kind of like Winmodems... offloading > the real work to the CPU and OS. Or am I still not getting this right?? yeah you are getting it right.... but the driver has to be binary only of course... because RAID1 is highly proprietary! (ha! there's like only so many ways you can do raid1 :) fwiw, in practice linux MD raid will perform better than the software raid done by such drivers. The reason is in the on disk layout; linux md raid has the guarantee that the start of the partition is on a 4Kb boundary (by virtue of being per partition); the fakeraid drivers have to raid teh entire disk, which means partitions can (and generally will, since such raid formats advertise an non-even number of heads/cylinders) start not-on-a-4k boundary. If you do raid 0, this means that the places where the strides go to the other disk, are also not on a 4Kb boundary. Now the kernel will almost always do 4Kb (pagesize) IO's, when then have to be split on every stride boundary, which 1) is not cheap, and 2) means that you may have to wait for 2 disks to seek instead of one, etc etc.
Attachment:
signature.asc
Description: This is a digitally signed message part