On Fri, Mar 15, 2002 at 03:50:56AM -0500, Arjan van de Ven wrote: > On Fri, Mar 15, 2002 at 02:20:53AM -0500, Marc Horowitz wrote: > > I've read the last few months of posts, and I'm still confused about > > something. > > > > If chipsets such as the Promise PDC20265 and Highpoint HPT372 are just > > extra IDE controllers, is the *only* reason for the special linux > > support to be able to share arrays with windows? > > Basically yes. The only other nice thing is being able to boot from it. > > > > And the $64,000 question (or maybe the $50 question :-) If I'm not > > going to run windows on a machine, is there any reason to get a > > motherboard with a "RAID" controller? The only reason I can think of > > is the extra IDE channels; if I do get a RAID board for this reason, > > is there any reason not to use /dev/mdX? > > I personally prefer md because it avoids a fundamental issue > with performance (ataraid needs to some requests up due to the offset it > gets from the partition table; md never has to "abuse" requets) Is it safe to run a mix of ataraid and md? My test system is as such: 1 set of raid1 drives configured in BIOS so that I can boot from them HOWEVER, once the system comes up, I use linux software RAID without the ataraid/pdcraid drivers loaded when I want to write a LILO record, I modprobe pdcraid and use disk=/dev/ataraid/d0 in my lilo.conf. After LILO stops, I remove the pdcraid/ataraid modules and never use them again. ..and I'm using the kernel support for the Promise chipset to handle the IDE controller itself. thoughts? john R
