On Thu, Jan 26, 2006 at 12:20:21PM +0100, Geert Hendrickx wrote:
> Hello, 
> 
> I'm planning to move our mail+file server to a software RAID-1.  I've been
> reading about RAIDframe, and even toyed with it in qemu[*], but I have no
> "real life" experience with it, so I still have a few questions: 
> 
> - Partitions.  Some people divide their physical disks (wd0, wd1, ...) into
>   multiple partitions, create multiple raid* devices on them, and then put
>   one (or more) filesystem partition(s) on each.  Others just create one
>   big partition on each physical drive, building one big raid0 device, and
>   put all their filesystem partitions on that (so raid0a, raid0b, raid0e,
>   ...).  Are there any specific advantages to either setup?  The only thing
>   I could think off is that you'll have more work recovering in the former
>   situation (more raid sets to rebuild).  

I setup 2 raid sets: one for root+swap and one for the rest. This way,
after a reboot because of the swap partition the root+swap raid1 will be
dirtly and have parity rebuilt, but as it's small it's not an issue.

> 
> - Swap.  Should I swap onto raid0b, or onto wd0b and wd1b?  In case of a
>   disk failure, swap on raid0b will keep working, whereas swap on wd?b will
>   not.  But I've read about problems with swap-on-raid in the past.  
>   (I know I should set swapoff=YES when swapping on raid, and I know how to
>   setup crash dumps onto a physical partition.)  

For reliability I recommend swap on raid. Using a small raid for swap makes
the parity issue a non-issue :)

> 
> - Configuration.  I've been using the configuration from the NetBSD guide: 
>   > START array
>   > 1 2 0
>   > 
>   > START disks
>   > /dev/wd0a
>   > /dev/wd1a
>   > 
>   > START layout
>   > 128 1 1 1
>   > 
>   > START queue
>   > fifo 100
> 
>   Is this ok?  I'm not sure whether/how the "layout" or "queue" sections
>   could be optimized.  


Nothing for layout, and as you have IDE disks, nothing for queues.
With SCSI disks and a smart SCSI controller, you can bump queue to 255 to
have as much queued command as possible at the drive level.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--