Jeff Rizzo writes:
> I've been playing around with raidframe a bit recently to figure out
> what my ideal configuration for the task at hand is, and I've run into
> a couple of things that I was unable to get working, so I'm not sure
> if they're unsupported, or if I was just doing it plain wrong.
> 
> In all the configs, my goal was to be able to have the root filesystem
> on the raid set.  I'm using port-i386, so I know that some of the
> configs aren't supported in the bootblocks, so I'm booting the
> kernel off a small FFS, and forcing the root fs onto the raid in the
> kernel with
> 
> config netbsd root on raid0a type ffs
> 
> in the kernel config.

You don't need a custom kernel with this.  See 'man raidctl' under the '-A' 
option for how to get raid0a used for /.

> (scenario #1)
> I have four disks, and want the best performance (with some failure 
> recovery, so straight RAID0 is out), so my first thought was to create
> two RAID0 filesystems and mirror them with RAID1. 

A RAID 0 of two RAID 1 sets is probably a better way to go. (RAIDframe isn't 
very happy when a RAID 0 set has a component failure...)

> I got the raid0s and
> raid1 created, but when autoconfig time rolls around, only the RAID0s 
> were configured, and thus it couldn't find the root fs.... is this to 
> be expected? 

Yes. (see below).

> Is there any way I can get the RAID1 autoconfigured at boot time?

No.  Multi-level autoconfig isn't supported right now. 

> (scenario #2)
> The second thing I tried was to set up a RAID5 of the four disks.  Since
> I only have four disks total, I wanted to set up the raid5 with three
> disks and a "dummy", so I could create a degraded raid5 that I could 
> add the fourth disk to once I copied the system to the new raid5.  I have
> gotten this to work successfully with RAID1 before, but raidframe would
> not let me configure (even with -C) the 4-disk raid5 with only 3 disks.
> Is this expected?  

No.. You should be able to create a 4-disk RAID 5 set with just 3 disks and a 
'dummy'.  Note that the 'dummy' disk must have a valid entry in /dev
(but the device doesn't have to physically exist).  

> Any feedback/input would be most appreciated... I've now got it running
> a 3-disk raid5, but the write performance is so bad I'm considering
> other options. 

With a little tweaking, you can probably improve the write performance a fair 
bit... e.g. what is the block/frag size on the filesystem, and what is the 
stripe width that you've selected?  With some playing around, you should be 
able to bump up the perf numbers a fair bit... 

> I'd prefer scenario #1 if I can do it, but my next
> try will probably be two separate raid1 arrays... I suppose this will
> be a good enough compromise.

Depends on what you're using the machine for... e.g. for / and swap, it's 
maybe just sufficient to have those on RAID 1 sets... but for /u1 (or 
whatever) then a tuned RAID 5 set might work the best?  Or possibly the 
RAID 0 over two RAID 1 sets as suggested earlier?

Later...

Greg Oster