Jukka Marin writes:
> I'm trying to benchmark raidframe (level 1) with different options.
> 
> First: Where can I find a good document about raidframe?  man raidctl
> is good for raidctl, but it doesn't tell too much about raid tuning.

RAID tuning could be a whole other document :)

> For example, what is the maximum value of sectPerSU?  I have tried 64
> and 128 and 128 gave me better performance, so I wanted to try 256,
> but when I raidctl -C, raidctl -I, and raidctl -iv, I get
> 
> WARNING: raid0: total sector size in disklabel (16776704) != the size of
> raid (16776960)
> Initiating re-wrrite of parity
> aid0: IO Error.  Marking /dev/wd1f as failed.
> raid0: node (Rod) returned fail, rolling backward
> WAKEUP CALLED
> raid0: IO Error.  Marking /dev/wd0f as failed.
> raid0: node (Rod) returned fail, rolling backward
> Unable to verify raid1 parity: can't read stripe
> Could not verify parity
> raid0: Error re-writing parity!
> Parity Re-write status:

Hmmmmm... 
 
> After this, raidctl -u panics the system, I have to boot in single user
> mode etc.  Is 256 an illegal value for sectPerSU?  If so, why doesn't
> the manual say so?

It shouldn't be an illegal value, but it seems that values over 128 do cause 
problems in some cases... (perhaps related to MAXPHYS?)

> (What is a clean way of wiping away the old raid parameters, anyway?
> When I raidctl -C with new parameters, it still claims the raid disk is
> clean!)

I ususally just 'dd' a bunch of zeros to the beginning of the component.
Not recommended if you like the data on those components, but it does work
for nuking the component labels.
 
> My disks are reasonably fast (49 MB/s per disk when reading two disks
> simultaneously using dd), but raidframe is slower than with my previous
> machine that had slower disks (and the same motherboard).

Use a stripe width of 128.  RAIDframe will be a little slower on writes, and 
should be at least as fast on a single read.  If you do 2 simultaneous reads, 
you should see a nice speedup.  Also note that not all disks behave the same, 
even if they are newer... (We have some 73GB U160 disks at work that will 
write at ~40MB/sec to a filesystem on them, but will only get about 5MB/sec 
writes if RAIDframe is writing to them.  A newer firmware rev of the same
disk gets us writing at 35MB/sec (through a RAID 1) and we're much happier 
now. (machine, RAM, controller, kernel, etc. were all the same -- it was just 
the disks :( ))

Later...

Greg Oster