Subject: Re: RAIDframe, anyone?
To: Hubert Feyrer <feyrer@rfhs8012.fh-regensburg.de>
From: Andrew Gillham <gillhaa@ghost.whirlpool.com>
List: current-users
Date: 11/22/1998 14:14:10
Hubert Feyrer writes:
> 
> Did anyone here have any success with RAIDframe? I've given it several
> tries on my i386 (runnign -current), and it always hung the machine for
> me. I've tried putting the raid onto several partitions one one disk and
> also tried spreading it on several disks, but it's always the same.

I am using it (well ok, just now got it configured) with 4x1.7GB disks.
They're Micropolis 2217 SCSI drives.  I have a 'e' partition that is
the whole disk, and sd1e/sd2e/sd3e/sd4e are a RAID-5 array.  Works out to
be a 5003MB array.  I newfs'd /dev/raid0d directly and have it mounted 
under /mnt.  Using iozone and bonnie I've been able to get some very basic
stats.  I don't have any for the bare drives, so I can't really compare.
The machine is only a 486/33, so performance is not great.  It took almost
2 hours to rewrite parity on the array. I'm seeing about 650K/s writes
and 1.5MB/s reads.  This is on an Adaptec 1542B controller.  I doubt
my machine could ever exceed 3MB/s on the bare drive, more likely less
than 2MB/s.  So.. my assumption is that reads on RAID-5 should approach
reads on RAID-0, so I'm curious how ccd will compare on these disks.
If I have time, I'll redo the drives as a ccd.

> I've observed (using "vmstat -m") that the hangs occurs as soon as the
> "devbuf" pool (?) is full.

I haven't seen this. (yet?)  I did crash my other PC when I tried to
create a RAID-5 array on 3 vnd devices/files.  Everything worked fine
up to the point of running Bonnie, when it panic'd with a kmem_map full.
Hitting 'c' in ddb caused a 'locking against myself' panic.  I have
the dump from this if anyone is interested.  I didn't build the kernel
with '-g' though.

One thing that does concern me, is that initially I had a bad block on
one of my 2217s.  During the parity rewrite, the SCSI controller choked
on the bad block, and subsequently RAIDframe panic'd.  My concern would
be that an actual failure might panic the box, causing the array to
be essentially useless for reliability.  Since I have my drives in an
external "hot pluggable" cabinet, I can easily check the results of
yanking one drive.

So far, RAIDframe seems pretty cool!

-Andrew
-- 
-----------------------------------------------------------------
Andrew Gillham                            | This space left blank
gillham@whirlpool.com                     | inadvertently.
I speak for myself, not for my employer.  | Contact the publisher.