Re: raidframe performance question

[Thor Lancelot Simon <tls%panix.com@localhost>]

On Thu, May 05, 2011 at 11:08:26PM -0600, Michael L. Hitch wrote:
> On Thu, 5 May 2011, Paul Goyette wrote:
> 
> >The next step was to initialize the raidset parity, so I used raidctl -i
> >
> >That was almost 10 hours ago, and so far it has done only 26% of
> >the job!  It appears that the writes are taking ~forever to
> >complete! systat(1) shows that wd5 (I'm assuming that's the
> >"parity" drive being written) has twice the transfer-ops and twice
> >the bytes-transferred, but the drive is 100% busy!
> 
>   I wouldn't consider it normal.  I would suspect that the write
> cache on the drive is off, and writes are very slow.  You could
> check that with dkctl, and enable the write cache for the parity
> initialization if the write cache is off.

I have seen this problem with parity rebuilds and RAIDframe on an
intermittent basis for years.  It seems to afflict some machines all
the time, others never.

I wonder if this has something to do with the missed wakeups Matt
reported elsewhere -- the I/Os complete, but the buffer's callback never
gets hit until the next clock tick, so at every clock tick the disk looks
busy.

Thor