I think this bug and possibly many other AIO bugs have been around for a
long time. On Metalink you can see patches for AIO issues, but they are
usually only fixed for RedHat with no mention of SuSE. That's not good.
We have been relatively happy with SuSE SLES since 8, the ONLY issues
we've ever had have ALL been related directly to AIO and Oracle.
We can reproduce this AIO bug in SLES9 w 9.2.0.8 while using sqlldr
heavily.
> From: Alexei_Roudnev [mailto:Alexei_Roudnev@(protected)]
snip
> You may try to get a statistics, WHO and in WHICH
> CONFIGURATIONS had this
snip
> - multipath IO
Not us. We can reproduce on a stock Dell 2650 with a perc 3di internal
raid controller on internal SCSI disk. Nothing special about that gear.
> - specific file system
ReiserFS (Stock settings)
> - specific IO scheduler
Stock, whatever is configured out of the box.
> - specific kernel parameters (unlikely, but who knows)
Just what orarun package sets.
> - specific response, coming from SAN storage under special
> conditions (so that oter SAN storage wil not show this bug).
No SAN, just direct connected SCSI disk on the internal backplane.
I am certain, this bug has nothing to do with exotic hardware. It has
everything to do with pushing a BUNCH of data into the database as fast
as possible. We load spam headers into a database for analysis using
sqlldr. Maybe 10 or 20 gig a day for a couple hours at a time. Not much
really. Doing that, within 10 days, we hit the bug. If we were to run
the loads continuously, I have no doubt we'd hit the bug within 24
hours.
> From: Stout, Warren [mailto:WStout@(protected)]
snip
> I've found it rather easy to create this problem. Just having
> archive logging on and running heavy I/O ( such as doing
> several imports ) will create this async problem. I had to
> turn off archive logging to get rid of the errors.
Snip
We can reproduce with archive logging turned OFF. No doubt it would
accelerate the problem, but it is not a prerequisite. The key is
sustained heavy IO for multiple hours. We're talking enough IO to keep a
dual proc 2.4Ghz 2Gig RAM with SCSI320 10Krpm drives busy to the point
of major sluggishness. Do that all day long for 10 days and if you
haven't hit the bug, then I'd say it is fixed.
Andy
--
To unsubscribe, email: suse-oracle-unsubscribe@(protected)
For additional commands, email: suse-oracle-help@(protected)
Please see http://www.suse.com/oracle/ before posting
