More Oracle Async IO

More Oracle Async IO

Post by Adam Walke » Wed, 04 Jul 2001 00:46:35



This thread is around still and I have heard speculation but nothing
definite
about AIX 4.3.3 (with ML08) and still receiving async oracle crash errors
during high volume.  Is there a way around this (besides not using async)
Is something missing?

DBW0: terminating instance due to error 27061
Instance terminated by DBW0, pid = 45860
User:oracle On someserver
SID:sumsid
/opt/oracle/admin/sumsid/bdump> oerr ora 27061
27061, 00000, "skgfospo: waiting for async I/Os failed"
// *Cause:  aiowait function returned error
// *Action: check errno

Adam

 
 
 

More Oracle Async IO

Post by KIRill Terechenk » Wed, 04 Jul 2001 03:28:00




Quote:> This thread is around still and I have heard speculation but nothing
> definite
> /opt/oracle/admin/sumsid/bdump> oerr ora 27061
> 27061, 00000, "skgfospo: waiting for async I/Os failed"
> // *Cause:  aiowait function returned error
> // *Action: check errno

Try #smit aio.

 
 
 

More Oracle Async IO

Post by Jose Pina Coelh » Thu, 05 Jul 2001 11:14:03



Quote:> This thread is around still and I have heard speculation but nothing
> definite
> about AIX 4.3.3 (with ML08) and still receiving async oracle crash errors
> during high volume.  Is there a way around this (besides not using async)
> Is something missing?

> DBW0: terminating instance due to error 27061
> Instance terminated by DBW0, pid = 45860
> User:oracle On someserver
> SID:sumsid
> /opt/oracle/admin/sumsid/bdump> oerr ora 27061
> 27061, 00000, "skgfospo: waiting for async I/Os failed"
> // *Cause:  aiowait function returned error
> // *Action: check errno

So, what was the value of errno ?  (It's one another of the oracle log
files)

--
Doing AIX support was the most monty-pythonesque
activity available at the time.

 
 
 

More Oracle Async IO

Post by Pete Kayte » Thu, 05 Jul 2001 11:35:47


Adam,

Not sure of ML08 (we're running ML07) but we've seen problems if aio is
configured with too few server (kernel) processes for the workload and your
Oracle datafiles are filesystem based. I'm wondering if the error your
seeing is really a timeout because all aio server procs are busy. The number
of aio server procs is configurable via smitty aio. The default settings are
definitely too low for Oracle and a reboot is required to activate new
settings.

Pete


> This thread is around still and I have heard speculation but nothing
> definite
> about AIX 4.3.3 (with ML08) and still receiving async oracle crash errors
> during high volume.  Is there a way around this (besides not using async)
> Is something missing?

> DBW0: terminating instance due to error 27061
> Instance terminated by DBW0, pid = 45860
> User:oracle On someserver
> SID:sumsid
> /opt/oracle/admin/sumsid/bdump> oerr ora 27061
> 27061, 00000, "skgfospo: waiting for async I/Os failed"
> // *Cause:  aiowait function returned error
> // *Action: check errno

> Adam


 
 
 

More Oracle Async IO

Post by Christian GILBER » Sat, 07 Jul 2001 08:46:49


There's an interoperability bug with Oracle on AIX 4.3+ !
You SHOULD NOT USE ASYNC I/O with Oracle unless the alert is closed !
Try using multiple dbwriters instead !!!

Hope this helps !!

Pete Kaytes a crit :

> Adam,

> Not sure of ML08 (we're running ML07) but we've seen problems if aio is
> configured with too few server (kernel) processes for the workload and your
> Oracle datafiles are filesystem based. I'm wondering if the error your
> seeing is really a timeout because all aio server procs are busy. The number
> of aio server procs is configurable via smitty aio. The default settings are
> definitely too low for Oracle and a reboot is required to activate new
> settings.

> Pete



> > This thread is around still and I have heard speculation but nothing
> > definite
> > about AIX 4.3.3 (with ML08) and still receiving async oracle crash errors
> > during high volume.  Is there a way around this (besides not using async)
> > Is something missing?

> > DBW0: terminating instance due to error 27061
> > Instance terminated by DBW0, pid = 45860
> > User:oracle On someserver
> > SID:sumsid
> > /opt/oracle/admin/sumsid/bdump> oerr ora 27061
> > 27061, 00000, "skgfospo: waiting for async I/Os failed"
> > // *Cause:  aiowait function returned error
> > // *Action: check errno

> > Adam


--
Christian GILBERT

IBM Certified Specialist - AIX System Administration
 
 
 

More Oracle Async IO

Post by cogg » Wed, 01 Aug 2001 07:04:59


unless the alert is closed? can you explain?

pardon my ignorance :/

--
Posted via dBforums, http://dbforums.com

 
 
 

More Oracle Async IO

Post by Cinzia Innocen » Wed, 01 Aug 2001 16:49:08



> unless the alert is closed? can you explain?

> pardon my ignorance :/

Hello!

We are experiencing a similar problem. We have an IBM H80, AIX 4.3.3,
ML08, 2 processors, 2GB RAM, 2 internal disks and an external disk
array, with Oracle 8.1.6, Oracle iAS9i, MQSeries, and Actuate
e.Reporting.

The box has been working fine until 15 days ago or so. It crashes,
apparently due to hardware errors (SCSI BUS ERROR scsi0, DISK
OPERATION ERROR hdisk0 hdisk1). We are mirroring all system
filesystems, including swap space. I say apparently because IBM has
not found any hardware errors when they come to check it (they have
been here a lot lately!) and think this is a software problem.

As soon as the OS detects errors, the databases "crash" giving the
following error:
Errors in file /u01/oracle/admin/db1/bdump/ckpt_34316_db1.trc:
ORA-00206: error in writing (block 3, # blocks 1) of controlfile
ORA-00202: controlfile: '/u01/oradata/db1/redo/log1/control01.ctl'
ORA-27063: skgfospo: number of bytes read/written is incorrect
IBM AIX RISC System/6000 Error: 5: I/O error
Additional information: -1
Additional information: 8192
Mon Jul 30 14:28:06 2001
CKPT: terminating instance due to error 221
Instance terminated by CKPT, pid = 34316

(BTW, the first set of redo logs are on rootvg, the other set on the
external disk array.)

From this moment, no command works, including OS. The only thing left
is to reboot the machine (shutdown doesn't work either) by pressing
the reset button.

We have been rebooting the machine 3-4 times everyday. It has even
crashed at 2 A.M. when no one is in the office.

We have increased the maximum number of asynchronous I/O servers to
100 (using pstat we've seen a maximum of 38 when everything works),
but nothing improved.

At most we've seen a total of 0.6% of swap space used (3GB swap space:
2GB /dev/hd6, 1GB on external disk array).

Can someone please give further explanations regarding the Oracle bug
with async I/O?

What do you mean by using multiple dbwriters?

Any help is appreciated!

Cinzia

Cinzia Innocenzi
Finsiel S.p.A.
Via Carciano, 4
00131 Roma, Italia

 
 
 

1. async IO on UDP ports and SIGIO

In porting xntpd 3 to OSx 5.1, I've discovered that when xntpd 'NOTTY's
itself off the terminal, xntpd then fails to respond to incoming datagrams
using the signal mechanism.

From what I can see, it's does all the right things, sets the fd up using
FNDELAY and FASYNC and does an F_SETOWN.

I vaguely remember getting this to work before, but I've since forgotten
how I did it O:-(

I also suspect OSx 5.1 to be 4.2BSDish.

Any tips ?

thanks,
Darren

2. forking

3. Solaris 2.2 async io problem

4. Ultra 2 scsi transfer rates

5. async IO

6. alpha/numeric paging tools/Drivers

7. ASYNC-IO on SVR4 system

8. RH7.3 Network configuration

9. Async IO using threads

10. Async IO in Linux...

11. async io

12. dynamic allocation of async io structs in solaris?