Re: [suse-oracle] SLES9 SP4 and Async IO - Fixed

Hi All,

For all, who are upgrading to SLES-9 SP4 on AMD machines and are detecting
real bad performance: Try to set the kernel-boot-parameter c1e_disable=off
;-)

See
http://www.novell.com/support/search.do?cmd=displayKC&docType=kc&externalId=
3408793&sliceId=SAL_Public&dialogID=59671473&stateId=0%200%2059669967

We experienced a performance-degration of batch-processes on our oracle-db
by factor 5 after apply the SLES9-SP4 Kernel....Going back to the SP3-Kernel
(286) performance was good again.

After Disabling the c1e-AMD-Mode (double-negation) the performance is back
again...with the SP4 kernels 308 and 311..

Is there anyone out there with the same problem/solution?

Bye
   Frank

-----Ursprüngliche Nachricht-----
Von: [email protected]
[mailto:[email protected]] Im Auftrag von Arun Singh
Gesendet: Mittwoch, 2. April 2008 21:50
An: [email protected]
Betreff: Re: [suse-oracle] SLES9 SP4 and Async IO - Fixed

Hi All,

We have fixed this problem and released kernel# 2.6.5-7.311.PTF.369361.0.
This PTF is available via Novell Tech Support - bug#369361 .

This fix will be part of next SLES9 kernel update (post .311).

Thanks,
Arun

>>>> <[email protected]> wrote:
>> AsyncIO problem have been retested in VMWare.
>>
>> Yes, it is 100% the bug.
>>
>> Here is instruiction, how to reproduce the problem:
>> 1) Instal SLES9 SP4 on x86_64 (I did not tested on i386 but I am almost 
>> sure
>>
>> that bug exists there, too).
>> 2) Install Oracle 10.2.0.3 (10.2.0.1 + patchset; may be, 10.2.0.1 is
>> enough). Use file system for data files.
>> 3) Create a database with Flash Recovery Area and Archive Logs. You can 
>> use
>> ext3 or reiserfs (with notail option) file system.
>>
>> - Set up ASYNCIO + DIERCT IO:
>>
>>  alter system set FILESYSTEMIO_OPTIONS=setall SCOPE=SPFILE
>>
>>  (and restart the database). You can configure it on initial database 
>> setup.
>>
>> instead.
>>
>> - Run few transactions to create archive logs (I did not tested with the
>> small logs; I tested after the full HAMMERORA/TPC-C paths; I suspect that

>> it
>> doesn't matter at all, but can't confirm it).
>>
>> - Now, run rman. As Oracle, do:
>>
>>    export ORACLE_SID=<db_sid>
>>    rman (remove /usr/X11R6/bin/rman to prevent shell confusion).
>>    connect target
>>    backup validate archivelog all;
>>
>> DONE. You have read error.
>>
>> Now change the kernel to the kernel 287 (rpm -U -oldversion ...., or any
>> other way).
>> Try rman again. Now it works.
>>
>> Another experiments show, that not only rman is affected, but log
shipping
>> (in oracle) is affected. The only difference is that log shipping works 
>> well
>>
>> if filesystemio_options=ASYNCH but doesn't work with SETALL, while rman
>> doesn't work with both SETALL and ASYNCH, modes.


_______________________________________________
suse-oracle mailing list
[email protected]
http://listx.novell.com/mailman/listinfo/suse-oracle


_______________________________________________
suse-oracle mailing list
[email protected]
http://listx.novell.com/mailman/listinfo/suse-oracle