HP-UX redo write performance

[former_member624711]

Hi,

We have a number of SAP DB instances on HPUX (oracle 11g2) using JFS (cooked filesystems rather than ASM). Filesystems are mounted with CIO (concurrent IO), and JFS does not support sync IO (so its sync io with cio support at the OS/FS).

REDO is written to only a single group member (ie not mirrored logs)

Redo write time are high, even thought back end SAN is caching all wites (writes are sync written to the DR storage from the local array as well).

What I an looking for is recommendations on how to configure the REDO devices (ie array based stripping, host based stripping of multiple storage devices).

As REDO is a single treaded (ie LGWR) and does siquentiual IO I am not sure what benefit can be obtained by stipping of the LUNs underlying the REDO filesystems (using /oracle/<SID>/origlogA and origlogB).

Advise gratefully accepted.

Regards

Glenn

[Former Member]

Hello Glenn,

the SAP note 1077887 was already mentioned. There is an excellent whitepaper available here,
http://h20195.www2.hp.com/V2/getdocument.aspx?docname=4AA1-9839ENW.pdf which explains why certain configuration parameters should be met or parameters be set.

You mentioned redo write times are high. What are the values for "log file sync" and "log file parallel write"? Is "log file sync" a dominant wait state? If so, you should use glance and check what the ora_lgwr process is waiting for.

Regards,

Jan Schreiber

[former_member188883]

Hi Glenn,

For HPUX please refer SAP note Note 1077887 - SAP on HP-UX:mount & filesystem options for best performance.

This has some good guidelines pertaining to performance.

Regards,

Deepak Kori

[jairo_pedroza]

Hello Glenn,

I have SAP System on Oracle with 19TB. We're using config as mentioned by Deepak in SAP Note.

Oracle

For sapdata's use 8kb blockize

For redologs (origlog and mirrlog) use 1kb blocksize.

We're using DIO !

Our "log file sync" response time is 7ms

Regards,

Jairo Pedroza