Itanium vs PA-Risc for Oracle Databases

[Chris Fadrowski]

We recently moved all of our 11 Oracle DB's from a v2500 PA-Risc box running 11.0 hpux to a RX4640 Itanium server running 11.23v2.

we have 16 gb of memory on both boxes, and we did NOT change any SGA's just ftp'd copies of the databases over to the itanium box. However, memory on the Itanium is at 99% and it was less then 60% on the V2500 server.

Can someone tell me if there is a difference on how the Itanium boxes or 11.23v2 allocates memory? I used Oracles documentation for kernel parameters but is there another one i should be looking at?

[Chris Fadrowski]

One more thing,

Buffer Cache is at 8gb on this system. dbc_max_pct is at the default of 50 for it's value

[Eric Antunes]

Hi Chris,

What is the RDBMS version? I'm asking because only Oracle 9.2 64-bit is certified on Itanium 11.23 and you should patch it to 9.2.0.6...

Best Regards,

Eric

[Chris Fadrowski]

9.2.0.6 Oracle. ONly version 9 runs on itanium. (but i am no DBA)

[Eric Antunes · Accepted solution]

Since Oracle has a buffer cache, you should reduce the OS buffer cache, thus avoiding double caching: your OS buffer cache should be something between 300-400Mb, so reduce dbc_max_pct and dbc_min_pct to, respectively, 3 (3%*16Gb=3/100*16384Mb=492Mb) and 2.

Furthermore, check recomendations for other significant kernel parameters when running Oracle in anexed doc...

Best Regards,

Eric

[Patrick Wallek]

The default buffer cache is absolutely ludicrous. Reduce at the first opportunity. On a 16GB system, a max value of between 3 and 5 % should be sufficient. I might consider dbc_min at 3 and dbc_max at 5.

[Eric Antunes]

Hi,

For more kernel parameters (since I'm not able to attach docs. witch more than 1Mb), see this thread:

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=915316

Best Regards,

Eric

[Jean-Luc Oudart]

Chris

I believe you just hit bug 3890965 "ORACLE CONSUMES LARGER MEMORY IN HP-UX ITANIUM PORT"

Regards
Jean-Luc

[Jean-Luc Oudart]

Hi

You can tune your memory as per other post., but I don't think the "bug" is corrected in 9.2.0.6. Check Metalink (or ask your DBA to do it for you).

From this :
[output of glance] - Itanium - Text RSS/VSS:112mb/112mb Data RSS/VSS:5.0mb/6.0mb Stack RSS/VSS:384kb/256mb Shmem RSS/VSS:229mb/242mb Other RSS/VSS:7.8mb/ 24mb - PA-RISC - Text RSS/VSS: 64mb/ 64mb Data RSS/VSS:2.0mb/2.0mb Stack RSS/VSS:128kb/192kb Shmem RSS/VSS:226mb/242mb Other RSS/VSS:4.4mb/9.0mb .
"Text Segment is about double the size but most of them are shared among each server processes so it is not a problem. . The problem is that Data Segment differs about 4MB. Data Segments are not shared among the server processes so this memory adds up when there are many connections. "

Regards
Jean-Luc

[Duncan Edmonstone]

Yup, Oracle on IA64 is one greedy sonovab****. Take a look at the release notes:

http://download-uk.oracle.com/docs/pdf/B10566_05.pdf

Section 6.1, page 21


Also see the discussion here. When I size customers moving from PA-RISC to IA64, I generally add 50% more memory for Oracle systems:

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=814468

HTH

Duncan

[Yogeeraj_1]

hi chris!

may i also suggest that you log an iTar with Oracle on Metalink.

It is true that there is a bug that they have detected. Oracle Database version 10.1.0.2 is also affect by this. but this was reported in january.

You may have found the resolution in the meanwhile.

hope this helps too!

kind regards
yogeeraj

[Zygmunt Krawczyk]

Hi Chris,

if you have configured enaugh swap space, at least 16GB, you can change the kernel parameter swapmem_on to 0 (default is 1). It saves the memory.

What is the date of your OS? If you have OS date prior to Sept 2004, you should upgrade your OS to Sept 2004.

Regards,
Zygmunt

[Chris Fadrowski]

i believe if i just lower the kernel parameter dbc_max_pct from 50 to 20, it will give some of the memory back. As others said, oracle has it's own buffer cache.

[Jean-Luc Oudart]

Hi

20% 16GB => 3.2GB I think this is still too high.
On HPUX11.0 I run between 300 and 400Mb.
I believe on 11i you should set to 800-1000Mb

Regards
Jean-Luc

[Ted Buis]

If you have OnLineJFS, you can eliminate the double buffering done by Oracle and HP-UX. Then there will be even less need for dbc maximum to be set so high. Also, if you are seeing pageouts, then it may be good to turn pseudoswap off if you have enough device swap. However, if you are not doing any page outs then you might as well keep pseudoswap turned on. It really doesn't consume memory, but only locks pages into memory, which elimitates the possibility of those pages being swapped out, thus eliminating the need to have any device swap space reserved for those pages. It does not take some of your memory and turn it into a swap area. It does effectively increase the virtual memory space to the sum of the locked pages plus the device swap space.

[Zygmunt Krawczyk]

In addition to Ted's response, read the doc:

"Best Practices for Oracle on HPUX"
http://h21007.www2.hp.com/dspp/files/unprotected/database/HP3KOracle.ppt

[Tim D Fulford]

Hi

I think every one has done stuff on Oracle.. you should also know HP-UX 11i v1 & v2 uses more memory than HP-UX 11.0.. You'll generally see a large increase in the system memory utilisation from a couple of 100 MB to GB's.

That said, HP say that this is not quite as bad as it may seem although system memory as a % has increased, User memory has decreased... such that there is 10-20% increase in total memory usage..

The above is independant of Oracle and its workings...

Tim