Memory upgrade - No improvement

[Akif_1]

Hi All experts,

OS:HP-UX 11.11
DB :10g R2
SGA 3.5GB (8GB RAM).\
shmmax : changed from 8 to 16GB
dbc_max_pct=35


The total SGA was 3.5 when 8GB ram , now i upgraded RAM from 8 to 16 , SGA still 3.5, But the swapinfo -tam 82/52% where as previous values are 98/52% just 10% difference.

My surprise is when i double the memory , The status should be much lower than previous value.


*Does it require any other changes
*It require any patches


Thanks & regards


I change

[Dennis Handly]

No improvement in what, SGA?
Please provide the full "swapinfo -tam" output.
Also, dbc_max_pct=35 may be still too big.

[Bill Hassell]

Your numbers aren't meaningful. It appears that you have lots of processes occupying memory. HP-UX is doing exactly what it is supposed to do -- allocate memory for processes that request it. A much more useful measurement is to sort the processes by memory and look at shared memory:

UNIX95=1 ps -e -ovsz,pid,ppid,args | sort -rn | head -20

and shared memory usage:

ipcs -bmop

[Akif_1]

Hi All experts,

I would like to remind you this is a RAC database based on HP-UX 11.11, Oracle database 10g R2.

The new update is :
*RAM updated from 8 to 16GB each node.
*SGA increased from 3.5GB to 10GB
*enclosed output UNIX95=1 ps -e -ovsz,pid,ppid,args | sort -rn | head -20

* enclosed ipcs -bmop output to file"present values

Kindly update me with your valuable input by looking present values enclosed file and below previous values details of swapinfo -tam and present values of ipcs,UNIX95 process.


previous values:
================

SGA=3.5GB RAM 8+8GB

DBRAC1:/#swapinfo -tam
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 4096 63 4033 2% 0 - 1 /dev/vg00/lvol2
dev 4072 60 4012 1% 0 - 1 /dev/vg00/lvswap
dev 4092 59 4033 1% 0 - 1 /dev/vgswapdb1_1/lvswapdb1_1
dev 4092 63 4029 2% 0 - 1 /dev/vgswapdb1_2/lvswapdb1_2
reserve - 11843 -11843
memory 12721 12721 0 98%
total 29073 24809 4264 50% - 0 -






DBRAC2:/#swapinfo -tam
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 4096 0 4096 0% 0 - 1 /dev/vg00/lvol2
dev 4072 0 4072 0% 0 - 1 /dev/vg00/lvswap
dev 4092 0 4092 0% 0 - 1 /dev/vgswapdb2_1/lvswapdb2_1
dev 4092 0 4092 0% 0 - 1 /dev/vgswapdb2_2/lvswapdb2_2
reserve - 11939 -11939
memory 12721 10774 1947 77%
total 29073 22713 6360 45% - 0 -




Thanks & regards

[Bill Hassell]

> The new update is :
> *RAM updated from 8 to 16GB each node.
> *SGA increased from 3.5GB to 10GB

These changes would make no significant difference in the percentage of RAM. You added 8GB, but you increased SGA by 6.5 GB...so your free RAM increased by just 1.5 GB (8-6.5=1.5) -- not very much. Your sorted processes by VSZ shows at least 20 processes using 123 MB each. Change the head -20 to head -60 and attach the results. Everything looks normal but you need to start counting the number of big processes and decide if the DBA and application admins are trying to overload your machine. If they want to run a lot of big programs, add another 8 GB of RAM (but leave SGA the same size), and definitely reduce dbc_mac_pct to 15%. 35% of 16 GB is 5.6 GB, about twice as large as can be effectively used on HP-UX 11.11.

[Akif_1]

Thanks bill for your valuable input.

Iam planning to reduce SGA from 10 to 8GB and dbc_max_pct=30.

Kindly check attached file for your review and feedback.

Definitely the db performance is crunch after memory upgrade and SGA allocation, need to fine-tune for much better performance than before.

Regard's

[Bill Hassell]

> Iam planning to reduce SGA from 10 to 8GB and dbc_max_pct=30.

These changes are very small -- I would not expect any differences. Leave SGA=8GB and change dbc_max_pct=15. Changing from 35 to 30 won't make any difference.

From your attachment, there are a *LOT* more processes that are not shown. Remove the head -60 and just list everything.

Now when the system is busy, run vmstat and report what you see. You are looking for the po (page out) rate -- 0 is good more than 50 is bad.

[Jean-Luc Oudart]

Hi

Before you make changes you should take a baseline of current performances. This should be done ast OS level and also at Oracle level.

CF Links for Oracle RAC monitoring / tuning :

http://download.oracle.com/docs/cd/B28359_01/rac.111/b28252/racmon2.htm
http://download.oracle.com/docs/cd/E18283_01/rac.112/e16795/monitor.htm
http://download.oracle.com/docs/cd/B19306_01/rac.102/b14197/monitor.htm

I always think that throwing hw at a problem does not fix it but rather hide it.

I heard sometimes it got cheaper to buy hw than analyzing the issues, but not sure it pays off in the long term.

Regards
Jean-Luc