Swapinfo High (100%)

[Akif_1]

Hi All,
On my production server rp7410 with OS 11i last week i upgraded memory from 5G to 8G x4, But my problem is that on one of the node(DB1) out of four, shows me swapinfo percentage were 100. where as other 3 where normal,
What i noticed is that i usuallay use DB1 for export a file of 5GB. After completion of this job the swapinfo were not reducing.
What were the solution for this


Thanks in Advance

[sajeer_2]

What is your swap size?Did you increase swap space after memory upgrade?.
Add a secondery swap and see the performance.

[Michael Steele_2]

You have 8GB of RAM in DB1. So for DB1 attach:

swapinfo -tam

vmstat 4 4

And check for memory leaks:

# UNIX95= ps -ef -o pid,ppid,vsz,sz,cpu,args|sort -k3nr|more

[Akif_1]

Dear Michel n Sajeer,
The output of above were as follows and bcz of small space in our drives we are unable to create new swap, as we have old swap which works before on 5GB..... Please follow attach file for more details.....



Cheez:)

[Steven E. Protter]

Shalom,

Due to the format of the attachment it is hard to read it and interpret the results, at least for me.

swapinfo -tam would be helpful.

It would appear however that your system is under memory stress from Oracle and java.

The real solution to this problem is to lower memory use, perhaps by adjusting the buffer or the oracle SGA or add memory to the system.

Too much swap will not make the system faster.

http://www.hpux.ws/system.perf.sh

SEP

[Michael Steele_2]

Steve, save it to disk and open it with word.

This all looks normal to me. Is DB1 at the same patch level as the other four nodes? Same with glance if that's what you're using.

Can you paste in:

ipcs -moba

[Akif_1]

Hi Steel,


Please follow an attach file of ipcs -moba, and dbc_max_pct 50 and dbc_max_pct 5, as on upgraded 8GM of RAM,,,,


Thanks in Advance

[Steven E. Protter]

Shalom,

Oracle is better at buferring its own data than the OS.

You can significantly decrease your memory use by changing dbc_max_pct to a lower number like 5 6 or 7.

This will reduce memory pressure by hunreds of megabytes.

You should also check with the DBA's and see if they can make memory consumption reduction in the oracle SGA's.

SEP

[Akif_1]

Hi Stee,
Sorry for mistake my dbc_max_pct 50 and dbc_min_pct 5, So is this improve my server performance, eventhough SGA max size is 2432M , with 4.5(Approx) of export data. and EVA has its own cache batteries on the controller.


Help me to release this 100% high memory in one of the 4 servers ... i.e:DB1.


Cheez:)

[Bill Hassell]

Well, the first change is to reduce dbc_max_pct from 50 to 3. That will return almost half of your memory. The default setting (50) is just plain bad for any sized system.

Also, in your previous attachment with a ps listing, the command was:

ps -ef pid,ppid,vsz,sz,cpu,args|sort -k3nr|more

which does not show memory usage and should have been:

UNIX95=1 ps -ef -o pid,ppid,vsz,sz,cpu,args|sort -k3nr|more

If you used the -o option without UNIX95=1 then ps will give you an error. To make memory usage easier to read:

UNIX95=1 ps -ef -o vsz,pid,ppid,sz,cpu,args|sort -k3nr|more

which put VSZ first and the largest processes will sort to the top. Just a note that glance (an priced product from HP) is invaluable in understanding and managing system resources in HP-UX.