Performance Problem in HP-UX 11.31 with Oracle 10 g R2

[Victor Burguillos_4]

Hi everyone,
I have a problem please anybody help me, I 've oracle enviroment with HPUX 11,31 but in glacen show me the memory status ever in 99%.

Anybody help me

Regards

Vic

[Tim Nelson]

For starters please provide the following.

amount of physical RAM.

output of swapinfo -tam

amount of SGA configured

output of ipcs -ma

[Victor Burguillos_4]

Hi Tim.

Physical RAM: 32 Gb

mls01:/> swapinfo -tam
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 32768 784 31984 2% 0 - 1 /dev/vg00/lvol2
reserve - 25518 -25518
memory 31040 6500 24540 21%
total 63808 32802 31006 51% - 0 -
mls01:/>
mls01:/> ipcs -ma
IPC status from /dev/kmem as of Thu Aug 14 10:26:01 2008
T ID KEY MODE OWNER GROUP CREATOR CGROUP NAT
Shared Memory:
m 0 0x411c07c0 --rw-rw-rw- root root root root 9
m 1 0x4e0c0002 --rw-rw-rw- root root root root 9
m 2 0x41202f91 --rw-rw-rw- root root root root 9
m 3 0x411c040f --rw------- root root root root 4
m 11960324 0x00000000 --rw-rw---- oracle10 dba oracle10 dba 4
m 65541 0x00000000 --rw-rw---- oracle10 dba oracle10 dba 4
m 65542 0x00000000 --rw-rw---- oracle10 dba oracle10 dba 4
m 65543 0x285402ac --rw-rw---- oracle10 dba oracle10 dba 4
m 425992 0x00000000 D-rw------- tronweb paquetes tronweb paquetes 8
m 196617 0x00000000 D-rw------- tronweb paquetes tronweb paquetes 8
m 30834698 0x0c6629c9 --rw-r----- root sys root sys 5
m 262155 0x06347849 --rw-rw-rw- root sys root sys 0
m 229388 0xffffffff --rw-r--r-- root sys root sys 6

[Victor Burguillos_4]

Hi Tim,

I attach the glance screem output.

Regards

VB

[Tim Nelson]

The ipcs -ma is cut off in the post and the attached.

Need the column for SEGSZ. Looking to see how large your Oracle SGA is.

Can you fix the number of columns in your emulator and repost as an attachment.

Anyway. The glance helps.

There is only 443MB of free mem

6GB of kernel
15GB of user
9GB of buffer cache.

First item. If this is an oracle db system then reduce the buffer cache way down.

Reduce to hard numbers maybe 500-800MB or maybe 2%

kctune filecache_max
kctune filecache_min

This will give you about 8-9GB back.

[Tim Nelson]

Also.. With only 32GB of device swap you MAY start getting cannot fork messages soon.

Gotta have a little buffer especially if you are riding on the edge. I would suggest at least another 4GB dev area just to carry the slop. Obviously you'd be swapping/deactivating.

If you choose to reduce the file_cache then you will be back to being OK. I would still allocate the little extra dev swap.

[Victor Burguillos_4]

Attach the complete output

[Tim Nelson]

With that I see about 6.6GB for Oracle shared mem segments.

So.

1) reduce your filesystem cache

2) need to find the other 8GB or so of user mem usage, maybe an application. Can use glance process list and sort by RSS.

3) create at least 4GB of additional device swap to stop the impending outage for now.

[Chitta]

Hi Victor,
Are you facing any performance issue while the memory utilization goes 99%?
checkthe value of "kctune -q dbc_max_pct" reduce to 10 or 20.Your swapinfo results shows the phhysical memory usage 21%, no issues.
also check taht there are a few performance patches you
should install in your system if applicable.


PHKL_37653 s700_800 11.23 mmap(2),mlock[all](2),madvise(2),msync(2)

PHKL_36669 s700_800 11.23 mlock performance fix

PHKL_37000 s700_800 11.23 Dynamic Buffer Cache patch

just forget about the Glance output.

Chitta.

[Hein van den Heuvel]

Victor,

What you show does NOT suggest a performance problem. There may be one, but it is not clear from the data provided.

Please help me understand why you gave this topic the subject 'performance problem'.

What you have is a number at 99% which apparently is not understood.

HPUX is DESIGNED to use all the memory.
It does so by that reasoning you should be happy!

HPUX uses all excess memory for fiel cache... it has nothign better to do with it, and thus the total memory consumption shoudl be expected to be 99%.

Now you can lower percentage this by restricting the file system cache... and the number will change, but the performance is unlikely to get better (on 11.31) and the performance may well get worse.

A better plan MIGHT be to use the memory 'better' by giving more to ORACLE.

Of course that will only be usefull if Oracle can use more. So go ask Oracle whether it likes more memory, or whether you can leave well-enough alone

Check the STATSPACK / AWR data, and so on.

Best regards,
Hein van den Heuvel
HvdH Performance Consulting.