Performance issue in hp-ux 11.23

[Sawant]

Hi Experts,

Our memory utiliation on production server running oracle 10g remains 100% most of the time hence affecting production hours.

Please find attached the kmeminfo details below:
Physical memory usage summary (in page/byte/percent):

Physical memory = 4193792 16.0g 100%
Free memory = 35711 139.5m 1%
User processes = 3310190 12.6g 79% details with -user
System = 837925 3.2g 20%
Kernel = 512446 2.0g 12% kernel text and data
Dynamic Arenas = 234942 917.7m 6% details with -arena
mdep arena = 57432 224.3m 1%
VFD_BT_NODE = 19114 74.7m 0%
vx_buffer_cache = 18784 73.4m 0%
vx_global_pool = 16774 65.5m 0%
spinlock = 15356 60.0m 0%
Other arenas = 107482 419.9m 3% details with -arena
Super page pool = 28851 112.7m 1% details with -kas
Static Tables = 264639 1.0g 6% details with -static
pfdat = 98291 383.9m 2%
nbuf = 90064 351.8m 2% bufcache headers
htbl2_0 = 32768 128.0m 1%
pfn_to_virt = 16381 64.0m 0%
bufhash = 8192 32.0m 0% bufcache hash headers
Other tables = 18941 74.0m 0% details with -static
Buffer cache = 325479 1.2g 8% details with -bufcache

Please check & revert soon.

[Duncan Edmonstone]

So do you actually _have_ a performance issue, or are you just thinking you do because yu are using all your memory (using all your memory might be a good thing for certain systems which don't change their usage patterns over time). Are you getting a lot of page-outs? (you can check with vmstat)?

Looking at the details you post, you seem to have 16GB of memory with 20% used for system purposes and 80% used by user processes (presumably mostly Oracle). So where do you spend your time diagnosing this... looking at the 20% that the system uses or the 80% that Oracle uses...

For the system usage, pretty much your only reasonable point of control might be to reduce the size of the buffer cache further (looks like you have dbc_max_pct currently set to about 8% which reults in a buffer cache of up to 1.2GB - I guess you could tune this down further, but its difficult to know the impact of this given you have told us nothing about what actually happens on this system).

On the Oracle side, you should talk to the DBAs about the size of the SGA and about controlling the size of Oracle shadow processes using PGA_AGGREGATE_TARGET

HTH

Duncan

[Laurent Menase]

3 or 4 things:
1) kmeminfo is a hp confidencial tool, so should only run on hp representative request.
- like indicated in the begining of the output.
the result should be interpreted by hp representative only.
2) you have a memory problem indeed and your system is probably swapping causing high perf problem.
2) kernel and buffer cache have a reasonable use
3) user process are using 80% of the memory,
---> you probably need to reduce oracle cache or add more memory to leave a little memory to the kernel

Also it depends on the number of cpu, number of 10Gether interfaces, number of io interfaces, number of oracle instances ......

you can reduce buffercaches but this may cause some other perf problems for other programs than oracle if you are using syncios.

The kernel need memory to be able to feed all the oracle process with io usually

[BDD]

Hi

Try to check following things

1. check option to reduce
Buf Cache:
Free Mem:
2. Check with DBA to reduce RAM used by SGA


BR
BDD

[Sawant]

Thanks all for your response.