High page out in vmstat

[Lin Fei]

Hi,

We have an Oracle8i database runs on a 24 CPUs and 32G HP server. There is 12G free memory. Why the po and fr numbers are so high from vmstat output? Below is the sample vmstat output:
page
pi po fr
1 19 35
0 9 0
0 16 0
0 5 37
0 15 11

[Pete Randall]

How are you determining that there is 12G of free memory? The fact that you're paging out indicates that you're under memory pressure which should indicate that you don't have any free memory.


Pete

[Patrick Wallek]

What the 'swapinfo -tam' indicate?

[RAC_1]

Can you the output of swapinfo -mat
Also look at "glance -w"

What it says?? What makes you think that you ahve 12gb free?? Check glance -m

Anil

[A. Clay Stephenson]

How do you know that you have 12GB free? What command are you using to determine this?
The vmstat data are not all that useful unless we also know the sampling interval. The Glance memory statistics would provide much more insight and swapinfo -t output.

[Xavier Gutierrez_2]

Hi, Lin.

If the box is swapping out, then you have run out of physical memory for sure.

Probably you're watching at the SWAP+PHYSICAL free memory instead of the free physical memory which should be quite low...

The default HPUX installations take as swap at least as much memory you have... in such a case you would be running low on swap also.

I would suspect Oracle initialization parameters that are likely using up all of the system memory when numerous Oracle parallel processes are started:

hash_area_size
sort_area_size

And so forth...

Cheers,

Javier.

[Steven E. Protter · Accepted solution]

If you really had 12 GB of free memory you would not be paging.

swap gets reserved when you start the process.

swap doesn't get used(paged) until there is not enough physical memory for all processes and inactive processes are paged from memory to disk.

This is a simplistic view because the entire process is not necessarily swapped.

Attachinig some performance monitoring scripts that might help.

SEP

[Lin Fei]

I don't have root permission to run swapinfo. And I don't have glance either. But I attached sar and vmstat info here. From vmstat, free=3155240, so the free memory is 3155240*4K/1024/1024=12G. That's how I know there is 12G free memory.

[Pete Randall]

If I interpret that correctly, you have 3155240 pages on the free list. This does not mean physical memory, it means virtual memory which could be either physical or swap space. I suspect that you have exhausted your physical memory and probably a significant amount of swap space as well.

The "po" column in vmstat is a pretty solid indication of swapping.


Pete

[Lin Fei]

Pete,

Thanks for your script. I ran the script. I don't have permission for swlist and ioscan. Attached is the zip file for HP, HP performance info.

All,

Thanks for the quick reponses. Please exam the attached files and let me know if we did ran out of swap space.

Thanks,
Lin

[Pete Randall]

Just a thought:

Since you don't have root authority to run the commands that would help us answer your questions, why don't you ask your sysadmin?


Pete

[Lin Fei]

Attached is the swapinfo info. I saw the other our HP-UX 11.0 hp servers have same behavior. Could this high po due to the 11.0 operating system?

[A. Clay Stephenson]

Memory-mapped files can cause significant pageouts even with lots of memory free -- and this is not a bad thing. You need to ask your application developers if memory-mapped files are in use.

[Bill Hassell]

Your page out rate is very low. Single digits can be ignored, double digits are of interest but not if they don't last. These low numbers in such a large computer can be ignored. However, I would STRONGLY recommend getting a copy of GlancePlus as soon as possible. With a machine this large (that cost a lot of money), not having Glance will cripple your ability to monitor performance issues.