HP-UX Memory and swap issue

[gab_in]

Hi,

We have a HP-UX Itanium system running 11.31 and Oracle 10 version. We are running some application in that and it is giving the below warning message

Warning: Your system is swapping out processes. Your system is overloaded. Your system may have started thrashing. Total number of jobs that are swapped out are <2>.
But our systems swapinfo output says that it has 32.GB swap space and 8GB Physical RAM. Please find below the swapinfo -m output .

Please suggest how do we go about it….


./swapinfo -m
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 32768 659 32109 2% 0 - 1 /dev/vg00/lvol2
reserve - 8158 -8158
memory 7763 1413 6350 18%


Thanks

[Dennis Handly]

>Please find below the swapinfo -m output.

We need "swapinfo -tam".
Since your memory is only 18% used, perhaps you don't have the problem right now? You need to use swapinfo when you see that Warning.

Do you know what tool is producing that warning?

[Sajjad Sahir]

please give the following output
then only i can give u completely details

swapinfo -tam out put


see the commands difference i want to see what is u total reserver under total column
so please send it swapinfo -tam output

data1#swapinfo -m
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 16000 115 8077 1% 0 - 1 /dev/vg00/lvol2
reserve - 3074 -3074
memory 6448 6448 0 100%
data1#swapinfo -tam
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 16000 115 8077 1% 0 - 1 /dev/vg00/lvol2
reserve - 3074 -3074
memory 6448 6448 0 100%
total 22448 9637 5003 43% - 0 -

[gab_in]

Thanks for the response. Please find below the output.
# swapinfo -tam
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 32768 3641 29127 11% 0 - 1 /dev/vg00/lvol2
reserve - 6436 -6436
memory 7763 1450 6313 19%
total 40531 11527 29004 28% - 0 -
#

[Sajjad Sahir]

Dear
one more thing
what about u physical memory
quckly revert
dmesg | grep phys

sajjad

[Steven E. Protter · Accepted solution]

Shalom,

1) If your system is swapping out at all it IS overloaded. Swapping keeps it running under stress but is not a good day to day situation.
2) You may wish to have your dba's reduce the oracle SGA requirements.
3) Your RAM to swap ratio is off. The maximum recommended amount of swap on a 8 GB physical RAM system is twice RAM or 16 GB. Your current configuration is inherently bad for performance.

Since your system is under stress and Oracle DBA's are well know for not giving up a byte of SGA, I'd suggest upgrading your system to 16 GB of ram is the quick solution. If you do that, you won't even need to add swap, you already did that some time in the past.

:-)

SEP

[gab_in]

# dmesg | grep Phys
Class Physmem Lockmem Swapmem
Physical: 8358468 Kbytes, lockable: 5937452 Kbytes, available: 6766224 Kbytes
#

Thanks

[Don Morris_1]

Well, swapinfo is just swap.. but it doesn't look at first glance (no pun intended) as if there's system wide memory pressure that should be causing swap outs. (Since usually if there's that much consumption, it eats into memory swap as well).

How about we just ask the kernel -- what does "vmstat -S 30" give as output, particularly the si, so, pi, po and fr columns? [so vs. po is interesting -- the application indicates that so should be non-zero (processes swapped out), po being non-zero is entirely normal especially on v3 -- the file cache works by paging out unneeded file pages when the cache is filling up, after all... I'm wondering if the application writer didn't understand the difference between po and so here]. Also for quick summary, post "vmstat -s" output.

The only other possibility that comes to mind is if you're using PRM (or WLM which uses PRM underneath as I understand it) and have the memory resources parititioned... that could allow for the system not being under pressure but the resource partition for a set of processes being under local pressure.

[Sajjad Sahir]

u swap configuration is correct becuase u physical memory is 7.971 GB also u swap space configured is also ok, u process running load is high thats why this message u recieving so u have to consult wiht application team in unix side u configure very well
For a process
to spawn it needs a sufficient amount of virtual memory to be placed in
reserve. There should be a sufficient amount of free device swap to open any
processes that may be spawned during the course of operations.

[Don Morris_1]

Well, that's certainly true -- but I don't see how it is germane to the problem in question. The submitter isn't getting fork() or malloc() failures due to ENOMEM, an application is claiming processes are being swapped out. Different issue, and the swapinfo output posted so far does not show a lack of swap reservation resources.

(And please for all our sanity -- the word is "you". I don't see where you're gaining any bandwidth or speed by saving the 'y' and 'o' characters there.)

[Don Morris_1]

Re-read the thread, and SEP's comment stood out this time (and that the device swap consumption jumped 9% between your initial post and the follow up, not a good sign).

I just missed the good old simple case -- you've got much more disk than RAM and pseudo-swap isn't needed for anything but the locked memory (which is probably just the kernel in your case). Yes, SEP is right [like this is a surprise, I know] -- you're just oversubscribed here -- by at least 4Gb judging by your later output.

Since that's 50% above your physical capacity, I seriously doubt trying to tweak filecache_min/filecache_max or other system settings is going to help -- you just simply need more RAM or to reduce the memory load. (Or get used to really lousy performance).

[Dennis Handly]

>Don: the word is "you".

Should we also be complaining that there are three different proper words:
you, your and you're (for you are)