SWAP Problem

[George Nikoloudis_1]

Dear All

I have a strange problem withthe swap. It is 93% full. Below is the output of swapinfo -tam:

Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 1024 0 1024 0% 0 - 1 /dev/vg00/lvol2
dev 4096 110 3986 3% 0 - 0 /dev/vg00/lvswap
dev 2048 108 1940 5% 0 - 0 /dev/vg00/lv_swap2
reserve - 6397 -6397
total 7168 6615 553 92% - 0 -
The server has 6 GB of Memory, 6 CPUs and is a V Class.
Can you help me in order to inverstigate this case, I can provide you with any other information you required.

Regards
George

[harry d brown jr]

What are your

dbc_max_pct & dbc_min_pct set to??


live free or die
harry

[Helen French]

Hi George:

One possible try is to check the dbc_max_pct kernel parameter value. Check it with:

# sysdef
# kmtune

If it is set to default 50%, then reduce it to a proper value ( 15% ?)

HTH,
Shiju

[Rita C Workman · Accepted solution]

The fact that the total is showing 92% means that the majority of processes are being run in memory.....and that is where you want things to run - in memory. That 92% does not mean that it swapped/paged out to disk 92%, since that 92% is cumulative of ALL swapping going on which includes both in memory and on physical devices.....(but your swapinfo does show that you ARE paging out to disk alittle bit at 5-8%..)

You have 6gb of mem. So as was mentioned, make sure that those two parms are not set too high.
Since the majority of your processing appears to be occuring in memory I shall presume that you have your bufpage and nbuff set to 0 to have the system dynamically control this...and that you have enabled swapmem.


Just a thought,
Rita

[A. Clay Stephenson]

Hi:

Note that only 8% is actually swapped out. The really big stuff is 'reserve' - processes that are running but are not paging out. You should do a vmstat and look at the pageout (po) rates. If that number is high then you need to add more memory or reduce the number of processes running or reduce the size of some kernel resources - like buffer cache. If you are lucky, you have some shared memory segments that are unused and that is where your memory is.

Ideally, you would like that pageout rate to be extremely low (zero).

[George Nikoloudis_1]

the Kernel Parameters are

dbc_max_pct 6
dbc_min_pct 6

[George Nikoloudis_1]

After Running the vmstat the result is the the pi/po is 0 for both.

[harry d brown jr]

george,

Then your machine is behaving very well.

live free or die
harry

[Rita C Workman]

If your getting 0 on pi/po than that hit/swap to disk was probably a short-term peak. If you see it a little too often, look at making adjustments as Clay mentioned.
Or possible it's one job (you could monitor to see), and you could maybe schedule it to run at an off-peak time to avoid impacting the system and swapping out.

Just a couple last thoughts,
Rita