Number of Pages freed per second is high

[David_246]

Hi,

I am monitoring my system (RP5430 single cpu / 3Gb mem) and I find that the Number of freed pages is extreem high now and than. Looking at the guidlines this value should stay under 10 p/sec. Sometimes this value passes the 500. Looking at the guidelines again this could indicate a shortage of memory, however looking at "swapinfo -tam" the system is not using all memory at all (3Gb) :
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 3072 0 3072 0% 0 - 1 /dev/vg00/lvol2
reserve - 656 -656
memory 2411 591 1820 25%
total 5483 1247 4236 23% - 0 -

Does anyone have a clue where I should find a solution/explenation for this issue.

Regards David

[G. Vrijhoeven]

Hi David,

If fr is high the process vhand must use lots of cpu load. You are having memory problems.

What you can do is check the number og signle bit errors in memory (stm), check if you buffer cache settings are. I heard a it is better to have dynamic buffer cache ( max_dbc_pct/and min_dbc_pct in stead of maxbufpages ( guess i am not sure about this parameter.
Check your memory usage with glance to get a better insight.
And at last increase memory

HTH,

Gideon

[David_246]

Hi Gideon,

Thanks for your reply. dbc_max_pct is set to 10 and dbc_min_pct is set to 2. Both values are static (currently).
Unfortunetly I do not have glance available (I use BMC Patrol). However 3Gb of memory should realy be sufficient. We do not have any memory consuming programms running on this server, it is a fallback node (MC Serviceguard) for an Oracle package, but currently not running on this node.

Do you have any explenation or understatement for your recommendation?

Regs David

[G. Vrijhoeven]

David,

The buffercache setup looks fine to me. Vhand is a process that swaps out memory pages that are not used since the last time vhand touched the page. If there is a shortage of memory vhands starts using CPU and touches the pages faster.
If vhand is not active it can be so that >10 number of pages ( not touched by a process ) get swaped out.

Gideon

[Jeff Schussele · Accepted solution]

Hi David,

I wouldn't worry about page frees, but rather page outs.

Frees could simply be processes ending normally & returning it's memory pgs to the free stack.
If you have a lot of work going on & it's short-lived stuff then you'll see a lot of frees as a normal result.

Run vmstat -s or a vmstat 10 10 & check the page out (po) values. If you see none or just a few then you're OK memory-wise.

Rgds,
Jeff

[Tim Sanko]

If you have a bunch of small processes terminating quickly, you may see this.

I wouldn't worry about anything except the timeslice tuning parameter, at this time. as lond as the system seems to perform adequately.

[David_246]

Hi All,

Thanks a lot for all your help. The pages freed happens right after my system reaches a load of 15 . Page outs are not there at that time, just a lot of user and system cpu usage.
The only weard thing was the high amount of pages freed. It's clear to me now (thanks Jeff) that it's just the amount of processes finishing which result in the high amount.

Again thanks a lot for your help.

Regs David