Originally posted by: SystemAdmin

Hello.

From some time we see very big numbers in vmstat output.
We know that we have some memory shortage and that's the reason of paging, but THAT big numbers can't be realistic

$ oslevel -s
6100-07-03-1207
$ vmstat 10 3

System configuration: lcpu=20 mem=40959MB

kthr memory page faults cpu

---

-----------

---

------------

---

r b avm fre re pi po fr sr cy in sy cs us sy id wa
4 1 9051050 19075 0 4611909543914966196 4630331938342107451 4652579111602582656 4662926806065236488 0 3027 192186 13634 15 17 67 1
6 1 9048320 24410 0 4617879396921796497 4633825319621027728 4654159360071464109 4658225288420287975 0 3126 197221 14366 20 17 61 2
5 1 9058070 18040 0 4621758645822803355 0 4656499482117394375 4663425876903641771 0 4874 187943 24416 30 15 53 1
$
#AIX-Forum

Originally posted by: Tibor_B

Why dont you just dismiss these numbers as erroneous?

I would say this is a problem of vmstat itself...
next reboot will fix it I believe :)
#AIX-Forum

Originally posted by: SystemAdmin

But I wanted that numbers to look how bad is it.
And no - reboot not fixed the problem. We rebooted system at weekend.
#AIX-Forum

Originally posted by: SystemAdmin

And yes. It looks problem in vmstat itself.
Short programm shows reasonable numbers:
Where should I post this problem to have it fixed some day&
#include <sys/vminfo.h>
#include <stdio.h>
main()
{
struct vminfo64 vi0,vi1,*p1,*p2,*tmp;
vmgetinfo(&vi0, VMINFO64, sizeof(vi0));
p1=&vi0;
p2=&vi1;
while(1) {
sleep(2);
vmgetinfo(p2, VMINFO64, sizeof(vi1));
printf("count of pages paged in: %lld + %lld\n",p1->pageins,p2->pageins-p1->pageins);
printf("count of pages paged out: %lld + %lld\n",p1->pageouts,p2->pageouts-p1->pageouts);
printf("count of page scans by clock: %lld + %lld\n",p1->scans,p2->scans-p1->scans);
printf("count of page steals: %lld + %lld\n",p1->pgsteals,p2->pgsteals-p1->pgsteals);
printf("------\n");
tmp=p2; p2=p1; p1=tmp;
}
}
#AIX-Forum

Originally posted by: Tibor_B

Well I see that you came up with more elegant solution in the meantime...

vmstat is part of http://procps.sourceforge.net/, but I am not sure if AIX uses "vanila" version from procps package or maintain own version...

It would be interesting to know what specific conditions you have on your server that vmstat returned so strange numbers...
#AIX-Forum

Originally posted by: Tibor_B

Hm....

As a workaround you can use

vmstat -s | grep 
"page " 20827235 page ins 3437977 page outs 0 paging space page ins 0 paging space page outs

numbers in first two lines are actuals, so you should run the command twice, calculate difference and divide by seconds :)
#AIX-Forum