What is normal swap space utilization on Solaris 10

Question

Hi all,
I'm running Oracle 11.2 on Solaris 10 on a couple of HP Proliant DL 360 servers.
Both servers have 72G of physical RAM with swap space set to 16G on both of them.
Server A has only one database and total memory free = 30G.
Server A
top:  Memory: 72G phys mem, 30G free mem, 16G total swap, 16G free swap
swap -s:   total: 27249744k bytes allocated + 13873764k reserved = 41123508k used, 1000552k available
prstat: 
NPROC USERNAME  SWAP   RSS MEMORY      TIME  CPU
   257 oracle     39G   38G    53% 222:11:52 5.6%
    31 root       57M   59M   0.1% 414:47:23 0.1%
     1 smmsp    1776K 7736K   0.0%   0:00:34 0.0%
     6 zabbix   4752K 4092K   0.0%   0:58:31 0.0%
     4 daemon   3864K 6456K   0.0%   0:00:35 0.0%Server B has two databases and total memory free = 9G.
Server B
top:  Memory: 72G phys mem, 9890M free mem, 16G total swap, 16G free swap
swap -s:  total: 29223360k bytes allocated + 627312k reserved = 29850672k used, 16926320k available
prstat:
NPROC USERNAME  SWAP   RSS MEMORY      TIME  CPU
   157 oracle     28G   28G    39%  15:38:41 0.4%
    34 root       58M   65M   0.1%   2:56:57 0.0%
     6 zabbix   5580K 4816K   0.0%   0:00:31 0.0%
     1 smmsp    1776K 5724K   0.0%   0:00:00 0.0%
     5 hpsmh      17M   13M   0.0%   0:00:00 0.0%
     4 daemon   3204K 5912K   0.0%   0:00:00 0.0%We are using zfs file system on both servers (which is pretty much the standard these days on Solaris).
Recently I got an OEM alert that my swap space on server A had crossed the 95% threshhold on one of the servers.
But when I checked the server, I found that the average swap space utilization was 97.45.
In fact, what actually happened was my swap utilization momentarily dropped below 95% and then returned back to its normal range above 95% which caused the alert to be triggered.
So this made me wonder why my swap space utilization was so high on server A, or is this just normal for Solaris (v.10).
Checking with server B, I see that my swap utilization is only at 63.6% (even though server B has much more physical memory in use by the two databases than server A).
Main question is why is swap utilization so high on server A, which is configured the same as server B and with less physical memory actually in use.
Next question is should I be concerned.
When I check vmstat, I do not see any paging in or out or blocked processes.
See below for server A
Server A
$ vmstat -S 5 5
kthr      memory            page            disk          faults      cpu
r b w   swap  free  si  so pi po fr de sr s0 s1 s2 s5   in   sy   cs us sy id
0 0 0 1059868 30507176 0 0  0  0  0  0  2  7 -0 123 30 13742 25008 7264 5 2 93
0 0 0 1024076 30982140 0 0  0  0  0  0  0 23  0  0 122 4433 14793 6854 6 2 92
0 0 0 1030292 30987500 0 0  0  0  0  0  0  0  0  0 102 4055 15049 7014 8 1 91
0 0 0 1044484 30999572 0 0  0  0  0  0  0  0  0  0 129 5905 19196 8127 6 1 93
0 0 0 1028584 30987636 0 0  0  0  0  0  0  0  0  0 114 10611 19925 7259 6 3 90

Answer 1

974632 wrote:
Looks like we don't have 'free' on these Solaris boxes (only the man pages).
I'm guessing that free is for linux (since it works fine on my linux boxes).
$ whereis free
free: /usr/man/man3c/free.3cdarn!
Realize that SWAP is purely an OS facility; which is 100% external to Oracle.
The OS send little used or idle processes into SWAP when RAM is scarce resource.
The fact that SWAP is being used is not a Bad Thing, in and of itself.
as long as vmstat shows that BI+BO > SI+SO I would ignore the Chicken Little warnings.