Performance issues

[Chris Fadrowski]

We are running OEM (oracle enterprise mgr) on the server. the problem is that every time we start it up, it appears that it goes right to swap and not memory. We have 1.5 gig of mem and 2 gig of swap. SWAP gets pounded. Is there a kernel parameter or a way to check why this is happening?

[Mladen Despic]

How do you observe this?
Are there a lot of page outs
and/or deactivations?
What does 'swapinfo -tam' tell you?
What is the amount of free memory?
What is the size of buffer cache?
What values do you have for dbc_max_pct and dbc_min_pct?
Do you have Glance?

[Duncan Edmonstone]

How have you determined that swap is getting pounded? Your not confusing swap space being *reserved* with swap space being *used* are you?

HTH

Duncan

[Chris Fadrowski]

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
reserve - 178 -178
memory 1128 484 644 43%
total 2152 662 1490 31% - 0 -


dbc_max_pct 25
dbc_min_pct 5

We could see swap hitting in SAR and in glance. but not sure how to tell if it is "reserved or used".

[S.K. Chan]

From your swapinfo output it doesn't look like device swap lvol2 is been used at all, I can see 178MB reserved though. Could you show the "sar" output as well ?

[Chris Fadrowski]

HERE IS A sar -b when i started the OEM app..

/u05/patches>sar -b 3 7

HP-UX starr B.11.00 A 9000/871 05/09/02

09:43:52 bread/s lread/s %rcache bwrit/s lwrit/s %wcache pread/s pwrit/s
09:43:55 0 38 100 2 4 46 0 0
09:43:58 9 204 95 5 10 55 0 0
09:44:01 89 111 20 1 1 0 0 0
09:44:04 81 974 92 5 10 48 0 0
09:44:07 16 376 96 4 4 0 0 0
09:44:10 14 166 92 5 7 25 0 0
09:44:13 19 217 91 6 6 11 0 0

Average 32 298 89 4 6 35 0 0

[Mladen Despic · Accepted solution]

Chris,

I don't understand where the problem is. Based on the information that you provide, I don't see a problem with swap.

To check for swap and memory issues, use Glance's memory report, or run 'sar -w 3 7'

Look for page outs (swpot/s in sar) and deactivations.

Mladen

[Chris Fadrowski]

we were using the perf monitor in SAM to see the swap area getting hit. I will try this however. thanks.

[S.K. Chan]

With your dbc_max_pct at only 25% (50% is the default setting) and you're getting such a good hit ratio for your buffer cache read (%rcache), I do not see any problem here. You got the %rcache in the range of 20-100%. On top of that %wcache (buffer write hit ratio) is good too. What you don't want is too much reads being done from the physical disk.

[Mladen Despic]

I'm guessing that you've observed swap utilization in Glance (character mode), which can be run from SAM. Near
the top of Glance's main screen there 4 lines like these:

CPU Util S | 2% 3% 5%
Disk Util F F | 3% 8% 14%
Mem Util S SU UBB | 45% 45% 45%
Swap Util R R | 47% 47% 47%


The last line should show the same percentage as PCT USED for "total" in 'swapinfo -tam'

This percentage includes swap reservations, which is quite normal to see even if no swap is actually used. However, if this percentage hits 100%, then no new processes will start up. Indeed, sometimes a running process will crash.

In order to prevent this problem, it is a good idea to monitor the percentage, so if it gets close to, say 80%, you can just add more swap device
(on disk).