WLM and Oracle 9i Configurations

[James Specht]

We have run WLM on HPUX 11i and Oracle 8 for over a year and a half without problems. Once we upgraded to oracle 9i we started seeing problems with WLM giving away too many CPU shares to our production instance of oracle to the point where the rest of the box is almost dead in the water. It never gives back the shares once it reaches the max value. We have to stop and restart WLM to clean the mess up. The logs show very high return times on the performance testing sql we run when it starts, but it returns to normal values pretty quickly, it's just that WLM won't give back the CPU shares. This happens several times a day and night. I got tired of sshing in in the middle of the night so I have turned off WLM, which has removed the problem, but has added others of course.

Has anyone run into this before with WLM and any other application? Or does anyone with a working setup of WLM and oracle 9i wish to share their WLM configuration and/or sql if applicable?

Thanks in advance.

-Jim

[melvyn burnard]

Just out of interest, which version of WLM are you using?
The latest version is 2.1, with a newer release coming in Decmber.
There are a few enhancements/fixes in the newer releases.
You may also want to check your patch levels for WLM/PRM

[James Specht]

Returned from what wlmd.

HP-UX WLM A.01.02.PHSS_25169 (2001_09_05_18_34_03)

I am pretty good about patching our production system whenever I can get the downtime. Usually every 2 months or so. I stay pretty current.

[melvyn burnard]

James, That is quite an old version, I would suggest you look at updating to the later version and see if this helps.
There are also Toolkits with the newer versions, which I think were not around with the version you are using.

[Jonathan Fears]

Hello Jim,
You say that WLM is giving away too many shares? How are you coming to that conclusion (prmmonitor output, top, ...)?
The return times may go back to normal values, but you might check WLM to see if it considers the SLO's goal satified. The "goalsatis" value in the /var/opt/wlm/wlmdstats file gives that information. If the goal is not satisfied and the SLO is at the max it can request, that is the entitlement the workload group will remain at. Also, be sure that you're metric is being refreshed by lookin at the "metfresh" value in the wlmdstats file. Make sure the data collector giving values to WLM for that metric is still alive. I don't know of any issues with Oracle9i, but these are some of the things that could cause WLM to "freeze" entitlements.