Percentage of system memory in Glance

[Tim Medford]

Is it purely my imagination or does the percentage of "system" memory used in hpux 11i almost double as opposed to hpux 11.0?

If you look at the first screen in glance, the bar graphs at the top show memory. Pardon my crude drawing, but under 11.0 it usually looked something like this:

SSUUUUUUUUUUUUUUUUUUUBBB

On the same machine, with roughly the same kernel config, it looks more like this:

SSSSSSSUUUUUUUUUUUUUUBBB

The server is running the exact same applications it was before I did the cold install to 11i.

Has anyone else noticed this behavior?

Thanks,
Tim

[John Poff]

Hi,

I seem to remember reading that 11i actually reports system memory usage more accurately than 11.00. I can't find the thread or docs where I read it, but that is what comes to mind.

JP

[Tim Medford]

Thanks John, let me know if you find that thread. Initially it looked to me like I might have a memory leak somewhere, but it seems to stay constant.

Tim

[Chris Vail]

Remember that 11i has some dynamic kernal tuning parameters. System memory utilization is going to be, by nature, a little heavier in 11i than in 11.0. I wouldn't expect it to double, but it could be possible. I suspect that the kernal is not now as efficient as it will be in later releases. Sun Solaris went through the same teething pains, and now t'is quite efficient.

Look for differences in the per-process numbers, if you have them. You didn't say the system was too slow, so I don't think there's much to worry about.

Chris

[Stefan Farrelly]

yes, weve noticed the same thing. But the dbc_max_pct is set to50% by default on an 11i install - far too high. IF we copy over a system file from an 11.0 server (which weve customised to our needs and explicitly set a lot of parameters) it drops back to SSUU....

[Pete Randall]

After reading Stefan's reply, I started checking my systems and found that the development box, which is 11i and has dbc_max set to 10 shows

SSSSSSSSSSSSSS (14 - I counted)

and the 11.0 production box with dbc_max set to 50% (I could have sworn I tuned that) shows

SSS

There definitely seems to be a difference.

Pete

[Tim Medford]

Thanks for all the replies.

Any ideas on what kind of system processes could be trimmed back? I noticed there's a whole bunch of new kernel parameters in 11i. I pretty much left them at the defaults other than maxssize, maxdsize, maxtsize and some of the IPCS parameters.

The only thing we run on this machine is 3 Oracle instances. I don't need Samba, Apache or many of the other things that come with 11i. I used LVM for the root volume, maybe I can disable all the vxvm startup scripts too?

I can't really say I'm having any performance problems. I just don't want to be wasting 20% of memory on a bunch of optional system modules that I'm never going to invoke.

Thanks, Tim

[John Poff]

Hi again,

I don't think that the issue was reported as a problem, but rather that the 11i kernel does a more accurate job of reporting the actual CPU usage for system time. The 11.00 kernels were under reporting the amount of time spent for system stuff.

I'm still looking for where I read about it. It's starting to drive me nuts. The older I get, the harder it is to remember where I read something. If I had known I was going to live this long I would have taken better care of myself! :)

JP

[parkerleung]

Hi Tim

If you are running oracle 9i (901,920) on 11i, then you will experience shortage of memory problem. I hit this. After consulting a document found in the oracle metalink website, I changed the virtual page size using chatr of the oracle executable. It seems ok now.
Actually the dynamic kernel in 11i is really troublesome. GlancePlus cannot tell the real truth of memory usage, both for the OS & Application such as Oracle