Kernel Parameter Monitoring

[Joe Kanakaraj]

Hi Guys,

I have a multi processor rp7410 running oracle instances and at frequent intervals the system slows down and becomes sluggish. I cannot afford downtimes since it is a production box. I have a downtime scheduled in a short while and would like to make the best use of it, including modifying of kernel parameters. My DBA has done a few calulations on his own and we are working out a set of values, but what I would like is a tool which or script whaich can help monitor kernel parameters in real time.. Sarcheck and EMS are being used right now, but EMS seems worthless and too slow and Sarcheck jus to confusing. Any of you guys know some tricks or ideas or tools will be of great help.

WOuld really appreciate the help.

Live free & Die,

Joe

[Pete Randall]

Joe,

The best tool is Glance. There is a trial version available on your application CDs if you don't have it installed already. Run gpm (this gives you the graphical version - though you can get along with running the terminal version: glance), and go into the Reports > System Info > System Tables Report. This gives you current, average, and high water mark metrics for all the key kernel parameters. There is no better way to get this info.


Pete

[Steven E. Protter]

Glance is the best tool to get started on the situation.

It has a special screen that will let you know if you are exceeding or close to maxing out on kernel values.

You can collect data over a longer period of time with a sar scripts.

I a good set I usually post but I have no idea how to access the darned things from Jerusalem. Perhaps someone could post them up.

SEP
Feeling separation anxiety from my cherished boxes.

[harry d brown jr]

I agree with Pete and Steven, if you aren't using glance/measureware/perfview/what ever it is called today then you aren't using the right tools. sar, sarcheck, vmstat, iostat, ... can NOT measure performance at the level "glance" does. sar and the other "tools" are only good for quick health checks and should never be used to determine how to tweak a system for better performance.

live free or die
harry d brown jr

[Ralph Grothe]

Hi Joe,

the easiest and quickest road to collecting metrics on HP-UX really is, as already has been mentioned, using GlancePlus.
You can even use the trial version for your purpose.

Because you said that sar was too confusing I couldn't really advise you to use the other free tool that I use on a few boxes that don't have such nice tools like glance (e.g. Solaris).
Since it's more or less just a data storing container with the capability to produce nice graphs, it is of course left to the discretion of the user to devise means to fill the database.
I'm talking of RRD Tool (which stands for round robin database), and that stores data in a kind of ring buffer very similar to the MWA perf logs so that these database files won't exceed the predefined size
(which is nice from a sysadmin point of view)
But you do have to feed it at regular intervals, that you also define as you create the database.
Thus it's predestined to hold metrics and counters like you need to store for getting a picture of your system's performance.
It also has special consolidation functions that do the interpolating of missed out data points.
In order to collect the data you usually would have to script a little and know of the commands and probably other tools how to retrieve the metrics.
If you are still interested to find out more go to

http://people.ee.ethz.ch/~oetiker/webtools/rrdtool/