CPU Utilization Breakdown by DB

[Stephen Serbe]

Ms. Rita,
Your Web page solution (or PW's) sounds interesting... where can I get a view? Is this a share-ware pkg or an OVO product (Service Reporter, etc.)?

Thanks!
Stephen

[Hein van den Heuvel]

Hey Jim,

How about asking Oracle instead of the OS.
Get some procedures in place as a framework:
- activate statspack on every instance.
- times_statistics is probably not need, just counters
- run a statspack snap after startup, befor shutdown.
- run a statspack snap once a week? day? hour?
- optionally roll up statspacks through queries or exports.

You may be able to use CPU time from the statspack, but why not take an easier number. Notably I would suggest BUFFERED GETS. Optionally toss in PHYSICAL IO COUNTS.
Now pro-rate the aggregate OS collected CPU cost for all oracles by buffered_gets (or executes, or usercalls, or sqlnetpackages, whatever tickles your imagination).

Statspack may be over the top. It may suffice to just run your own simple queries against V$SYSTAT. Check out the ORACLE DATABASE REFERENCE MANUAL. Appendix C: "Statistics Description". The table indicated whether timed_statistics is needed or not for a given counter.


Or how about looking at some externally visible attribute of Oracle. Admittedly only one thing comes to mind: The redo/archive log volume. You can probably tell how many bytes were written. This will be proportional to the database modification work (insert, update, delete) and may or might not be proportional to all the work. A read-mostly DB would not see its full share of system resource usage reflected, but that coudl be our little secret :-).

Just thinking out aloud,
Hein.

[Stephen Serbe]

Good thoughts! I was looking at a "mgmt" level view... especially a web view for managers. I'm using MW extracts to Oracle RDBMS for a relational repository that I can coorelate with other data (configuration and change). Always looking for a way (tools) to improve the process.

The trouble with our times is that the future is not what it used to be.
--poet Paul Valery

[Tim D Fulford]

Hi

I use MeasureWare and run Informix. With multiple instances of Informix what I do is split up the oninits by soft linking them to a "dummy" name. eg oninit-inst1 or oninit-inst2. This means in your parm file you can list all the application as
application = ifx-inst1
file = oninit-inst1
application = ifx-inst2
file = oninit-inst2
etc...

I also collect usage data from each Informix instance (select * from syspfofile); I'm sure Oracle has similar tables.

I find that this allows me to corolate usage spike, with say reductions in caching ot High IO etc. It is worth attcacking the problems from as many angles as possible.

Regards

Tim