Re: Plotting Statspack data
- From: Rhojel_Echano@xxxxxxx
- To: oracle-l@xxxxxxxxxxxxx
- Date: Fri, 5 Mar 2004 09:08:22 +0800
Thanks Jared. Sorry my post's format, it just looks OK on my mail client
so I didn't bother
writing it in a better form.
Jared.Still@xxxxxxxxxxx
Sent by: oracle-l-bounce@xxxxxxxxxxxxx
03/05/2004 01:46 AM
Please respond to oracle-l
To: oracle-l@xxxxxxxxxxxxx
cc:
Subject: Re: Plotting Statspack data
See yapppack.zip at miracleas.dk
Jared
PS. Please use paragraphs. I hope I am not
the only person that finds your post difficult to read.
Rhojel_Echano@xxxxxxx
Sent by: oracle-l-bounce@xxxxxxxxxxxxx
03/04/2004 12:34 AM
Please respond to oracle-l
To: oracle-l@xxxxxxxxxxxxx
cc:
Subject: Plotting Statspack data
Hi Guys,
I want to hear your comments regarding this present endeavor of mine....
My bosses wants to see some form of automated "database performance
indicator reporting" scheme. They want me to create some form of "early
warning device" from this kind of scheme, that is, based on certain
metrics and some predetermined threshold values, the scheme would notify
the admins (including me) if the database needs tuning. I talked to my
boss and it was clear to me that he was thinking of ratios (I guess he
stopped learning about the Oracle RDBMS after 7.3 ;-)). I explained to him
the current paradigm being used by most Oracle DBA's today (I hope) in
diagnosing database performance, i.e. wait events analysis. And I told him
that the metrics that can tell us systemwide performance is provided by
Statspack which is implemented in our site. So, in order to deliver to my
bosses this scheme that they wanted, I thought of plotting statspack data
using RRDTool. This is different from what Mr. Burleson did in his book ,
however. What I did is I generat ed the values that SPREPORT will produce
in its output (specifically the "load summary -- per transaction
metrics", "instance efficiency percentages", and "top 5 wait events"
sections), store it in the database, and plotted it to show daily, weekly,
and monthly "trends". The statistics are generated daily for two snapshots
(with one hour interval, i know this is quite long) taken during operation
peak period. So how I got around the "treshold" values that they demanded?
I told them that a baseline will be determined, and metrics that shoots up
or down dramatically against this baseline will be investigated.
I just want to know your thoughts on this. I've been so immersed on this
project that I may be missing some important points. = )
TIA
Rhojel
Other related posts:
