Re: cpu average load
- From: Mladen Gogala <gogala@xxxxxxxxxxxxx>
- To: Paula_Stankus@xxxxxxxxxxxxxxx
- Date: Sat, 04 Dec 2004 04:37:00 +0000
On 12/03/2004 11:05:28 PM, Paula_Stankus@xxxxxxxxxxxxxxx wrote:
> (as a whole) then drilling down from there. Also, proactively
> monitoring system resource utilization on a regular basis if you are
> supporting a number of databases operationally has proven useful to
> me.
There is a good carpentry book about some wooden tables and some other =20
forms of furniture in which one among the authors mentions an old IBM
criteria: job that consumes more then 8 CPU seconds is considered a =20
batch. I believe that the similar criteria can apply today:
If you have dedicated server configuration and if any of the server =20
processes on an OLTP system has accumulated more then 8 seconds of CPU =20
is considered a candidate for an investigation. In the days of yore IBM =20
didn't use the 8 CPU seconds criteria as an indicator, they used as a =20
criteria to separate batch, OLTP and DW. That philosophy is still =20
relevant today.
The illusion of proactive monitoring revealing you a problem is just =20
snake oil. You cannot monitor perception and it is user's perception of =20
the application that counts. You can only monitor consumption of =20
computer resources: CPU, memory, I/O bandwidth, semaphores and network =20
bandwidth. The fact that some process resource consumption is above
a chosen mark does not mean that the process should not do it or that =20
you have a problem with the resources. Tuning is always reactive, =20
because in order to start tuning, something must be out of the tune.
Resource consumption alone can not tell you that. Sometimes a form
made with a disregard for the rules of good UI design can make users =20
unhappy and nagging about the "database being slow", and a simple =20
redesign and adding a select list and rearranging fields on the screen
so that logically grouped information is visually close can make users
stop complaining. May be Pythias would be able to predict that, after =20
all, her predictions where said quixotically on laurel (SQL for short)
but I am not. Any regular report you give to your management is =20
exclusively for their own amusement and is, generally speaking, useless =20
as a health indicator.
--=20
Mladen Gogala
Oracle DBA
--
//www.freelists.org/webpage/oracle-l
Other related posts:
