Re: Evaluating overall performance
Date: Thu, 8 Apr 2010 16:54:37 -0700 (PDT)
Message-ID: <9917ace1-b93a-40b6-9a69-5ff1c266576e_at_g11g2000yqe.googlegroups.com>
On Apr 8, 1:53 pm, Helma <helma.vi..._at_hotmail.com> wrote:
> On Apr 8, 4:58 pm, AD <alain.den..._at_gmail.com> wrote:
>
>
>
> > Hi,
>
> > I've been on the development side of things most of the time for the
> > past years,
> > but lately getting to do more dba work.
>
> > I yet need to find material on "seeing the big picture" so to be able
> > to describe and evaluate the performance of an Oracle system.
>
> > The books I own are very good on drilling down to the fine details of
> > a query, but I need to do the opposite and improve my skill on
> > "measuring" the performance of the whole server, enabling me to
> > communicate with others why the system is in good shape or why it
> > needs attention.
>
> > ..sort of the opposite of "tuning" for a specific tasks, and more like
> > assessing the actual performance of the server...
>
> > any good books, articles and links are welcomed,
>
> > Thanks,
>
> > Alain
>
> Hello Alain,
>
> As others have pointed out, there is a business dimension to
> performance. This means that even if the ' big picture' isn't good,
> ( e.g. with lots of inefficient queries), this still can be acceptable
> and not worth tuning. Tuning without the business dimension can often
> lead to behavior what is called 'Compulsory Tuning Disorder' : keep on
> tuning without goal.
>
> Having said that, there are indeed ways to get ' the big picture' -
> this is mainly used for capacity planning. ( e.g. can the database
> handle an load of 50 extra users?).
>
> The person to check out is Craig Shallahamer. His ' Forcasting Oracle
> Performance' and ' Performance firefighting' books are very good for
> the big picture of the system. He stresses the point that there are 3
> area's ( OS, application and RDBMS) that need to participate in this
> big picture.
>
> Helma
There are some unspoken assumptions in both Mladen's application centric view and the business-centric view, the former usually being that the system is somewhat in the ballpark and the latter that the business has a rational view of system performance.
Most of us have seen how really, really bad some applications can be developed. While it is true that most of these problems come from bad design and bad coding, it often goes hand-in-hand with myths or other incompetence configuring systems for Oracle. In the past, it was a much bigger deal on the Oracle side, though these days the defaults are fairly reasonable, they can still be out of whack when installations are done by drones. The advisors are somewhat helpful if one takes the time to learn what they are really saying. On the OS side, things can be way out of whack, even by experienced admins. Think of all the variants on async just to start. Throw in ISM, GUI monitoring bugs, paging algorithms, NUMA, emctl and its shenanigans, java, perl, and there is a big ugly picture that needs to be seen. So while you can say that most problems come from bad SQL, the unspoken assumption is that all those other things are not a problem - and that's just wrong.
The business-centric viewpoint really is rationalization for outside consultants. Not that there's anything wrong with that, it just needs to be admitted. Many places have docile users who have been convinced (or just plain told) that whatever performance they see is the way the system works. It may be true, or not, but business people often make snap myth-based judgment calls about computer systems. Sometimes they think they are being rational by looking at pretty pictures from expensive tools they have bought. That of course implies that the tools are good and they are being interpreted correctly. Sometimes that is true. Performance consultants are likely to have a skewed view of what to do, as they are usually called in when the myths hit reality, so they assume most things are close to properly configured. But it is also a big opportunity for BS artists to come in and unnecessarily throw hardware at it, or worse (ie, twiddle with undocumented init.ora parameters). So there is definitely a need for an unbiased big picture for local staff. In addition, a business may not see justification for proper tools, kind of like people can never properly evaluate their own risks, so they won't buy insurance unless compelled. Even free tools are expensive, and Oracle options are ripe for a rant.
By the way, Craig has a lot of free tool downloads at orapub.com, but I haven't had time to check them out.
jg
-- _at_home.com is bogus. http://api.ning.com/files/shK09qQmvXLD4jLnouPQyt9AyjX0Xsk0wdAwqaPfHgbtQ9gBO01H0VnFDUI85fee/SunOracle.pngReceived on Thu Apr 08 2010 - 18:54:37 CDT
