Re: Training for Oracle Performance tuning

Sorry for the delayed reaction but I've been on a long overdue vacation ...

Learning to tune Oracle is a never ending process; I'm not an expert and doubt I ever will be, thank god. That's what I love about Oracle, there's no shortage of new stuff to figure out. Method R is one of my favorite things, and the results I've been able to achieve by using it have been the highlight of my performance reviews for several years. In this year's review, my manager actually described the newly tuned code as 'smokin' (and that was a first). The report he was describing is near and dear to many people's heart's as it provides the sales numbers that predict many of their bonuses - definitely a best seller at year end and everyone starts running it repeatedly at the same time and ... but I digress.

I'm not a consultant. I am responsible for about 2 dozen application databases including SAP and a data warehouse, all of which support a 24x7 global manufacturing operation. Our customers tell us on a regular basis where their pain is, specifically by process, job or report. We look at the instance level performance metrics as well, but primarily to benchmark normal performance so we can identify abnormal when it occurs. Once that process is identified, a trace file shows the issue, and more importantly, captures it in something that can be used to marshal the resources needed to fix the damn thing. In a couple of cases, I've found that being about to pull out the historical documentation from past tuning efforts has been invaluable to demonstrate where the issues lies. (Sure it ran in 5 minutes last year, but we've increased by xx% in the blah blah market. If we need in 5 minutes now, some modifications might be required. etc etc)

All that being said, I'd vote for the Orapub class and the Hotsos class. You won't go wrong with either but keep in mind that learning both approaches is best. I haven't had the chance to attend Craig's class, but his book is great and it seems to me the classes address complimentary skill sets. I don't think I'd trust any class that claimed to be the complete source for performance tuning knowledge anyway.

hth ... Robyn

On 9/28/07, Sunil Kanderi <sunil.kanderi_at_gmail.com> wrote:
>
> Thank you for thoughts. My thoughts were exactly in line with what you
> lay out here. I read chapter 1 in Cary Milsap's book and for all the ERP
> applications and also database instances that host multiple applications
> on the same database, it would be hard for me to get the customers to
> give me their most important tasks for business. I believe it is a
> definitely a great approach, but in my case it was hard to see the
> practicality of the approach for my specific situation. That is why I
> opted for their Oracle SQL tuning class instead of their PD101 which
> concentrates on Method-R. Having said that, I probably shouldn't be
> making judgments without taking the class. Based on the feedback I had,
> Hotsos classes are regarded very highly and that is why I had them as
> one of my option. I will probably take PD101 at some point. For now, I
> have decided to take the Orapub class. I will give my feedback to the
> group once I am done with the class. Thanks for all the feedback.
>
> Thanks,
> Sunil.
>
> fmhabash wrote:
> > I read the OraPub class description here
> >
> http://resources.orapub.com/ProductDetails.asp?ProductCode=RPM&Show=TechSpecs
> > . My advice, take this class first then if you still need to, use the
> > HotSos one. If you care to know why, keep reading ....
> >
> > I have read and attended many Oracle perf approaches. Last one was
> > HotSos Metho-R one. One thing that distinguishes HS from others is
> > that they have a science-evidence-based approach. Compare this to what
> > others keep calling 'art' or at best the 'bed-time' stories that some
> > publish on their web sites as 'gurus'. I think they are, but their
> > approach (if you we can call as such) is nothing more that a
> > collection of experiences they have collected over the years. IMHO,
> > any perf turning approach must yield accurate and reproachable results
> > using a clearly defined tools and steps. And it should be something
> > that can be taught and used by DBAs without necessarily having 20
> > years of experience.
> >
> > Having said that, I have 4 reservations on HS M-R approach ...
> >
> > - This approach refuses to give any attention to instance level
> > tuning. On some occasions, they won't even read a statspack report.
> > Its entire focus is on identifying most important tasks for the
> > business and collecting diagnostic data on it. This sounds great in
> > concept, but in real life, it does not work this way. First, most
> > often than not (at least in my experience), users and their managers,
> > have no time (and sometimes knowledge) even to agree on a list. System
> > is slow and we need to know why NOW. You will find out when such issue
> > strikes, you won't have the time or initiative to ask people on the
> > call to go back and a gather a list so you can have time to instrument
> > your database to collect stats. Specially, if your application users
> > MTS, connection pools, or in large RAC environment.
> >
> > - Has no focus on real-time performance monitoring and
> > troubleshooting. I was disappointed when the class had not even
> > addressed it. In real life, you must be able to decide on some tools
> > that allows you to look at a DB real-time and be able to spot an issue
> > immediately. This is the most critical of all. This class will not
> > prepare you for this. This is what I do almost every day.
> >
> > - This class will not prepare you for necessary tools and knowledge to
> > able to research and report on a DB performance historically. This is
> > again very critical for me. What tools can I use to be able to do
> > this. Furthermore, I'm asked repeatedly by upper management to provide
> > such reports on periodically basis. This class will not help you in
> > this regard.
> >
> > - Class focuses at least by 80% of its time on the method itself.
> > Little emphasis is put on Oracle internals which are mandatory to
> > understand performance. Also, I did not think enough emphasis was put
> > on the resolution part and how these measure will work to resolve the
> > issue.
> >
> > IMHO, the HS M-R class, is an advanced class that I will not recommend
> > as an initial step. I think this method worked great for HS group due
> > to the nature and magnitude of problems they get consulted for. At
> > that level of consulting and visibility, all the necessary
> > pre-requisite work is probably have been done for them. But for
> > everyday DBA performance issues, it will not help as much.
> >
> > I have been involved in so many performance issues since I attended
> > this class, all of them have been diagnosed and resolved. However,
> > only in a few of them I had to get to the level where I really needed
> > to use M-R.
> >
> >
> >
> --
> http://www.freelists.org/webpage/oracle-l
>
>
>

--
http://www.freelists.org/webpage/oracle-l