Re: Queueing Theory in Oracle

I did some work using QT and Oracle in 2002 or 2003 (just 2 - 3 years after I finished college).
At that time I was brave enough to starting building a queuing model of major Oracle components
- of course if was very complex and even simulation tools I used for it was unable to give me results close to my expectation. Thankfully I read Cary's book and early Craig papers about it and decide that simple M/M/n should work for 90 % of capacity planing exercises.

As Andrey mentioned above we can use G/G/n instead of M/M/n but it will be more complex.

In terms of exponentiation arrival rate I think is more important to be use that metric that we will use will be independent and identically distributed.

Probably title of the book is not a best one for science brains but I would recommend to read this book if you are interested in modeling system - "The art of computer system performance analysis" by Raj Jain - http://www.amazon.com/The-Computer-Systems-Performance-Analysis/dp/0471503363 - not a newest one but math is math.

regards,

-- 
Marcin Przepiorowski
http://oracleprof.blogspot.com



On Sun, Mar 16, 2014 at 3:51 PM, îÉËÏÌÁÅ× áÎÄÒÅÊ óÅÒÁÐÉÏÎÏ×ÉÞ <
Andrey.Nikolaev_at_rdtex.ru> wrote:



> Hello!

> >I am asking if anyone has successfully used queuing theory in Oracle ...

>

> This is very interesting and still developing research area. Typically,

> the Oracle database internal algorithms are much more complex then the

> standard QT models.

>

> In my opinion, the QT provides us with phenomenological background about

> the observables (utilization, arrival rate, queue length, wait time,

> service demand ...) and their interdependences (Little's Law, (1-u)^-1

> behavior,  ...)

>

> This result in the phenomenological rules such as: Amdahl's law, Universal

> Scaling law by Neil Gunther, "Magic of 2" by Cary Millsap, ...

> All the above allows the performance capacity planning and predicting.

>

> In addition to previously discussed in this thread the classical books by

> Cary Millsap,  Craig Shallahammer, and the Hotsos presentations by Henry

> Poras, I would like to recommend the inspiring books by Dr. Neil Gunther.

>

>

> >I guess to use M/M/n we just have to assume certain database metrics are

> >exponentially distributed such as ...

> >If not we arent suppose to use queueing theory because it is based on

> >exponentially distributed data.

>

> Unless you encounter a heavy tailed distribution, the Law of large numbers

> is on your side.

> Approximate G/G/n formulas show behaviors close to the corresponding M/M/n

> formulas and generally differ only by multipliers.

> Therefore, you can use M/M/n expressions for estimations.

>

> However, Dr. Neil Gunther at Hotsos 2012 demonstrated that the heavy

> tailed distributions and the "fractal behavior" do exist in Oracle world.

> Such cases require more complicated treatment.

>

> >Maybe some wait events about latches/mutex/... can be seen as pure

> QueueTime...

>

> Despite the classical QT works well for estimations, the detailed

> investigation of the queuing properties for even the simplest Oracle

> objects, such as mutexes is more complicated.

>

> For example, the "wait" in Oracle has the different meaning then wait in

> Queuing Theory. Even the "average wait time" has another definition.

>

> Some time ago, I presented about Oracle mutexes and the corresponding

> queuing models.

> If you are interesting, you can find the article at

> http://arxiv.org/abs/1212.6640

> (Warning: Math ahead)

>

> With Best Regards

> Andrey Nikolaev

> http://andreynikolaev.wordpress.com

>

>

> --

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

>

>

>


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