Re: Mathematica RDBMS Model
Date: Mon, 22 Jan 2001 07:17:46 +0100
Message-ID: <3A6BD08A.EDCF4AD4_at_itam.zabrze.pl>
That is it!
So, you got the picture!
When i construct such model i need to find how it
statisicaly behave when i put in some kind
of neural hashing module :-) (for instance)
The problem of locking .. hmm.. i'know.
I schould find a statistical connection
between size of database, type of signals
which come to model, and types of answers.
but of course i can't make such assumption - 90%.
I looking around to find appromimate mathematical tool.
Maybe M/M/1 is not the best - by the way - thx for advice!
But is there any other beter statistical tool ?
Very interesting is high load problem in such model. - isn't it ?
In some cases i feel that neural module may be best solution.
There are still lot's of job to be done. :-)
Jan Hidders wrote:
> Michal Widera wrote:
> > Jan Hidders wrote:
> > > Michal Widera wrote:
> > > > Jan Hidders wrote:
> > > > > Michal Widera wrote:
> > > > > >
> > > > > > Is there anybody knows something about mathematical (queued
> > > > > > maybe...) model of simple RDBMS ?
> > > > >
> > > > > What kind of model are you looking for? [...]
> > > >
> > > > I'm looking for a simplified theoretical model of database
> > > > engine.
> > >
> > > So the next question is what you want the granularity of the model
> > > to be? [...]
> >
> > Well, i think that the model should be projected step down - from the
> > black box to the most posible granuarity. In first aprox. i think
> > that the simple queriers should be black box'es, there also should be
> > some kind of process modeling. (Data base writter, Listener,
> > Archiver, etc... like internal structure of Oracle Database) What
> > kind of matematic tool i'll use - it still open problem. I think
> > about construct a simple M/M/1 ? What you think about it ?
>
> Oh. So you want a statistical model! The question is of course how
> appropriate such a model would be. The assumptions of the M/M/1 model
> are probably not very valid; the queues esentially are FIFO queue but
> transactions can be postponed if they need a lock that is not free. But
> of course you can make some general assumptions about this like "in 90%
> of the cases the lock will be available" or "if a lock is not available
> it usually becomes so in a few secondes" but that is not very
> realistic, especially if you want the model to describe accurately what
> happens tot the database under a heavy load.
>
> So the central question has to be: what is it that you would like to
> achieve with this model? How realistic do you want it to be? How
> complicated do you want it to be? Obviously there is a trade off here.
>
> --
> Jan Hidders
Received on Mon Jan 22 2001 - 07:17:46 CET