Re: true definition of a transaction

Well one of the reasons for the existence of benchmarks is exactly that they do allow direct comparisons - AIUI the definition of a transaction in tpc-c is the same irrespective of what database or tp-monitor is being used. Similarly I *believe* with ECPerf for J2EE.

Where you don't get straight comparisons - for example Oracle's use of the Java Petstore demo as a 'benchmark' being flat out beaten by the .Net 'Petstore' then you are in the realm of marketing - so far as I can tell nearly anything is allowed there.

So your original guy - was he quoting a recognised benchmark or just some unverifiable figure?

=20

On 5/4/05, ryan_gaffuri_at_comcast.net <ryan_gaffuri_at_comcast.net> wrote:
> I'm familiar with the difference between a logical and physical transacti=
on.
> I brought this up because it appears that people 'benchmark' large system=
s
> by the number of transactions they do. This appears to be a flawed method
> since a transaction in one system may be 1 action and a transaction in
> another could be many.=20
> > I don't consider a transaction to be a technology thing but a logical=
=20
> > concept relating to the application. It should be defined something=20
> > like 'a logical grouping of work tasks that must all be performed=20
> > together and should all succeed or all fail'.=20
> >=20
> > I see the classic banking example has already been given. Let me give
> another.=20
> >=20
> > We run invoices monthly. The process that does this creates updates=20
> > and deletes a lot of data in an Oracle database (well we *hope* its a=
=20
> > lot of data anyway) and then prints some crystal reports on headed=20
> > paper, and for some customers I believe, now generated pdfs that get=20
> > emailed out. The *transaction* in this case as covering all the data=20
> > manipulation, plus the generation and sending of the printouts (if=20
>=20

--=20
Niall Litchfield
Oracle DBA
http://www.niall.litchfield.dial.pipex.com

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