Re: Database performance comparison paper.

On 2/20/2007 3:51 PM, Andrej Ricnik-Bay wrote:
> On 2/21/07, Guido Neitzer wrote:
>
>> It would be more or less the same, if you compare copy against insert
>> performance on PostgreSQL and state that insert should be as fast as
>> copy without saying why.
>>
>> Btw: these guys claim to be database consultants.
> Guess one should consider oneself lucky not to be their
> customer, then, since they seem to base their decisions
> on thin air and personal preference...

As the original author of the PHP TPC-W implementation you can find on
pgfoundry, I know pretty good what it takes to make MySQL perform about
as good as PostgreSQL under a real benchmarking scenario. I implemented
all the database access parts basically two times. Once for PostgreSQL
as an experienced DB developer would do it, once turning half the
queries upside down in a horribly unintuitive way to give MySQL+InnoDB
clues how to do it. Of course did I NOT run any of those tests using MyISAM.

In the end, both implementations performed more or less the same,
measured at the HTTP interface. What the PHP+PG implementation did more
elegantly in SQL, the PHP+My implementation had to do with more PHP
code. And that is where all those crappy wannabe-benchmarks just fail to
make sense to me. They measure some common denominator SQL statements at
an abstracted DB API level. That is just nonsense. It doesn't matter how
fast a specific index scan or a specific insert or update operation by
itself is. What matters is how many parallel simulated users of a well
defined business application the System Under Test (middleware plus
database) can support within the responsetime constraints.

All that said, what really scares me is that these clowns apparently
don't even know the system of their preference. No serious DB consultant
would even bother testing anything using MyISAM any more. It is a table
handler only considered for "disposable data".

Jan

--
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me. #
#================================================== JanWieck(at)Yahoo(dot)com #