Re: Postgres performance comments from a MySQL user - Mailing list pgsql-general

Hi Jay,

 From his comments about PostgreSQL's performance, I'd extremely
strongly suspect he has only tested it with untuned (default) memory
settings.  The default memory settings in all of the previous releases
of PostgreSQL have been extremely conservative to maintain the widest
compatibility.  However, they _significantly_ reduce the throughput and
transaction speed of PostgreSQL.

The _very first_ thing to do with any PostgreSQL installation is bump up
the memory settings (at least "sort_mem" and "shared_buffers") in the
postgresql.conf and restart it.

Tell him to test it with decent settings (try about 4000 for each as an
initial start), and he'll find that a decently tuned PostgreSQL matches
the speed of a MySQL installation with any table type.  An in
write-intensive applications, the MySQL server will always fall behind.
  _Especially_ as the number of simultaneous clients rises.  MySQL falls
behind, as does Oracle 8i a bit further on (not sure about 9i and 9iR2),
and PostgreSQL keeps on performing at pretty much the same throughput
for far higher numbers of client connections.

And _that_, is for real.

Regards and best wishes,

Justin Clift


Jay O'Connor wrote:
> Some backs-story.  I'm in the process of converting our internal file based
> data storage to an RDBMS.  After looking a bit at PostgreSQL and MySQL, I
> chose Postgresql.  My boss has heard of MySQL and has not heard of
> PostgreSQL and every now and then ahe make allusions that we shuold be
> using MySQL.
>
> One comment he got from the architect of another web site is as follows
>
>
>>If we were to start again
>>from scratch now, I'd still use InnoDB over postgres unless the
>>performance picked up with postgres recently.
>>
>>Keep in mind our application is very write-heavy so your numbers may
>>be different. Does postgres still keep the old row versions in the
>>primary-key B-Tree? If it does I doubt performance improved much for
>>write-heavy apps, that was a very poor design decision by them. InnoDB
>>takes the Oracle route of moving old row versions to a seperate
>>on-disk data structure.
>
>
> Does what he say make sense? If so, has the situation changed?  BNasically,
> I need something intelligent to say to my boss to either counter or
> mitigate his perception.
>
> Thanks
>
> Take care,
> Jay
>
> ---------------------------(end of broadcast)---------------------------
> TIP 5: Have you checked our extensive FAQ?
>
> http://www.postgresql.org/docs/faqs/FAQ.html