Re: Datatypes and performance

If disabling FSYNC means that some my transaction log records will
reside in
some OS memory cache instead of been written on a disk after transaction
is commited,
I do now want to disable that.

-----Original Message-----
From: Mattias Kregert [mailto:mattias(at)kregert(dot)se]
Sent: Monday, July 07, 2003 3:59 AM
To: Alvaro Herrera; Maksim Likharev
Cc: PostgreSQL List
Subject: Re: [GENERAL] Datatypes and performance

First of all, disable FSYNC... that will speed things up a lot!

/M

----- Original Message -----
From: "Alvaro Herrera" <alvherre(at)dcc(dot)uchile(dot)cl>
To: "Maksim Likharev" <mlikharev(at)aurigin(dot)com>
Cc: "PostgreSQL List" <pgsql-general(at)postgresql(dot)org>
Sent: Saturday, July 05, 2003 8:15 AM
Subject: Re: [GENERAL] Datatypes and performance

> On Fri, Jul 04, 2003 at 10:50:17PM -0700, Maksim Likharev wrote:
> > Wasn't any sarcasm in my words,
>
> Well, I had really thought so. I'm no native english speaker so I
> frequently misunderstand people.
>
> > but I have benchmarks and my observations about performance in
general.
> > Just want to say, I haven't seen much other DBMS systems other then
> > Microsoft SQL, and Postgres 7.3.x, so my performance observations
> > based on those systems. Selects, do not have any problems in
general
> > equal or slightly slower that MS SQL.
>
> > Inserts/Updates sometimes slow, sometimes a nightmare,
> > in general painfully slow.
>
> This is probably unrelated to the technique I was describing (it's
> called TOAST, by the way). Try using another approach, for example if
> you have lots of inserts, wrap them in a transaction like
> BEGIN
> INSERT ...
> INSERT ...
> ...
> COMMIT
>
> Or use a COPY statement. Both of those approaches are much faster
than
> using hundreds of standalone inserts.
>
> Be sure to tune your server, for example bump up the shared_buffers
> setting, VACUUM frequently, ANALYZE when there's significant
statistical
> change in your data, etc. There are lots of documents on how to do
> this (I'm sure somebody will post appropiate URLs for those).
>
> --
> Alvaro Herrera (<alvherre[a]dcc.uchile.cl>)
> "Por suerte hoy explotó el califont porque si no me habría muerto
> de aburrido" (Papelucho)
>
> ---------------------------(end of
broadcast)---------------------------
> TIP 9: the planner will ignore your desire to choose an index scan if
your
> joining column's datatypes do not match
>