Re: Hardware tuning (Was: Performance question)

On Sat, 22 Sep 2001, Justin Clift wrote:

> Good to hear this has been of benefit.  From reading your email, you
> haven't altered the value of sort_mem, just shared_buffers.
May be I was a little bit unclear.  I stopped increasing sort_mem after
there was no siginificant increase in speed.  My latest value was
   sort_mem = 65536
which is higher than any suggestion I´ve got here.
I observed no increase in speed against
   sort_mem = 32768
and
   sort_mem = 131072

> It might be worthwhile checking things out with sort_mem at different
> levels too before homing in on the "best" value(s) for your application
> & setup.
I observed the only difference when I increased shared_buffers.  This
difference was not significant regarding speed but memory usage (measured
by top).
Today I observed that I seemed to reach the border for simply doubling
the shared_buffers value.  If I set shared_buffers = 262144 and double
the kernel.shmall and kernel.shmmax values in /etc/sysctl.conf postmaster
says:

psql: connectDBStart() -- connect() failed: No such file or directory
        Is the postmaster running locally
        and accepting connections on Unix socket '/var/run/postgresql/.s.PGSQL.5432'?

(this is the same error I´ve got when I increased shared_buffers without
adjusting sysctl parameters.)

So my question remains:  Why isn´t the databased cached if there is
enough space to do so?  I think it is not cached because there is
no speed difference in doing the first query after Server restart and
repeating the same query.  Moreover top shows an continual increase
of memory usage over a longer query.  If the database would be cached
the memory usage should be high also at the start of the query (IMHO).

Kind regards

         Andreas.