[Pgpool-general] best location for pgpool in a 2 machine arrangement

[Andrew Sullivan]

Aside: your mail almost got plonked by my spam filters.  Please don't
send HTML mail to lists.

On Tue, May 02, 2006 at 06:47:07AM -0600, David Boreham wrote:
> Really ? (as in , do you really expect a 'significant' benefit ??).<br>
> Any database server I ever worked on had a code path much heavier<br>
> than the associated network stack processing. Perhaps PG is different,<br>
> but somehow I doubt it.<br>

You seem to be in some confusion here over the nature of socket
interfaces.  Certainly on every UNIX I know of, if UNIX domain
sockets are available, they are faster than the TCP/IP stack. 
They're even faster on FreeBSD, which has what is widely regarded as
a very fast stack.  If you can use domain sockets, they're usually
faster.

Now, whether the connection is your bottleneck is quite another
matter.  Often it isn't: I've heard people say, "Oh, use domain
sockets: TCP/IP is too slow," but when you look at the application,
they're sending 40 round trips per transaction.  That's going to lose
even if you use the super-experimental-secret quantum computing
connection.  So I agree that they're no magic bullet.

> that avoids burning the CPU time to read in and write out the bits<br>
> on the database machine (which would typically be the place that<br>
> you want to avoid using the CPU if you can avoid it).<br>

It's incredibly rare that database systems are CPU bound.  (Besides,
if performance is in question, I'm assuming we're talking about SMP
boxes.)

A

-- 
Andrew Sullivan  | ajs at crankycanuck.ca
The plural of anecdote is not data.
		--Roger Brinner