From: "Aaron Werman" <awerman2@hotmail.com>
To: <ischamay.andbergsay@activestateway.com>,
	<pgsql-performance@postgresql.org>
References: <20040912042442.49915.qmail@web41508.mail.yahoo.com>
	<pB21d.193816$X12.122954@edtnps84>
Subject: Re: Data Warehouse Reevaluation - MySQL vs Postgres -- merge tables
Date: Tue, 14 Sep 2004 09:24:35 -0400
MIME-Version: 1.0
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: 7bit
Message-ID: <BAY18-DAV122akGAyW80002c65d@hotmail.com>

> Mark Cotner wrote:

> > The time has come to reevaluate/rearchitect an
> > application which I built about 3 years ago.  There
> > are no performance concerns with MySQL, but it would
> > benefit greatly from stored procedures, views, etc.
>

From: "Mischa Sandberg" <ischamay.andbergsay@activestateway.com>

> If your company is currently happy with MySQL, there probably are other
> (nontechnical) reasons to stick with it. I'm impressed that you'd
> consider reconsidering PG.

I'd like to second Mischa on that issue. In general, if you migrate an
*existing* application from one RDBMS to another, you should expect
performance to decrease significantly. This is always true in a well
performing system even if the replacement technology is more sophisticated.
This is because of several factors.

Even if you try to develop in totally agnostic generic SQL, you are always
customizing to a feature set, namely the ones in the current system. Any
existing application has had substantial tuning and tweaking, and the new
one is at a disadvantage. Moreover, an existing system is a Skinnerian
reward/punishment system to the developers and DBAs, rewarding or punishing
them for very environment specific choices - resulting in an application,
dbms, OS, and platform that are both explicitly and unconsciously customized
to work together in a particular manner.

The net effect is a rule of thumb that I use:

NEVER reimplement an existing system unless the project includes substantial
functional imporovement.

Every time I've broken that rule, I've found that users expectations, based
on the application they are used to, are locked in. Any place where the new
system is slower, the users are dissatisfied; where it exceeds expectations
it isn't appreciated: the users are used to the old system quirks, and the
improvements only leave them uncomforable since the system "acts
differently". (I've broken the rule on occation for standardization
conversions.)

My expectation is that pg will not get a fair shake here. If you do it - I'd
like to see the results anyway.

/Aaron