Re: Feature Request --- was: PostgreSQL Performance Tuning - Mailing list pgsql-performance

From david@lang.hm
Subject Re: Feature Request --- was: PostgreSQL Performance Tuning
Date
Msg-id Pine.LNX.4.64.0704271804240.31111@asgard.lang.hm
Whole thread Raw
In response to Re: Feature Request --- was: PostgreSQL Performance Tuning  (Josh Berkus <josh@agliodbs.com>)
List pgsql-performance
On Fri, 27 Apr 2007, Josh Berkus wrote:

> Dan,
>
>> Yes, this is the classic problem.  I'm not demanding anyone pick up the
>> ball and jump on this today, tomorrow, etc.. I just think it would be
>> good for those who *could* make a difference to keep those goals in mind
>> when they continue.  If you have the right mindset, this problem will
>> fix itself over time.
>
> Don't I wish.  Autotuning is *hard*.  It took Oracle 6 years.  It took
> Microsoft 3-4 years, and theirs still has major issues last I checked. And
> both of those DBs support less OSes than we do.  I think it's going to
> take more than the *right mindset* and my spare time.

I think there are a couple different things here.

1. full autotuning

   as you say, this is very hard and needs a lot of info about your
particular database useage.

2. getting defaults that are closer to right then current.

   this is much easier. for this nobody is expecting that the values are
right, we're just begging for some tool to get us within an couple orders
of magnatude of what's correct.

the current defaults are appropriate for a single cpu with 10's of MB of
ram and a single drive

nowdays you have people trying to run quick-and-dirty tests on some spare
hardware they have laying around (waiting for another project) that's got
4-8 CPU's with 10's of GB of ram and a couple dozen drives

these people don't know about database tuneing, they can learn, but they
want to see if postgres is even in the ballpark. if the results are close
to acceptable they will ask questions and research the tuneing, but if the
results are orders of magnatude lower then they need to be they'll just
say that postgress is too slow and try another database.

an autodefault script that was written assuming that postgres has the box
to itself would be a wonderful start.

I think the next step would be to be able to tell the script 'only plan on
useing 1/2 of this box'

and beyond that would be the steps that you are thinking of where the
useage pattern is considered.

but when every performance question is answered with "did you change the
defaults? they are way too low for modern hardware, raise them by 2 orders
of magnatude and then we'll start investigating"

David Lang

pgsql-performance by date:

Previous
From: Josh Berkus
Date:
Subject: Re: Feature Request --- was: PostgreSQL Performance Tuning
Next
From: "Harald Armin Massa"
Date:
Subject: Re: Feature Request --- was: PostgreSQL Performance Tuning