Re: Presentation - Mailing list pgsql-performance

From Shridhar Daithankar
Subject Re: Presentation
Date
Msg-id 3F843274.30500@persistent.co.in
Whole thread Raw
In response to Presentation  (Jeff <threshar@torgo.978.org>)
Responses Re: Presentation
Re: Presentation
Re: Presentation
List pgsql-performance
Jeff wrote:
> Let me know if there are blatant errors, etc in there.
> Maybe even slightly more subtle blatant errors :)

Some minor nitpicks,

* Slide 5, postgresql already features 64 bit port. The sentence is slightly
confusing
* Same slide. IIRC postgresql always compresses bytea/varchar. Not too much sure
about which but there is something that is compressed by default..:-)
* Tablespaces has a patch floating somewhere. IIRC Gavin Sherry is the one who
is most ahead of it. For all goodness, they will feature in 7.5 and design is
done. There aren't much issues there.
* Mysql transaction breaks down if tables from different table types are involved.
* Mysql transactions do not feature constant time commit/rollback like
postgresql. The time to rollback depends upon size of transaction
* Mysql does not split large files in segments the way postgresql do. Try
storing 60GB of data in single mysql table.
* Slide on informix. It would be better if you mention what database you were
using on your pentium. Assuming postgresql is fine, but being specific helps.
* Slide on caching. Postgresql can use 7000MB of caching. Important part is it
does not lock that memory in it's own process space. OS can move around buffer
cache but not memory space of an application.
* Installation slide. We can do without 'yada' for being formal, right..:-)
(Sorry if thats too nitpicky but couldn't help it..:-))
* initdb could be coupled with configure/make install but again, it's a matter
of choice.
* Slide on configuration. 'Reliable DB corruption' is a confusing term. 'DB
corruption for sure' or something on that line would be more appropriate
especially if presentation is read in documentation form and not explained in a
live session. but you decide.
*  I doubt pg_autovacuum will be in core source but predicting that long is
always risky..:-)
* Using trigger for maintening a row count would generate as much dead rows as
you wanted to avoid in first place..:-)

All of them are really minor. It's a very well done presentation but 45 slides
could be bit too much at a time. I suggest splitting the presentation in 3.
Intro and comparison, features, administration, programming and tuning. Wow..
they are 5..:-)

Can you rip out informix migration? It could be a good guide by itself.

Thanks again for documentation. After you decide what license you want to
release it under, the team can put it on techdocs.postgresql.org..

Again, thanks for a good presentation..

  Shridhar



pgsql-performance by date:

Previous
From: Jeff
Date:
Subject: Re: Sun performance - Major discovery!
Next
From: Rod Taylor
Date:
Subject: Re: Presentation