Thread: stonebraker diffs RDBMS

From:
Robert Bernier
Date:

Two links here, comments anybody?

http://www.computerworlduk.com/technology/applications/databases/news/index.cfm?newsid=5059

http://www.databasecolumn.com/

Robert

From:
Josh Berkus
Date:

Robert,

> http://www.databasecolumn.com/

This looks like a last-ditch attempt to drum up interest in Vertica.  Since
Vertica is in trouble and probably doomed, I'm not inclined to pay much
attention to it ...

--
--Josh

Josh Berkus
PostgreSQL @ Sun
San Francisco

From:
"Rodrigo De León"
Date:

On 9/10/07, Josh Berkus <> wrote:
> This looks like a last-ditch attempt to drum up interest in Vertica.  Since
> Vertica is in trouble and probably doomed, I'm not inclined to pay much
> attention to it ...

Interesting. Why do you say that?

From:
Neil Conway
Date:

On Mon, 2007-09-10 at 16:09 -0400, Robert Bernier wrote:
> http://www.databasecolumn.com/

Stonebraker has been saying essentially the same thing for a few years
now. Some more literate presentations of the same ideas can be found
here:

    http://www.cs.brown.edu/~ugur/fits_all.pdf
    http://nms.csail.mit.edu/~stavros/pubs/hstore.pdf

-Neil



From:
Brian Hurt
Date:

Robert Bernier wrote:

>Two links here, comments anybody?
>
>http://www.computerworlduk.com/technology/applications/databases/news/index.cfm?newsid=5059
>
>http://www.databasecolumn.com/
>
>Robert
>
>
>

OK, I'm being stupid here- can someone please explain to me the
difference between a column-oriented database and a fully normalized
row-oriented database?  Most databases are insufficiently normalized,
this I know (including mine).  And this causes pain.  But it seems to me
that full normalized (aka column-oriented) databases also cause pain,
just a different type of pain.  Collecting rows I'm very likely to all
want together onto the same page strikes me as being a signifigant
performance advantage.  Having to read in a page for every column- or
worse yet, O(log N) pages for each column (as I have to traverse an
index)- strikes me as being a huge performance hit.

But maybe I'm not seeing something, and someone can enlighten me.

Brian


From:
Christian Voelker
Date:

Am 11.09.2007 um 14:34 schrieb Brian Hurt:

> Robert Bernier wrote:
>
>> Two links here, comments anybody?
>>
>> http://www.computerworlduk.com/technology/applications/databases/
>> news/index.cfm?newsid=5059
>>
>> http://www.databasecolumn.com/
>>
>> Robert
>>
>>
>
> OK, I'm being stupid here- can someone please explain to me the
> difference between a column-oriented database and a fully
> normalized row-oriented database?

I guess, these are questions belonging to core implementation decisions.
With every select you make, whole rows are loaded depending on
conditions
you specify in your query. This is opposite to searching columns first,
finding rows that meet all conditions in the results. For me it looks a
bit like using an index for each field and everything, but that me be
a very basic understanding of what is dicussed here. In the end, there
will be reasons why RDBMS are such successful. The object DBs have been
around for a while and are gaining momentum slowly but steadily.

The performance argument has always been two edged for a basic decision
on development directions. While performance increase of a magnitude
might open a whole range of new applications it gets eaten up by
hardware acceleration overtime. Novell had a really tough time when
turning from Netware 3 to 4 partly because they made the base assumption
that PC hardware will always be to slow for complex software. What was
their success in Version 3 made them too slow when developing new
features and they lost against Microsoft with their brand new but
then performance encumbered Windows NT. You know what has happened
meanwhile.

Bye, Christian



From:
Simon Riggs
Date:

On Mon, 2007-09-10 at 16:09 -0400, Robert Bernier wrote:
> Two links here, comments anybody?
>
> http://www.computerworlduk.com/technology/applications/databases/news/index.cfm?newsid=5059
>
> http://www.databasecolumn.com/

As Stonebraker himself points out, "one size fits all" isn't true. His
claim than column-oriented databases can perform 50 times faster than
row-oriented (i.e. us) is IMHO true, but only on certain queries and he
neglects to comment that on many queries they suck really badly too.

My advocated approach is to learn from the technique and include lessons
learned into PostgreSQL, that way we have the benefits of multiple
techniques as and when required.

So, yes, one day I expect to include column-oriented techniques into the
database server, but it probably won't be a straight implementation of
verbatim Stonebraker... There is some discussion on -hackers on this
now, but please everybody don't start a surge on this now or soon, we
have a release to get out of the door.

--
  Simon Riggs
  2ndQuadrant  http://www.2ndQuadrant.com


From:
Ileana Somesan
Date:

Hi,

I still don't see the difference between column-orientation and vertical
partitioning.
Any idea?

Best regards,
Ileana