Databasen - Revisited

[Paul Makepeace]

On 10/20/06, Matt Sergeant <msergeant at messagelabs.com> wrote:
> On 19-Oct-06, at 2:31 PM, Paul Makepeace wrote:
>
> >>  Even in a replicated setup, databases don't scale
> >> brilliantly across servers (too much replication overhead,
> >> application becomes too complex etc).
> >
> > Depends...
> >
> > http://labs.google.com/papers/bigtable.html
> > http://video.google.com/videoplay?docid=7278544055668715642
>
> Bigtable is NOT any form of RDBMS.

Yes, you said "database". (I know you were using an RDBMS question to
begin with.) To trot out a cliche' Paul Graham did fine without an
RDBMS.

More in reply to the anti-MySQL zealots but worth mentioning here too
as it's a good example of massively replicated RDBMS: Google uses
MySQL for its ads database. (I was going to reply with this earlier
but had to dig something that showed it was public info)

http://xooglers.blogspot.com/2005/12/lets-get-real-database.html --
the comments are pretty interesting too

For those bleating about the unreliability of mysql, if you had an ads
system where failures directly translate to revenue losses of a scale
that most people can barely comprehend you would choose something
reliable. (QED)

The folks going on about the purity, ACID or whatever of Pg are only
addressing one very small part of the equation governing a choice of
DB, namely technical featureset and Computer Science undergrad
compliance. And even that's losing its relevance given that MySQL does
most of that stuff now anyway. They're not deploying MySQL 3.23 any
more, Toto.

(If I were starting a personal project from scratch I would use Pg, btw.)

> It's a highly optimised domain
> specific storage implementation designed specifically for scalability
> and redundancy. I suspect the people who wrote it are probably very
> careful about optimising every bit of their code too. Or perhaps they
> should just trust their optimiser. :-)

Heh.

P