On 6/1/07, Joshua D. Drake <jd@commandprompt.com> wrote:
> Alexander Staubo wrote:
> > You mean aside from the obvious one, scalability?
>
> Multimaster doesn't give you scalability (at least not like a lot of
> people think it does).
That depends on your particular definition of multimaster.
> Databases are a bottleneck when you get a bunch of so called web 2.0
> developers thinking they know an inch about databases.
That's a fairly bizarre and disrespectful blanket statement that
includes some good friends of mine, and probably many of people on
this list. I wonder you what you mean by it; regardless of the type of
application, a single machine has a very real, very finite number of
transactions it can process per second, and a finite number of
spindles you can attach to it. No amount of handwaving about the
incompetence of others changes that.
Note that I am not talking about multimaster replication here, just
scalability. Like Andrew Sullivan I think multimaster replication is
infeasible by design, at least with a system such as PostgreSQL.
> What you are basically saying below is... web 2.0 developers such as
> rails developers have so fundamentally broken the way it is supposed to
> be done, we should too...
I don't know if I said that, but I would love to hear how they have
broken it, and what you propose the solution to be.
The eBay architecture is one interesting example, demonstrating a
situation where no single box could possibly handle an entire
database. Their requirements are extreme, but there are much smaller
datasets and applications with similar performance characteristics
that easily saturate a single box, so at a high level the same
principles apply.
http://glinden.blogspot.com/2006/12/talk-on-ebay-architecture.html
The MySpace scaling story is similarly interesting, if mostly because
of the egregious blunders made underway:
http://www.baselinemag.com/article2/0,1540,2082921,00.asp
You could argue that they solved the problems in an inelegant,
irrelational way, but it seems that they solved it, and this is
reflected on their balance sheets, which in the end is probably the
most appropriate metric of success.
Alexander.
