Re: multimaster (was: Slightly OT.)

On Fri, Jun 01, 2007 at 12:39:42PM -0400, gonzales(at)linuxlouis(dot)net wrote:

> ever need? I say yes, I can. BUT it would so much cooler IF,
> there was a multi-master environment configured for the sake of
> doing it.

Yes, it would be. Unfortunately, when I was in the position where I
had to explain to my bosses that we still didn't have any
demonstrable progress after a couple of years of effort and expense,
they told me they weren't willing to continue funding that work. I
presume other people had similar experiences with their bosses.
Moreover, because we had something else that solved our _particular_
problem -- i.e. zero-transaction-loss failure recovery within our SLA
-- it was hard to justify the continued work.

> I'm glad that not everyone has this disposition about 'why do we
> need it.' The pursuit of knowledge is the facilitator of
> innovation.

Others have argued that necessity is the mother of invention. I have
no idea what causes better software in general; what I know is that,
for companies that are likely to spend money to pay people to work on
multimaster replication, a strong need and evidence of progress are
two necessary conditions. Bruce mentioned (at PgCon in Ottawa) that
one of the UCB people, in handing the project off, said that Postgres
needed a few people with a lot of time rather than a lot of people
with a little time. Multimaster replication is like that, and
therefore finding someone who will work on it (either because they're
paid to or because they are incapable of leaving the problem alone)
is going to be the requirement to get it done.

> Why? Because I'd like to.

My point was really to ask, "How much do you want it?" If the answer
to that is, "Enough to use it when someone else shows up with the
goods," then I'm trying to suggest that that's not enough to cause
the happy result you desire. Someone needs to do the work, which
means that someone has to want the result badly enough to put the
time and talent into getting the result.

> would not be deterred from using Postgresql because it doesn't have a
> quote unquote, 'multi-mater' replication.

It is certainly true today that if what you really need is RAC, you
have exactly one company to go to, and that's Oracle. If you need
something else, there might in fact be a Postgres answer for you.
Part of the problem here is that people often say "multimaster"
without stating what they're asking for.

Does MySQL have multimaster replication? Well, yes and no, depending
on which set of features you might want to use and what data breakage
you're willing to tolerate.

What about SQL Server? See above.

Ok, what about DB2? Well, no, not really, but it scales well and you
can do HA with it using hot-standby failover clusters at the OS
level, if you believe their literature. There's also some component
that sort of hangs on the side and does some data synchronisation
between points, allowing something that looks like multimaster but
sounds like it has some nasty gotchas (and I get that feeling just
from reading the brochures. OTOH, I am not completely rational when
it comes to claims made by IBM). You can in fact build a Postgres
system that does the same things today, though.

These are all different solutions to different problems, so it's not
surprising that they look different. This was the reason I asked,
"What is the problem you are trying to solve?"

A

--
Andrew Sullivan | ajs(at)crankycanuck(dot)ca
Unfortunately reformatting the Internet is a little more painful
than reformatting your hard drive when it gets out of whack.
--Scott Morris