Re: Experiences with extensibility

On Tue, Jan 08, 2008 at 10:59:56PM -0700, Guido Neitzer wrote:
>
> Easy multi-master clustering with just two machines.

To my knowledge, _nobody_ actually offers that.

There are three companies I know of that have done effective marketing of
systems.

Company O has a very advanced system with plenty of features. When you see
it functioning, it is very impressive. Casual acquaintance with anyone who
has attempted to implement it, however, will yield many stories that give
the lie to any claims of "easy" multi-master. Some implementors would be
happy to get to "hard to do, but working" multi-master, as far as I've been
able to ascertain.

Company M has a nifty 80% solution for in-memory clustering. It's a cool
hack. But it has a remarkably large number of failure modes and corner
cases that make it a risky answer for really high-value data. If my bank
were using this technology, I would transfer my money to another bank.

Company I actually has the most mature technology in this area, if you're
willing to use VMS. It relies on the features of VMS to do this. Given
that those features were delivered precisely for the finance and insurance
industries where extremely valuable data was being stored, there is a long
history of reliable field deployments. Unfortunately, the continued life of
VMS is in some doubt, and skilled VMS operators are increasingly hard to
find and expensive.

There are other systems, including PostgreSQL, that can do a kind of
"clustering" with multiple machines, shared disk, and some sort of heartbeat
arrangement.

A