Re: Concurrency in an RDB

Pickie wrote:
> David wrote:
> <snip>
> > My claim can't easily be proven true, but it should easily be falsified
> > if it turns out to be wrong. You haven't yet provided an example.
>
> This sounds reasonable, but it might not be. How many colours are
> required for a map?

My point was that it is common for universally quantified assertions to be difficult to prove but easy to refute if a counter example can be found.

> <snip>
> > I don't. I merely claim it is desirable (and in practise possible) to
> > use fine-grained mutative transactions.
> <snip>
> > My statement was quite simple: Distributed transactions (in contrast
> > to in-process transactions) have large overheads and therefore issuing
> > many fine-grained transactions instead of fewer coarse-grained
> > transactions can degrade performance.
> <snip>
>
> I'm confused. Are fine-grained transactions good or bad?

Fine-grained transactions are good until you use distributed transactions. That doesn't worry me because IMO distributed transactions are a poor approach that can and should be avoided most of the time. This is not the conventional viewpoint. That may change one day if Operational Transform proves to be useful for replicated databases.

Cheers,
David Received on Wed Dec 20 2006 - 02:36:12 CET