Re: Concurrency in an RDB

Sampo Syreeni wrote:
> On 2006-12-12, David wrote:
>
> > It is kind of like an optimistic concurrency control except that the
> > mathematics of OT allows for convergence to be achieved without any
> > concept of conflict, even though different sites apply operations in
> > different orders.
>
> How is that different from compensation processing, as applied in the
> context of semantic concurrency control, and more specifically
> level-specific conflict resolution in the multilevel and open nested
> transaction models? As I've understood it, that sort of thing is in
> common use in transaction processing environments, e.g. in the banking
> industry.

I've had a (very) brief look at the paper "Semantic Concurrency Control in
Object-Oriented Database Systems" by Muth et al. Is this representative? In that paper they look for methods on objects that commute and use this to provide a locking protocol with increased concurrency.

That has some similarities to OT. However OT assumes that *all* changes on the data (however simple or complex) can be represented by operations that can be applied in different orders at different sites in order to synchronise replicated data. It is not about providing increased concurrency on top of a single, centralised DB with (ultimately) a pessimistic locking model.

OT doesn't limit itself to commuting operations (say). In fact this is a trivial case (albeit rather common) where no operation transformation is required at all.

Cheers,
David Received on Fri Dec 15 2006 - 14:59:04 CET