Re: Proposal: Commit timestamp

On Feb 8, 2007, at 8:41 PM, Richard Troy wrote:
> It'd be great if Jan considers the blending of replication; any  
> given DB
> instance shouldn't be only a master/originator or only a slave/ 
> subscriber.
> A solution that lets you blend replication strategies in a single  
> db is,
> from my point of view, very important.


It might be constructive to define what a minimal "complete" set of  
replication primitives actually is in addition to which ones should  
be implemented.  In addition to master/slave models, you have Paxos  
algorithms and dynamic reconfiguration models in literature that can  
utilize many of the same primitives but which are very different in  
implementation.  I see the value of Jan's proposal, but perhaps it  
would be better to step back and make some assertions about the  
nature of the core capabilities that will be supported in some  
broader picture.  Having a theoretically (mostly) complete set of  
usable primitives would be an incredibly powerful feature set.

Cheers,

J. Andrew Rogers
jrogers@neopolitan.com