Re: [Nel] Suggestion for the NeL network library / architecture

[MIGUEL ANGEL BLANCH LARDIN]

> That's where PGM/multicast intervenes. In that model, the agent, 
when he
> changes some of his variables, notifies the replica that "XYZ are 
now...".
> I spoke about how the world services organise where an agent 
resides. In
> some models, you have about one, maybe two replicas, which you can 
notify
> using a standard TCP stream. Or RDP. But, if you want a 
load-balancing
> system, you quickly end up with a replica of an object in most world
> service processes. It becomes then more efficient to multicast the 
updates
> as above to all processes, and update them all in one sweep.

Ok, I always have though of load balancing as load-work division 
between servers.

I think that the above kind of working can have several inconsistences 
, just take a look to some distributed Mutual exclusion algos.

Just imagine that both copies of a object multicast different 
positions, you have to choose what is the best one, and whatever your 
choose is, it will be wrong, and servers can't be wrong.

Replycating work can be a real pain.
And for what you have said multicast isn't a solution for end-users.

I think that IPv6 will fix these things, by having a multicast in the 
standart, isn't it?