[erlang-questions] Unreal engine

[Ulf Wiger (TN/EAB)]

James Hague wrote:
> 
> "One game object per process" gets mentioned
> occasionally, and I think that's not as beautiful
> in practice as it sounds. 

I think you're right. There should of course be more
than one process per actor to start with. (:
I also like the suggestion I saw on LtU to have one
process per bullet...

Seriously, it's a bit like the telecoms analogy
"one process per phone call". We've actually played with
lots of different models:
- one process per half-call
- process model following the H.248 model with context
  and terminations
- having just a context process per call, with the
  terminations handled as attributes, rather than 
  as separate processes.
- everything from one process for all calls to 
  10 processes per call.

There are pros and cons with each model, and which
one is best depends to some extent on what kinds of 
call we're talking about. SIP calls are much different
from ATM UNI point-to-point POTS calls, which are 
different from ATM PNNIR point-to-multipoint calls
(Mats C's absolute favourite.)

In short, you just have to take a deep dive into the
domain, and really understand the semantics, before 
you can come up with a good process model.

BR,
Ulf W