[Ns-developers] Internal Time Keeping Question

[Hagen Paul Pfeifer]

* Tom Henderson | 2008-07-22 17:57:37 [+0000]:

>But will you get any good lookahead performance?  I mean, you are relying on
>the user putting nodes into containers that will result in good performance,
>but the user may not know that he or she needs to be conscious of this.  This
>is kind of an implicit API that is arguably not a feature.  Plus, nodes may
>exist in many containers simultaneously, for different configuration reasons.


Joseph likes the idea! ;) No just kidding. The idea was that this is a
mechanism with no impact to the scenarios (one big point). The other aspect I
thought was if the user simulate a scenario which is small then the impact
should be not that much (e.g. eventually he/she does not even enable
parallelization). BUT: if the user in knowledge of this mechanism he can put
nodes in one container and boost the simulation. And, from my point of view,
the probability of nodes that are grouped together in NodeContainers are
communicate much more, then with nodes from other containers was a naturally
assumption. It can be false, sure.

Tom, I searched for similarities to build federates, without user
interferences, then I stumbled across NodeContainers and thought, ... perfect
suited!  (or something like that)

As George said in the other mail. Maybe we can add an additional functionality
to let this be decided in a manual manner. Sure, I had no complains for tips,
additional methods - in opposite. If you see the code where the node federates
are build you will see some ideas as comments how to build federates. There
could be some of them. We must consider a trade-of between
user-friendly-but-effective and user-effort-but-highly-optimized - where the
first will be not that effective in all scenarios. ;-)

Hagen