[Ns-developers] Internal Time Keeping Question

[Hagen Paul Pfeifer]

* Tom Henderson | 2008-07-22 08:50:15 [-0700]:

> I think it would help to bump up a level and clarify exactly what is the  
> end functionality goal of the effort, what are the requirements on the  
> ns-3 core for your effort to succeed, and what will be the API impact on  
> users.  For instance, you state in your project description that your  
> main goal is CPU local parallelization but it seems that your emphasis  
> is on MPI recently.

o "What are the requirements on the ns-3 core for your effort to succeed"?

	Until now there is nearly no impact in the core. There are some wishes to
	support the parallelization even smoother. But I focused to reduce the
	impact.

o "what will be the API impact on users"?

	Only the mentioned flag to enable parallelization and the MPI wrapper.
	But as I wrote in another email, I kept in mind that the underling
	synchronization mechanism is substitutable. So far I see _no_ cause why not
	to replace MPI with a shared memory approach (this means limited to one
	processor, and implementation of barriers based on locking mechanisms, sure)

o "CPU local parallelization but it seems that your emphasis is on MPI
  recently"

	That was one advice from George and currently I see that this is the only
	way to go. The advantage is that MPI is ready to use. But more important:
	MPI forces me to parallelize the simulator completely! Consider OpenMP,
	there are only some minor advantages for ns-3, because there are _really_ no
	big chunks to parallelize if you do this in this manner. I also spent some
	time to utilize OpenMP but I see no usage for us! IMHO, MPI is the way to
	go, after MPI is working we can think about it to replace this with a other
	synchronization technique. (BTW: the bulk of parallelized approaches are
	based on MPI, if not, they implement their own synchronization mechanism)

> After that, we should identify the paths to mergeable code within the  
> project timeframe.

Sure! I am in confidence that at the end we had a working ns-3. There are open
question but the interplay parallelization vs. ns-3 isn't that simple.

Hagen