Re: [sumo-devel] SUMO simulation engine integration

[Björn H. <hen...@ib...>]

Hi Artem,

On Wednesday 17 April 2013 11:05:38 you wrote:
> I am sorry for writing on your email address directly, but it seems
> there is no possibility for me to answer the thread on the "sumo-devel"
> list.

AFAIK there is no technical difference between posting to a mailing list and 
posting to a personal address. But maybe the list owners know more.

> I think the virtual machine approach will not work for us. A usual
> workflow of ours assumes handling dozens or hundreds of simulation runs
> at a time. Operating that many virtual machines might be tedious.

It depends on how different the installtions need to be. Maybe it would be an 
option to set up one virtual machine and then simply clone it as often as 
needed. You could also run the simulation up to your fork point in one machine 
and then clone it to several machines.

> Regarding the second possibility you mentioned, could you please clarify
> that a bit? Do you mean that, in order to continue a simulation after
> interruption at a certain point, one would need to re-run it from the
> very beginning with the same initial conditions and seed? Or do you mean
> that it would be possible to save a state at the interruption point and
> use it as a new initial condition with the same random seed?

Yes, you need to re-run it from the beginning, but if you apply a fixed random 
seed it will produce exactly the same output as long as you have the exactly 
same configuration. I haven't tried this, but it might be possible to configure 
SUMO such that the configuration settings only diverge from a fork point on, 
such that the simulation before the fork point is always the same. Just try it 
out.

Cheers
Björn




> >	Hi Artem,
> >	
> >	On Wednesday 10 April 2013 18:14:21 wrote Artem Konev:
> >	> *What we are particularly concerned with is the possibility to save
> >	> simulation states, modify them and continue the simulation from a
> >	> just-loaded state. According to **this
> >	
> >	As a low-level solution you can run SUMO in a virtual machine and save
> >	the state of the whole virtual machine. Some time ago I've tested that
> >	there is nearly no performance loss due to the virtual machine.
> >	
> >	External (from outside the virtual machine) control of SUMO by TraCI
> >	interface should still be possible, because TraCI runs over a TCP
> >	connection.
> >	
> >	If you put configuration and result files of SUMO into a folder shared
> >	with the host it would be easy to access these files from outside.
> >	
> >	Other than that, if you run SUMO with a fixed random seed (which is the
> >	default) every run with the same configuration settings reproduces
> >	exactly the simulation. Thus, it might be possible to set up a
> >	configuration which only differs at a later simulation time, such that
> >	the start of the simulation is reproduced on each run. But, I didn't
> >	tested such an approach.
> >	
> >	Cheers
> >	Björn