init with event list clears the list

[umontreal-simul]

init with event list clears the list

[chwyean]

Good catch @matthewsowders !

@pierrelecuyer: Do you know why it is coded this way? We can call init() earlier, so the evlist won't be emptied. However, the old eventList will be emptied.

  public void init (EventList evlist) {
      if (evlist == null)
         throw new NullPointerException();
      init();
      eventList = evlist;

Otherwise, we'll have to remove that sentence from the javadoc.

[chwyean]

@matthewsowders : Could you describe/explain the reason you would need to initialize the simulator with a non-empty event list (instead of adding the events after initialization)?
We are trying to assess the need to allow init to keep a non-empty event list, because the events can be added after the initialization. We could create a new method that would allow the user to replace the event list.

[matthewsowders]

I would like to run simulations that will run for 12 hours or more. I would like to make the system fault tolerant so I'd like to be able to serialize the state of the simulator (time, event data and schedule) at certain events. I would store the intermediate state in some service like S3 and use SimpleWorkflowService to manage the long running process that way if the simulation fails 11 hours into a 12 hour simulation, I can take the most recent state and continue the simulation from that point. I've attached some sample code. Let me know if I'm trying to do this a dumb way.

sample-code.zip

[pierrelecuyer]

Matthew: It makes sense that you may want to do checkpoints in your simulations for potential restarts. The role of the current init method is just to empty the event list and reset the simulation clock. The documentation you mentioned is indeed misleading and would have to be changed. To do what you want, it seems you will need a method that clones the event list completely, clone all the events inside, and clone all objects in the simulation. We do not have that at this time, so I think you will have to implement it (which you can do, of course).

…

-- Pierre L'Ecuyer

On 2/13/2017 3:49 PM, Matthew Sowders wrote: I would like to run simulations that will run for 12 hours or more. I would like to make the system fault tolerant so I'd like to be able to serialize the state of the simulator (time, event data and schedule) at certain events. I would store the intermediate state in some service like S3 and use SimpleWorkflowService to manage the long running process that way if the simulation fails 11 hours into a 12 hour simulation, I can take the most recent state and continue the simulation from that point. I've attached some sample code. Let me know if I'm trying to do this a dumb way. sample-code.zip <https://github.com/umontreal-simul/ssj/files/772344/sample-code.zip> — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#5 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKp1oxnxarRkFBYIj8q9H-qvd7iNaJwuks5rcMHmgaJpZM4L8y0l>.

-- Pierre L'Ecuyer, Professeur Titulaire Chaire du Canada en Simulation et Optimisation Stochastique CIRRELT, GERAD, and DIRO, Université de Montréal, Canada http://www.iro.umontreal.ca/~lecuyer

[pierrelecuyer]

I had a quick look at your code and it seems this is what you did: redefine a method that restores the event list to a list that was saved earlier. I suggest this method could be called restoreEventList (...) to avoid confusion with init() that already exists. You can just have a class extension that does this (and perhaps something else...) for you. If you really insist, we could add such a method in SSJ; however we are a bit concerned that could be dangerous for bugs, because one must make sure that all the objects to which the event list refers indirectly haev been saved and restored! Regards,

…

-- Pierre L.

On 2/13/2017 4:54 PM, Pierre Lecuyer wrote: Matthew: It makes sense that you may want to do checkpoints in your simulations for potential restarts. The role of the current init method is just to empty the event list and reset the simulation clock. The documentation you mentioned is indeed misleading and would have to be changed. To do what you want, it seems you will need a method that clones the event list completely, clone all the events inside, and clone all objects in the simulation. We do not have that at this time, so I think you will have to implement it (which you can do, of course). -- Pierre L'Ecuyer On 2/13/2017 3:49 PM, Matthew Sowders wrote: > > I would like to run simulations that will run for 12 hours or more. I > would like to make the system fault tolerant so I'd like to be able > to serialize the state of the simulator (time, event data and > schedule) at certain events. I would store the intermediate state in > some service like S3 and use SimpleWorkflowService to manage the long > running process that way if the simulation fails 11 hours into a 12 > hour simulation, I can take the most recent state and continue the > simulation from that point. I've attached some sample code. Let me > know if I'm trying to do this a dumb way. > > sample-code.zip > <https://github.com/umontreal-simul/ssj/files/772344/sample-code.zip> > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <#5 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AKp1oxnxarRkFBYIj8q9H-qvd7iNaJwuks5rcMHmgaJpZM4L8y0l>. > -- Pierre L'Ecuyer, Professeur Titulaire Chaire du Canada en Simulation et Optimisation Stochastique CIRRELT, GERAD, and DIRO, Université de Montréal, Canada http://www.iro.umontreal.ca/~lecuyer

-- Pierre L'Ecuyer, Professeur Titulaire Chaire du Canada en Simulation et Optimisation Stochastique CIRRELT, GERAD, and DIRO, Université de Montréal, Canada http://www.iro.umontreal.ca/~lecuyer

[matthewsowders]

pierrelecuyer@

It could be made as an extension point such that the client is responsible for passing in a serializer/deserializer (I used jackson, but other users might want java serialization or something else). Then again, it was easy to write so maybe it is best left out of the simulation package. Maybe it would be better as an extension package outside SSJ, something like SSJ-serialization. Do you think it will be useful to others?

[pierrelecuyer]

Matthew: Maybe would be useful to add it, maybe not, not sure. If you want to add and submit an extension you are welcome! What I believe would be even more useful are facilities for simulations running in parallel. Unfortunately, on our side we are very busy with other work. Sincerely, Pierre L.

…

On 2/14/2017 5:19 PM, Matthew Sowders wrote: pierrelecuyer@ It could be made as an extension point such that the client is responsible for passing in a serializer/deserializer (I used jackson, but other users might want java serialization or something else). Then again, it was easy to write so maybe it is best left out of the simulation package. Maybe it would be better as an extension package outside SSJ, something like SSJ-serialization. Do you think it will be useful to others? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#5 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKp1o_9sCOcNXP-yt5W5Rci5g_QVG2XDks5rcih5gaJpZM4L8y0l>.

-- Pierre L'Ecuyer, Professeur Titulaire Chaire du Canada en Simulation et Optimisation Stochastique CIRRELT, GERAD, and DIRO, Université de Montréal, Canada http://www.iro.umontreal.ca/~lecuyer

[chwyean]

I have corrected the javadoc: 2d82af5
It now states that the event list evlist will be cleared. If the user wants to initialize with a non-empty list, then he can extend the class Simulator and implement his initialization method.

There are many ways to manage serialization, and it is likely implementation dependent. It is best left for advanced user to extend the Simulator, which is not too difficult as shown by @matthewsowders.

[matthewsowders]

Thanks to both of you for looking into this!