GSL PRNG seed: separate seeds for Libmboard and FLAME simulator, or a single one

[svdhoog]

@shawnchin @sysfal @mariamkiran
Some issues related to the GSL PRNG seed:

1 Is it useful to have Limboard and the simulator use the same seed? Libmboard introduces randomness if you use the tag for message randomization in model.xml. Since Libmboard is executing on its own thread, it also initializes its own random seed. It currently uses srand() for this, which of course is not so good. I suppose that for reproducibility the same seed should be used for both Libmboard and the simulation, or both seeds should be stored in the 0.xml file.

2 Is it useful to have each agent have its own random seed? Or even initialize its own PRNG? This is an old discussion that Rob Axtell once brought up about synchronization effects if all agents use the same seed, and then essentially sample from the same random number sequence.

[sysfal]

1 For reproducibility when using a seed all the rand() calls would have to happen in the same order. I'm not sure FLAME can guarantee this although I can see it happening. This probably goes out the window when running in parallel though!
2 Yes for reproducibility all agents would need their own seed and their own RNG!

[svdhoog]

1 Since Libmboard executes on its own thread, is it at all possible to store that random seed? Or to link it to the random seed that the simulator thread uses?

2 "all agents would need their own seed" would imply we need to make the seed a "special" memory variable for each agent, similar to the ID variable, such that it is stored in the xml.

3 when a function is run for all agents that need to run it (same function), the order in which it gets activated is one time ascending in the agent IDs, the next time it is descending. Is there any reason why the x-machines iterator activates agents in this way?

[sysfal]

1 Need to ask Shawn Libmboard questions
2 Yes
3 I can't remember why it works that way

One way to get agents to always use the same random numbers could be to update the seed variable with the last random number and just before calling rand again set the seed using the saved number. This way each agent would always be using their own seed to get new random numbers?