Hi,
I am noticing that in every round, the simulated from one collaborator to the other is always increasing. A quick example:
```
INFO Collaborator 19 took simulated time: 7.64 minutes experiment.py:476
...
INFO Collaborator 17 took simulated time: 9.1 minutes experiment.py:476
...
INFO Collaborator 14 took simulated time: 14.33 minutes experiment.py:476
...
INFO Collaborator 8 took simulated time: 15.87 minutes experiment.py:476
...
INFO Collaborator 22 took simulated time: 16.06 minutes experiment.py:476
...
INFO Collaborator 11 took simulated time: 16.52 minutes experiment.py:476
...
INFO Collaborator 10 took simulated time: 18.09 minutes experiment.py:476
...
INFO Collaborator 2 took simulated time: 20.04 minutes experiment.py:476
...
...
INFO Collaborator 1 took simulated time: 1554.36 minutes experiment.py:476
```
It seems a little weird. Is the order of local trainings defined by the simulated time, or can it be a bug in the computation of the overall simulated time of a round?
All the best,
Matthis.
Created by Matthis Manthe Matthis @Matthis That seems to be intended behaviour indeed, I am getting the same. Hi,
Thank you for your reply, this is much clearer.
However, I am wondering if the simulated times I am getting are normal. Collaborator 1 always gives around 1550 minutes of training for one epoch, so around 1 day of training, which seems extremely high. Is it normal/do you observe the same behaviour?
All the best,
Matthis. Hi Matthias,
The simulation code samples from historically based distributions of times taken to train (in real world FeTS training), assigns the time to each collaborator, then orders collaborator training according to the simulated time with which it took to train. So what you notice is by design.
Let me know if this is not clear.
Brandon