I am using ask and tell to optimize a simulation which takes days to run in parallel batches of 3. Is there any significant difference between the following two workflows?

1. Every time I have new evaluations to add, just create a new Optimizer, tell all the existing evaluations, and ask for 3 new points.

2. Create the Optimizer once. Every time I have new evaluations to add, restore the pickled Optimizer, tell the new evaluations, ask for 3 new points, and pickle the Optimizer again.