[Caffe2] What are the extra workers for in the data parallel model?

Hi,

I wonder…

  1. What are the extra workers for in Caffe2’s data parallel model?

  2. Why is extra_workers set to 8 for rendezvous?

Here is the link to the Python code that I am talking about : https://github.com/caffe2/caffe2/blob/master/caffe2/python/data_parallel_model.py#L137

Thanks!