An Analysis of Python Serialization towards Distributed Systems

We explored the current state of Python serialization in the context of distributed computing. The challenging points addressed are: object-type support, speed and size, version compatibility, and implementations. There are three well-known serialization modules in our analysis. They are: standard python pickle, cloudpickle, and dill. We found that while the current serialization modules work well, there are certain fundamental issues on __reduce__ method that can be addressed. There is also a room for improvement that can have significant impacts for the community.