The talk will focus on a novel idea of efficient structured data sharing between Python subinterpreters. I'll be presenting a new Python framework "memhive", which implements a worker pool of subinterpreters, efficient RPC mechanism between them, async/await-ready API, as well as fundamental data structures ( tuples, mappings and scalar types).

I'll demonstrate that it's possible to unlock true parallelism with subinterpreters without paying the overhead of data serialization. I'll explain how this is possible and what algorithms under the hood drive this.