Smaller (and lazy/persistent) states
graydon opened this issue · comments
The system state representation in the actor subsystem is relatively large and eagerly copies vectors and btrees full of contents when cloned. This makes it relatively memory-hungry, which in turn limits the size of the state-space it's possible to explore.
How would you feel about changes that reworked the representation to use lazy/persistent types (that copy-on-write only the substructures written) like those in the im crate? They are a bit slower than the eager types but I think they can take up substantially less room.
I think it’s worth exploring, and I’m curious to learn how the performance differs for your use case.