A discussion on adapting shared-memory concurrent data structures like hash maps into actor-based distributed designs, motivated by work on the Servo browser engine in Rust. The post explores several approaches: striping a hashmap across tasks (one per core), dynamic rebalancing via task splitting, and self-building hashtables for distributed CSS matching. It also touches on a concurrent B-tree where each node is an actor, noting that locking becomes trivial in this model. The author reflects on the advantages of avoiding shared mutable memory for correctness and security, and expresses interest in prototyping these designs in Rust.
