GoKawiilNew open-source library Mizu enables zero-copy Python-R parallelism via shared memory
A new project called pymizu, built on the C library libmizu, offers lock-free inter-process communication for Python and R using shared memory instead of sockets. It claims task round-trip times under 1 microsecond, down from roughly 100 microseconds over sockets, and supports zero-copy sharing of data structures like numpy arrays, dicts, lists and Arrow tables between Python and R. The project is described as experimental software, available via PyPI or GitHub, requiring Python 3.10+ on 64-bit platforms.
GoKawiil's interpretation of the reporting above, not reported fact.
Python's Global Interpreter Lock forces CPU-bound parallelism into separate processes, which traditionally pay a steep communication cost; faster shared-memory IPC could let developers parallelize at a much finer grain, closer to how multithreading works in other languages. This could appeal to teams mixing Python and R in data science pipelines, since it promises zero-copy interoperability between the two ecosystems. However, as an experimental, pre-1.0 project, its API and performance claims are not yet proven at scale in production environments.
- pymizu claims sub-microsecond task round trips versus ~100µs for socket-based IPC.
- It enables zero-copy sharing of numpy arrays, Arrow tables, and other structures between Python and R processes.
- The library is explicitly labeled experimental, with an unstable API subject to change.
Source: github.com, 2026-11-05
Published there as: “Mizu – zero-copy parallel R and Python over shared memory”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.