MATLAB Engine API for Python: Persistent Session vs. Per-Call Startup for Low-Traffic Workloads

Integration Boundary

The MATLAB Engine API for Python connects a Python client process with a MATLAB runtime session. For workloads that invoke MATLAB functions infrequently, the session lifecycle becomes a design decision at this boundary.

Goal and Constraints

Determine whether to maintain a single persistent MATLAB Engine session across sparse calls or to start and terminate a session for each call. MATLAB desktop licensing is user- and core-based, not traffic-based, so cost reduction does not come from session count. However, an idle Engine session retains a MATLAB process, holding memory and CPU resources. Conversely, repeated cold starts add latency (typically several seconds) and may increase license checkout frequency.

Open Questions

1. What is the measured cold-start latency for matlab.engine.start_matlab in the target MATLAB release and Python version?
2. How much resident memory does an idle Engine session consume on the deployment platform?
3. Does the licensing model in use impose any per-session checkout limits that would affect high-frequency teardown?