LabVIEW Benchmark Actor
Reproducible, cross-plane LabVIEW benchmarking — coordinated by an actor mesh, driven by agents, with no central data hoarding.
A free, open-source VS Code extension for the LabVIEW CI/CD community. It turns a personal, sandbox-isolated "golden" VM (Ubuntu 24.04 + LabVIEW 2026 Community Edition + VIPM) into a benchmark actor that runs headless LabVIEW benchmarks, coordinates on-demand distributed runs with other actors, and produces portable, fail-closed receipts you own — so results can be objectively compared across operating systems, hardware, and LabVIEW versions.
Everything is governed by construction: every capability is proven on real data with a committed receipt and a fail-closed gate (180+ gates on develop), and the requirements ⇄ tests ⇄ architecture decisions stay in a checked correspondence graph (ISO/IEC/IEEE 29148 / 42010 / 29119).
What it does
- Benchmark time-cursor viewer — a metric chart with a draggable vertical time cursor; drag it and the captured frame indexed at that instant is shown below, keeping the metric and the visual evidence synchronized. Capture is frame-locked at exactly 12 FPS — the shared clock across every plane.
- The actor mesh — register your golden VM as an actor; a requester dispatches a cross-plane benchmark run GitHub-natively (
repository_dispatch / Actions as the queue — zero central infrastructure), and volunteer actors return plane-tagged receipts. A run is fulfilled only when enough independent cross-plane actors returned a valid receipt for the same benchmark identity. No central results database — the receipts are the result.
- Verify-before-consume trust — an opt-in verified tier signs each actor receipt with an enrolled key, records it in an RFC-6962 transparency log, and proves the log is append-only. A single fully-attested verdict tells a consumer a run is trustworthy end-to-end (identity + signature + transparency inclusion + append-only).
- Cross-plane comparison at scale — reproducible parity + comparison receipts across the OS and hardware axes, with the mesh-stress calibration discounting results captured on a contended actor so comparisons stay fair.
- Coordination bus (
lbabus) — actors on a LAN coordinate over a local TCP + UDP message bus; the bus carries inter-actor messages only — run data never crosses it.
- MCP tools for agent mode — the extension contributes a dependency-free Model Context Protocol server so Copilot agent mode can call its tools directly (host capabilities, the deterministic benchmark series, the coordination bus).
Install
Install LabVIEW Benchmark Actor from the VS Code Marketplace (publisher svelderrainruiz), or from a packaged .vsix. Requires VS Code 1.101+. The extension is fully free and non-commercial; running LabVIEW benchmarks happens locally in your own activated golden VM.
Quickstart
- Provision a personal golden VM — one command mints a clean Ubuntu 24.04 VM, installs LabVIEW 2026 Community Edition + VIPM from NI's apt repo, and hands off to you for the one irreducibly-human step: signing in to your NI account and activating LabVIEW CE. The tool then confirms activation with a headless probe VI and mints your local golden VM.
- Register it as an actor — the activated VM is registered in your local actor registry; it stays local to you (no boxes are published to a shared registry).
- Join the mesh — follow the join-the-mesh quickstart to answer a dispatched run and return your first plane-tagged receipt.
Full walkthroughs: the getting-started guide, the user guide, and the command reference.
Principles
- Reproducibility over telemetry — the mesh coordinates runs, not a central results database; evidence is a portable receipt owned by whoever produced it.
- Real data, no fakes — every capability is proven on a real capture with a committed receipt and a fail-closed gate.
- Sandbox = the golden VM — distributed runs execute inside your isolated VM; provenance/attestation is an opt-in stronger tier, not a gate on participation.
- Hybrid labor — agents drive the automation; a human does the irreducibly-human steps (the NI-account activation, hardware, physical provisioning).
- Free + non-commercial — LabVIEW Community Edition is the substrate; the extension ships free.
How it's governed
The project is developed as fail-closed, receipt-backed increments. Requirements, tests, architecture decisions, and the executable gate suite stay in a checked correspondence:
The multi-year vision and near-term slice live in the roadmap.
Absorbed model: the bounded-RAM mprr ring-buffer model (dual-packet policy + a frozen replay transport) is absorbed in-repo as dependency-free mirrors — the project owns it and tracks no external repository (ADR-0009).
Standards baseline: authored against repo-standards-review v0.2.19 (commit d44f210d).
Contributing & support
Community, non-commercial project. Not affiliated with or endorsed by National Instruments; "LabVIEW" is a trademark of its respective owner.
