Performance

How fast @ntcore-ts/client and @ntcore-ts/react can process NetworkTables updates, and how CI catches regressions.

Typical FRC traffic is tens to low thousands of topic updates per second. The numbers below are far above that. The bottleneck in a match is the network and the NT server, not this library.

Headline numbers

Measured with npm run bench on an Apple M1 Pro, Node 24. React benches run in jsdom (flushSync commit), not a full browser layout/paint.

Workload	Rate	Mean
Client, 1 message / frame	2.63M/s	0.4 µs
Client, 10 messages / frame	1.01M frames/s (10.1M msg/s)	1.1 µs
Client, 100 messages / frame	143k frames/s (14.3M msg/s)	7.1 µs
Client, 100 frames through onMessage	16.1k batches/s (1.61M msg/s)	63 µs
React useTopic (one update)	88k/s	11 µs
React usePrefixTopic (one update)	85k/s	12 µs
React 50-hook dashboard (batched)	4.5k/s	240 µs
React 50-hook dashboard (unbatched)	800/s	1.3 ms

Decode of a double is about 0.4 µs. Applying it through useTopic is about 11 µs. Batching 50 hook updates into one React commit stays under 250 µs mean.

What is measured

Client (packages/client/src/lib/socket/*.bench.ts)

• Hot path: handleBinaryFrame — msgpack decode, lightweight binary shape check, onTopicUpdate callback.
• onMessage: 100 binary frames through the WebSocket message handler.

React (packages/react/src/lib/hooks.bench.tsx)

• Subscribe callback → setState → React commit in jsdom.
• Values change every iteration so React cannot skip the render.
• The client is mocked; this is hook + render cost only.

Not included: real TCP/WebSocket, NT server time, struct/protobuf decode, browser layout and paint.

Running locally

npm run bench

Package-only:

npm run bench -w @ntcore-ts/client
npm run bench -w @ntcore-ts/react

CI regression checks

Every push to main and every pull request runs .github/workflows/bench.yml under CodSpeed.

• Benchmarks run in CodSpeed's CPU simulation mode, which counts simulated CPU cycles instead of wall-clock time, so results are stable on shared GitHub runners.
• CodSpeed comments on each pull request with the change against main and reports a status check that flags regressions above the configured threshold.
• Pushes to main record the baseline and history shown on the CodSpeed dashboard.

Simulation runs V8 without its optimizing compilers and with a fixed GC schedule, so it tracks how much work each path does rather than optimized-JIT or GC behaviour. Absolute numbers will not match the M1 Pro table above. The plugin does nothing outside CodSpeed's CI, so npm run bench locally is unchanged.