Performance

Stress-tested with 1000 simultaneous bot users on a single board, all moving cursors.

Stress test results

Key optimizations

• Canvas 2D instead of SVG for cursors (zero DOM diffing per frame)
• Bitmap label cache - user names rendered to offscreen canvases once, then drawImage() per frame
• rAF cursor throttle - outbound cursor moves coalesced to one per animation frame
• Per-topic broadcast budget - server caps cursor broadcasts at ~60/sec per board regardless of user count
• RAF event batching - incoming WebSocket events coalesced into one Svelte store update per frame
• Batched publish - arrangement actions publish all note updates + activity in a single ctx.batch() call instead of N+1 individual publishes
• Batch SQL - arrangement actions update all notes in a single unnest() query instead of N+1
• Direct DOM drag - note dragging bypasses Svelte reactivity during the drag for smooth touch performance
• Delayed handoff - local drag position held for 300ms after release to avoid snap-back jitter

For OS-level tuning (sysctl, ulimits, conntrack), see the svelte-adapter-uws production docs.