Performance design

Concern	Decision	Effect
Robot flooding	500ms window per connection; invalid frames also consume it	At most 2 frames per second per robot reach Redis
Redis round trips	One multi field HSET per 250ms batch	4 writes per second regardless of fleet size
Snapshot latency	Hash values are stored as JSON and joined without parsing	1,000 robots in single digit milliseconds
MongoDB volume	Persist changes plus a 10 second heartbeat	About 70 documents a second instead of 1,000
MongoDB latency	Unordered insertMany, fire and forget	Persistence never blocks the live path
Dashboard fan out	One frame per batch per client; clients over 1 MB buffered are skipped	Server memory stays bounded
React reflow	Frames buffered, committed every 300ms	About 3 commits per second
Map rendering	WebGL, clustered GeoJSON, refit only on filter change	1,000 markers without dropped frames
Table rendering	TanStack Virtual with overscan 10	Only visible rows exist in the DOM
Reports	Aggregation over the history window, joined to inventory in memory	Sub second for one hour of data