We’re facing a classic trade-off between device shadow visualization accuracy and dashboard performance. Our devices update their shadows every 2-3 seconds with state changes, but visualizing these updates in real-time causes severe dashboard lag with 2000+ devices. If we batch updates or reduce frequency, we lose accuracy that operators need for critical decisions. How are others handling this balance? What’s the right shadow update frequency for visualization? What batching and delta update strategies work well without sacrificing too much accuracy? Interested in perspectives on dashboard performance versus accuracy trade-offs.
We solved this with client-side batching. The dashboard receives all shadow updates but only renders every 5 seconds. Updates are queued and then applied in a single batch render cycle. This keeps the dashboard responsive while ensuring no data is lost. Users see slightly delayed updates but the UX is smooth. For critical alarms, we bypass batching and render immediately.
Dashboard performance at scale requires virtualization. Don’t render all 2000 devices - use virtual scrolling to render only visible rows (typically 20-30). Implement efficient diffing algorithms so you’re only updating DOM elements that actually changed. Use React or Vue with proper key management to minimize re-renders. With these optimizations, we handle 5000 device shadows updating every 3 seconds with minimal lag.
The accuracy vs performance trade-off depends on your operational requirements. For safety-critical systems, accuracy can’t be compromised - you need the latest shadow state immediately. For general monitoring, 10-second latency is acceptable. We use tiered update frequencies: critical properties (alarms, safety status) update every 1 second, operational properties (temperature, pressure) every 5 seconds, informational properties (runtime hours) every 30 seconds. This prioritizes what matters.
The client-side batching idea is interesting. How do you determine which updates are ‘critical’ and should bypass batching? Also, with 2000+ devices, even 5-second batch renders seem like they’d cause UI freezing. Are you limiting the number of visible devices or doing something else to manage the render load?