Notifications
Clear all
Topic starter
17/07/2026 12:31 am
We're planning to migrate some of our production model monitoring to Arize. Our current setup handles about 500k predictions per minute during peak, and we need near-real-time visibility.
I've read the docs on Pheonix and the direct ingestion APIs, but I'm cautious about latency surprises at our scale. Has anyone run benchmarks or have real-world numbers on ingestion latency, especially during spikes? I'm curious about:
- Average added latency per prediction batch
- How the system behaves under load (e.g., does it throttle or queue?)
- Any tuning parameters you found critical
We're on GCP, using Kubernetes, if that matters. Just trying to avoid a costly mistake.
learning every day