Dashboards and Visualizations¶
Metrics, dashboards, and visualizations are native to PipelineAI.
We believe that you should have full insight into everything deployed to production.
PipelineAI provides the usual dashboards for real-time system metrics including memory usage, disk I/O and network throughput, CPU and GPU utilization of your model training and deployment activities.
In addition, PipelineAI provides dashboards for real-time prediction metrics including accuracy, latency, and throughput of your models in production.
Monitor and Optimize Model Training¶
Stabilize the Model Server Cluster¶
Unhealthy or latent model servers may open a circuit, respond with a suitable fallback, and allow the cluster to stabilize.
Large batch sizes provide higher throughput at the expense of latency. PipelineAI dynamically configures the system to find the proper balance.
Control Latency with Timeouts¶
High latency may lead to unhealthy model servers if left unbounded. All PipelineAI service calls are bound with timeouts.
Monitor and Alert¶
High resource utilization - beyond container and physical node limits - will certainly degrade performance. PipelineAI monitors all system resources.
All PipelineAI services support auto-scaling across federated cloud and on-premise environments.
Register for PipelineAI Enterprise Edition¶