Milvus outputs detailed time-series metrics during runtime. You can use Prometheus, Grafana, or any visualization tool that you think appropriate to display and render the following metrics:
Milvus Performance Metrics
System Performance Metrics: Metrics relating to CPU/GPU usage, network traffic, and disk read speed.
Hardware Storage Metrics: Metrics relating to data size, data files, and storage capacity.
Milvus performance metrics
Metric
Description
Insert per Second
Number of vectors that are inserted in a second. (Real-time display)
Queries per Minute
Number of queries that are run in a minute. (Real-time display)
Query Time per Vector
Average time to query one vector. Divide the query elapsed time by the number of queried vectors.
Query Service Level
Query service level = n_queries_completed_within_threshold1 / n_queries Generally, it is recommended to set 3 time periods - threshold1, threshold2, and threshold3, to track the query service level.
Uptime
How long Milvus has been running. (Minutes)
System performance metrics
Metric
Description
GPU Utilization
GPU utilization ratio (%).
GPU Memory Usage
GPU memory (GB) currently consumed by Milvus.
CPU Utilization
CPU utilization ratio (%). Divide the time that the server is busy by the total elapsed time.