Metering and Billing
Describes metering and billing in HPE Ezmeral Unified Analytics Software.
Metering involves the measurement and collection of metrics from monitored targets. The metering process is essential for billing purposes and to monitor the usage of key resources within the Kubernetes cluster, such as pods, PVCs, and jobs. Prometheus facilitates the monitoring of these components.
Prometheus is an open-source monitoring and alerting system designed for gathering time-series data. Prometheus provides a flexible querying language (PromQL) to retrieve and analyze metrics.
HPE Ezmeral Unified Analytics Software consists of platform components and application components such as Kubeflow, Airflow, and others. In HPE Ezmeral Unified Analytics Software, metering refers to tracking the resource usage (in particular, the CPU and GPU usage) of the components. Metering is used to support accurate billing and to observe the aggregated usage of these components over a specific time period.
- Application core workloads
-
The application core workloads are initialized by the application installer. These workloads allow the instantiation of user-initiated workloads, such as notebooks and jobs. Some examples of these workloads include notebook controllers, EzPresto core querying workloads, and inference job controllers.
- Application user workloads
-
Application user workloads refers to the user-initiated workloads. Some examples of these workloads include notebooks, inference jobs, and Spark jobs.
- Infrastructure or platform workloads
-
The platform workloads perform core-platform functions such as monitoring UI, managing users, connecting to Data Fabric, and so on.
- Bring your own application workloads
-
Bring your own application workloads refers to the workloads that are imported using Import Applications functionality. You must manually specify the labels for workloads in the workload resource yaml files.
Configure the resource metadata with the following labels:hpe-ezua/type="vendor-service"hpe-ezua/app="<name_of_the_app>"
Pod labels are used to categorize each workload into the four types of workloads. CPU usage is measured every five seconds. After obtaining the initial data, application-related pod labels are used to retrieve the aggregate application usage. For example, the total Kubeflow usage can be calculated by summing up the usage of all pods associated with Kubeflow.
Billing and Metering
In HPE Ezmeral Unified Analytics Software, billing is performed on an hourly basis. A cron job runs at the top of each hour to compute the average of the aggregated application usage. Daily and monthly aggregations are also calculated, and custom metrics are pushed back to Prometheus. The billing feature leverages these custom metrics to calculate monthly charges.
- Dashboard
-
An administrator can view the Billing & vCPU Usage dashboard on the HPE Ezmeral Unified Analytics Software homepage. An overview of the daily and monthly aggregated usages for applications is available through the Billing & vCPU Usage dashboard.
In this dashboard, you see the daily usage for the Overall category, which includes all applications, including imported applications. To view the resource usage charts for a specific application, select the application from the dropdown.
The left side provides the total estimated charges for the month with the daily breakdown. The right side displays the usage data.
In HPE Ezmeral Unified Analytics Software, GPU usage is metered per application, and each vGPU counts as an individual GPU for metering.
The GPU metric used for metering is
DCGM_FI_PROF_GR_ENGINE_ACTIVE.The GPU metrics sampling interval is every five seconds.
The hourly usage is the average GPU utilization over a one-hour period.
For example: With seven small vGPUs and four applications using six vGPUs as follows:
Total GPU utilization for the billing record is 1.3 + 0.8 + 0.5 + 0.5 = 3.1 vGPU-hourApplications Pods -vGPU Avg 1hr per vGPU Avg 1hr utilization per application App1 Pods11 - vGPU0 0.8 1.3 Pods12 - vGPU1 0.5 App2 Pods21 - vGPU2 0.8 0.8 App3 Pods31 - vGPU3 0.5 0.5 App4 Pods41 - vGPU4 0 0.5 Pods42 - vGPU5 0.5 Top Frameworks provides the list of monthly aggregated usage charges data for all applications in HPE Ezmeral Unified Analytics Software.
The total usage for billing is aggregated for all included and imported applications.