User Guide

Getting Started

Domain Power Service (DPS) is a datacenter power management system that monitors power consumption, enforces power policies, and integrates with BMCs and cluster management systems. It provides the nvidia.dcpower.v1 gRPC API for programmatic access and addresses the growing challenge of optimizing power consumption while maintaining performance and reliability at scale.

Domain Power Service (DPS) provides flexible deployment options to meet your needs, from quick testing to full datacenter power management.

Key Concepts

Before deploying DPS, it’s helpful to understand a few core concepts:

Topology - A representation of your datacenter’s power distribution network, from utility feeds down to individual compute nodes
Power Policies - Rules that define power limits and management strategies for different workload scenarios
Resource Groups - Collections of compute resources (nodes/GPUs) that share power budgets and policies
dpsctl - The command-line interface for managing DPS

For detailed explanations of these and other concepts, see the Concepts section.

Architecture

Deployment Options

1. Try It Out - DPS SDK Simulator

The DPS SDK Simulator provides a lightweight, pre-configured environment for testing and development without requiring full datacenter infrastructure. This is ideal for:

Learning DPS concepts and workflows
Developing and testing custom integrations via the SDK
Partner integration development
Proof-of-concept demonstrations

The simulator includes a mock topology and supports all DPS APIs, allowing you to explore functionality without hardware dependencies. It also provides:

Grafana dashboards for visualizing power metrics and system behavior
Simulation scripts for testing resource group scenarios and Grid event responses
Pre-configured examples for common use cases

Simulator User Guide: SDK Simulator User Guide

2. Full Datacenter Deployment

For production power management, DPS can be deployed as a comprehensive solution for your datacenter. This deployment provides:

Real-time power monitoring and policy enforcement
Integration with workload schedulers (e.g., SLURM)
Multi-level power distribution management
GPU power optimization capabilities

Prerequisites:

Kubernetes cluster for DPS services
Network access to BMCs for all managed nodes
Understanding of your datacenter’s power distribution architecture

Deployment Steps:

Deploy DPS - Set up the DPS server infrastructure using Helm charts
Install dpsctl - Install the command-line client for administration
Create Topology - Model your datacenter’s power distribution network (see Topologies)
Import Topology - Load your topology configuration into DPS
Configure Policies - Define power policies appropriate for your workloads

Note: Administrators will need to create and import a custom topology that accurately represents their datacenter’s power distribution network. See the Managing Topologies guide for detailed instructions.