Ray

Provides a brief overview of Ray in HPE Ezmeral Unified Analytics Software.

Ray is a unified framework for scaling AI/ML and Python applications, handling distributed workloads, and parallelizing serial applications. As a distributed computing framework, Ray simplifies scalability and fault tolerance. Ray offers flexible programming for parallel tasks and actors, making it suitable for data processing, reinforcement learning, and simulation.

To learn about API changes for Ray 2.0, see Ray 2.0 Migration Guide.

Ray Core
Ray Core provides core primitives to build and scale distributed applications. The core primitives are:
Ray Libraries
HPE Ezmeral Unified Analytics Software supports the following Ray libraries:
Purpose
  • Simplify development by providing high-level abstractions and automatic management of complex distributed systems.
  • Accelerate the development process by reducing the complexity of building distributed systems.
Use Cases
  • Data Processing: Efficiently handle large-scale data processing tasks.
  • Reinforcement Learning: Scale RL experiments across multiple machines for faster learning.
  • High-Performance Computing: Parallelize complex computations for faster execution in HPC scenarios.
  • Event-driven and Real-time Systems: Process events or data streams in parallel for timely processing.

Features and Functionality

Ray in HPE Ezmeral Unified Analytics Software supports the following features and functionality:

Ray Cluster Reconciliation

HPE Ezmeral Unified Analytics Software provides an automatic Ray cluster reconciliation feature using Helm hooks.

When you upgrade Ray in HPE Ezmeral Unified Analytics Software, all Ray workloads, including head nodes, workgroup nodes, small group nodes, and computational resources such as CRDs, config maps, services, and others are managed autonomously.

The Ray cluster reconciliation feature improves the user experience for AI application development.

Notebook Integration

A pre-existing image is created in Kubeflow notebooks with Ray library. See Creating and Managing Notebook Servers.

To submit jobs using Ray, you can connect to Ray cluster. See Connecting to Ray Cluster.

Ray Dashboard
Ray dashboard in HPE Ezmeral Unified Analytics Software allows you to:
  • Understand Ray memory utilization and debug memory errors.
  • See per-actor resource usage, executed tasks, logs, and more.
  • View cluster metrics.
  • Kill actors and profile your Ray jobs.
  • See errors and exceptions at a glance.
  • View logs across many machines in a single pane.
  • See Ray Tune jobs and trial information.

To access Ray dashboard, click the Tools & Frameworks icon on the left navigation bar. Navigate to the Ray tile under the Data Science tab and click Open.

To enable Metrics view in Ray dashboard, see Enabling Metrics in the Ray Dashboard.

Security

To configure Ray to use TLS authentication for client-server communication, see TLS Authentication.

To learn more about Ray, see Ray documentation.