Overview
The Anyscale Platform is a fully managed, scalable compute platform built on Ray that enables any organization or AI developer to effortlessly build, tune, train, and scale machine learning (ML) and Python workloads.
Thousands of organizations use Ray to increase developer velocity and scale AI. For an overview of the Anyscale platform, see the Anyscale data sheet.
Why Anyscale
Ray is used by thousands of organizations globally to increase developer velocity and scale AI. Anyscale offers several key features on top of the open source Ray:
A production-grade scalable compute platform
- Fully managed service: Anyscale operates clusters of machines on demand so infrastructure teams don’t have to operate them. ML practitioners can focus on coding in an interactive, scalable compute environment.
- Bring your own cloud: Anyscale is built from the ground up with customer data security in mind. It runs in your organization’s cloud account.
- Optimize compute costs: Anyscale provides features such as spot instance support, autoscaling, and auto-suspend to help your team reduce compute costs. You can also leverage existing agreements with public cloud providers (for example, AWS Reserved Instances or saving plans).
- Governance and compliance: Anyscale provides user access controls for Projects, Workspaces and Clusters as well as cost tracking mechanism. Anyscale has SOC 2 Type 2 attestation.
- Expert support: With Anyscale, teams get direct access to the Ray creators. In contrast to OSS where support is provided on a best-effort basis by the community and occasionally by committers, Anyscale offers dedicated support provided by Ray and Anyscale engineers.
Unified development environment for ML and Python workloads
- All-in-one development environment: Anyscale Workspace provides an IDE-centric experience for code development, dependency management, and observability similar to the local laptop but on top of a scalable cluster.
- Use the tools you love: Anyscale provides integration and instant setup for popular tools such as VSCode and Jupyter notebook, with GitHub, Weights & Biases, and more.
- Collaborate: Share or clone experiments with a click of a button. Different users can access a Workspace with all the same configuration and environments and be productive instantaneously.
Seamlessly move between development and production
- A unified environment for development and production: develop, run, debug, and test your code at scale on the same cluster configuration with the same software dependencies for both development and production.
- Flexible and extensive dependency management: Anyscale provides different options to manage your dependencies across your cluster. You can use existing base images, bring your own Docker, or use Ray’s runtime environments for faster iteration.
- Jobs and Services: Jobs and Services API and SDK provide an easy interface to advance your workloads into production and integrate with your existing deployment tools. Anyscale Jobs support cron jobs, ephemeral cluster creation and retry capabilities, while Anyscale Services provides replica management, no downtime upgrades and high availability.
- Managed logs, monitoring, and observability: Anyscale provides a production-grade monitoring and observability stack with a managed Grafana and Ray dashboard. Additionally, Anyscale provides production monitoring and notifications for added trust that pipelines are running well.
Compare open source Ray to Anyscale-managed Ray
| Open source Ray | Anyscale-managed Ray | |
|---|---|---|
| Ray Python package | Standard Ray | ✅ Optimized Ray with additional features that enhance performance and cost efficiency |
| Fully managed clusters | N/A | ✅ Deploy and manage clusters in your cloud provider account |
| Networking, Security configuration | N/A | ✅ No public IP, access control, secret management, etc. |
| Cost saving | N/A | ✅ Idle termination, cost monitoring, etc. |
| Development experience | Ray jobs API, SSH, Ray client, and manual integration with MLOps tools | ✅ Anyscale Workspace that provides out-of-box integrations with tools like VS Code, Jupyter, Weights and Biases, etc. |
| Enhanced observability | Basic Ray dashboard and logs that need manual instrumentation and setup | ✅ Optimized Ray dashboard + Grafana, persisted logs |
| Developer environment | Basic docker support and runtime environment | ✅ Optimized cluster environment + runtime environment that greatly reduce startup time |
| Instance startup time | Not optimized | ✅ Optimized instance startup time |
| Production-grade job management | Basic Ray jobs | ✅ Anyscale Jobs support automatic retries, scheduled jobs, etc. |
| Production-grade deployment management | Basic Ray Serve deployment | ✅ Anyscale Services support automatic restart, high availability, zero-downtime update, etc. |
| Expert support | Community support | ✅ Dedicated engineer for 1 month + prioritized support |
For more information, see how the Anyscale platform enhances capabilities of open source Ray.
Supported cloud providers
Anyscale currently supports AWS and Google Cloud. Azure support is currently under development. Please contact us if you'd like to request more information.
