SAN MATEO, Calif., May 20, 2025 — Alluxio, a leader in the AI and data acceleration realm, has unveiled Alluxio Enterprise AI 3.6, a transformative update designed to enhance model distribution, optimize model checkpoint writing, and bolster multi-tenancy support. With this release, organizations can significantly expedite AI model deployment cycles, minimize training durations, and ensure unified data access across various cloud environments.

The escalation in AI model sizes and distributed inference infrastructures across multiple regions poses substantial challenges for AI-powered organizations. Large-scale model distribution from training zones to production introduces bottleneck latencies and increased cloud expenditure, while time-intensive checkpoint writing processes decelerate model training cycles considerably.

“We are thrilled to extend our AI acceleration platform to not only streamline model training but also to enhance and simplify AI model distribution to production inference environments,” said Haoyuan (HY) Li, Founder and CEO of Alluxio. “By working closely with pioneering AI-focused clients, we consistently set new benchmarks beyond what was imaginable only a year ago.”

Alluxio Enterprise AI version 3.6 introduces several pivotal features:

High-Performance Model Distribution

By leveraging the Alluxio Distributed Cache, Alluxio Enterprise AI 3.6 accelerates model distribution tasks. Models need to be transferred from the Model Repository to the Alluxio Distributed Cache just once per region rather than once per server. Inference servers access models directly from the cache, benefiting from optimizations like local caching on servers and expanded memory pool utilization. Tests demonstrate remarkable throughput with the Alluxio AI Acceleration Platform hitting 32 GiB/s—outperforming the available network capacity of 11.6 GiB/s by an additional 20 GiB/s.

Fast Model Training Checkpoint Writing

Building on previous advancements, the 3.6 version introduces the ASYNC write mode, reaching speeds of up to 9 GB/s write throughput in 100 Gbps network environments. This advancement profoundly cuts down the time needed for model training checkpoints by making use of the Alluxio cache rather than the underlying file system, thereby bypassing network and storage constraints. Checkpoint files are asynchronously written to the file system, enhancing training workflows.

New Management Console

The introduction of a comprehensive web-based Management Console in Alluxio 3.6 aims to amplify observability and streamline administrative tasks. This console details vital cluster information like cache usage, coordinator and worker status, and essential statistics such as read/write throughput and cache hit rates. It also empowers administrators to oversee mount tables, set quotas, configure priority and TTL policies, launch cache jobs, and gather diagnostics—all within a user-friendly interface, eliminating the need for command-line dependency.

Enhanced Alluxio Administrators Features

• Multi-Tenancy Support – Robust multi-tenancy capabilities are now available through seamless integration with Open Policy Agent (OPA). This allows administrators to define fine-grained role-based access controls for multiple teams using a unified, secure Alluxio cache.
• Multi-Availability Zone Failover Support – Alluxio 3.6 introduces data access failover support in multi-Availability Zone configurations, ensuring higher availability and resilience of data access.
• Virtual Path Support in FUSE – Users can now define custom access paths to data resources, creating an abstraction layer that hides physical data locations in underlying storage systems.

Availability

Alluxio Enterprise AI version 3.6 is now available for download. Interested users and organizations can access it through this link.

Supporting Resources

• Learn more about Alluxio Enterprise AI 3.6 on its official blog.
• Access a trial version: Download here.

About Alluxio

Alluxio leads in providing accelerated data access platforms for AI workloads. Its distributed caching layer facilitates AI and data-intensive processes by enabling high-speed data access across varied storage systems. By establishing a global namespace, Alluxio consolidates data from multiple sources, including on-premises and cloud environments, into a singular, logical view, eliminating data duplication or complex movement challenges.

Engineered for performance and scalability, Alluxio minimizes I/O bottlenecks and latency by positioning data closer to compute frameworks such as TensorFlow, PyTorch, and Spark. With its intelligent caching and data locality optimization, Alluxio seamlessly integrates with modern data platforms, serving as an indispensable tool for teams developing and scaling AI pipelines across hybrid and multi-cloud landscapes. Backed by top-tier investors, Alluxio supports leading technology, internet, financial services, and telecommunications companies worldwide, including 9 out of the top 10 internet enterprises. For more information, visit www.alluxio.io.

Media Contact:
Beth Winkowski
Winkowski Public Relations, LLC for Alluxio
Phone: 978-649-7189
Email: [email protected]