ZettaLane's MayaNAS on Google Cloud Marketplace: High-Performance Lustre on Object Storage for AI Workloads

ZettaLane Systems, a Google Cloud Technology Partner, has announced the general availability of its MayaNAS solution on the Google Cloud Marketplace. MayaNAS offers a high-performance Lustre parallel file system that operates directly on Google Cloud Storage. This integration allows for the delivery of enterprise NAS capabilities (NFS, SMB with Active Directory) alongside Lustre and pNFS, with bulk data residing in Google Cloud Storage. The solution has demonstrated impressive performance in MLPerf Storage v3.0 benchmarks, achieving 32.42 GiB/s for checkpoint writes and 21.20 GiB/s for reads on a Llama 3 70B workload using standard Compute Engine instances and Google Cloud Storage. This development is crucial for organizations heavily invested in AI and machine learning, especially those working with large datasets and complex models like large language models. The traditional challenge has been reconciling the need for high-throughput, low-latency file system performance with the scalability and cost-efficiency of object storage. MayaNAS addresses this by enabling a parallel file system to leverage object storage as its backend, eliminating the need for separate, often complex, storage tiers or data movement. This directly impacts data scientists and MLOps engineers who can now access and process massive AI datasets more efficiently, accelerating training times and reducing operational overhead. The broader trend in cloud and DevOps is a continuous drive towards abstracting infrastructure complexity and optimizing for specific workload demands. Object storage has long been recognized for its scalability and cost-effectiveness for unstructured data, but its performance characteristics historically made it less suitable for high-performance computing (HPC) and AI workloads that demand POSIX-compliant file system semantics and extreme I/O. Solutions like MayaNAS represent a significant evolution, building on the growing maturity of cloud object storage and its integration with specialized file systems. This aligns with the industry's move towards AI-native storage architectures that prioritize performance, scalability, and cost-efficiency for AI training and inference. In practice, this means practitioners can now deploy and manage Lustre environments directly within Google Cloud, with the underlying data seamlessly managed by Google Cloud Storage. This simplifies procurement, deployment, and scaling of storage for AI projects. Teams should evaluate MayaNAS for workloads that require the performance of Lustre but also benefit from the elasticity, durability, and cost model of cloud object storage. It offers a compelling alternative to managing dedicated, on-premises HPC storage infrastructure or complex data migration strategies between different cloud storage types. Organizations should consider their specific performance requirements, existing Google Cloud investments, and the potential for simplified data lifecycle management when assessing this solution. The ability to run demanding AI workloads without data migration between storage types is a significant operational advantage.