What are you looking for ?
VergeIO
RAIDON

ZettaLane Systems Launches MayaNAS on Google Cloud Marketplace: Lustre Running Directly on Google Cloud Storage

MayaNAS runs enterprise NAS and the Lustre parallel file system from one engine on standard Google Cloud infrastructure, with bulk data in Google Cloud Storage

ZettaLane Systems LLC., a Select Google Cloud Technology Partner, announced that MayaNAS, the only MLPerf Storage v3.0 submission verified running the Lustre parallel file system directly on cloud object storage, is available on Google Cloud Marketplace.Zettalane LogoMayaNAS presents standard enterprise NAS (NFS, and SMB with Active Directory) alongside the parallel file systems Lustre and pNFS, with bulk data stored in Google Cloud Storage.

In MLPerf Storage v3.0, run entirely on standard Compute Engine instances with Google Cloud Storage, MayaNAS sustained 32.42 GiB/s checkpoint write and 21.20 GiB/s read on the Llama 3 70B workload with two client nodes and four storage nodes. On the 3D U-Net training workload, a single client node kept three simulated Nvidia B200 accelerators about 92% busy. The result shows that one file system on standard Google Cloud infrastructure can serve both enterprise file workloads and the most demanding parallel ones, with Google Cloud Storage as the storage layer.

Most cloud NAS stops where HPC and AI begin. On Google Cloud, MayaNAS lets teams scale from everyday file shares to GPU-class parallel throughput without deploying a second storage system. Bulk data lives permanently in Google Cloud Storage, so there are no tiering or staging jobs between a fast tier and the bucket. Capacity grows with the bucket, and teams pay for performance only while they need it: serving nodes scale up for a run and back down afterward while the data stays in place. The file system, buckets and encryption keys remain inside the customer’s own project, supporting data-residency and sovereignty requirements. MayaNAS runs active-active for HA and uses the standard open-source clients already in Linux, with no proprietary software to install.

Beyond AI, the same parallel throughput serves genomics and life-sciences pipelines, electronic design automation, media rendering, financial risk analysis and seismic processing. MayaNAS is built for enterprise IT, research institutions, regulated industries and sovereign AI initiatives that need parallel file performance at object-storage scale while keeping full control of their data.

Click to enlarge

MayaNAS brings the Lustre parallel file system directly to Google Cloud Storage, now available on Google Cloud Marketplace

“Cloud object storage scales without limits and handles durability on its own. We built MayaNAS to use it directly as the storage layer beneath a parallel file system, so teams get the enterprise file services they already run and the parallel Lustre and pNFS throughput their GPUs need, with no disks, RAID or separate fast tier to manage,” said Supramani Sammandam, founder, ZettaLane Systems. “MLPerf Storage v3.0 showed that standard Compute Engine instances and Google Cloud Storage make a powerful combination for serving the AI data pipeline, from training and checkpointing to inference cache, with the data staying in the customer’s own project.”

“Bringing ZettaLane Systems to Google Cloud Marketplace will help customers quickly deploy, manage, and grow the company’s MayaNAS storage platform on Google Cloud’s trusted, global infrastructure,” said Dai Vu, MD, marketplace & ISV G programs, Google Cloud. “ZettaLane can now securely scale and support enterprises using MayaNAS to deliver high-performance, shared storage for demanding workloads, including AI training.”

Customers can adopt MayaNAS the way they already buy cloud: pay-as-you-go on Google Cloud Marketplace with a free trial, deployed in minutes into their own VPC and billed through their Google Cloud account, with eligible purchases counting toward existing Google Cloud commitments; through private offers for negotiated terms, directly or via Google Cloud reseller partners; or as code, using ZettaLane’s open-source Terraform modules, which deploy the same Marketplace image and also cover pNFS, for teams that run their infrastructure as code.

Click. to enlarge

Official MLPerf Storage v3.0 results for MayaNAS on Google Cloud, verified by MLCommons Association

MayaNAS runs enterprise NAS and the Lustre parallel file system from one engine on standard Google Cloud infrastructure, with bulk data in Google Cloud Storage.

“ZettaLane can now securely scale and support enterprises using MayaNAS to deliver high-performance, shared storage for demanding workloads, including AI training,” said Dai Vu, MD, Marketplace & ISV G Programs, Google Cloud.

Read also :
Articles_bottom
AIC