en.Wedoany.com Reported - June 3, 2026, Manchester, UK – PEAK:AIO today announced the release of Lattice, the industry's first open-source pNFS metadata server. Developed in long-term collaboration with Los Alamos National Laboratory (LANL), Lattice employs a novel distributed metadata architecture designed to eliminate persistent performance bottlenecks in large-scale AI and high-performance computing infrastructure.
Current AI infrastructure demands are reshaping the storage market. AI workloads—including training large models, large-scale inference, and serving agentic AI applications—require ultra-fast, parallel, and consistently reliable access to massive datasets. GPU computing power has expanded significantly, but the storage layer, particularly the architecture coordinating metadata, has not kept pace. Across 23,000 production clusters, average GPU utilization is only 5%, primarily not due to hardware limitations but because the software systems feeding data cannot keep up. Metadata bottlenecks in parallel storage systems have become a critical constraint on AI workload performance.
Built as a Linux-based userspace pNFS metadata server, Lattice is designed for scalability, modularity, and distributed coordination. Released under the Linux Foundation, this open-source project divides the metadata control plane into four distinct layers: the Protocol State Plane, Lattice Core, MD Catalog Authority, and Data Server Control Plane. This architecture makes metadata services elastic for the first time, allowing them to be dynamically started on commodity hardware at any time, scaling from a single server to over 1,000 metadata servers. The announcement will be formally made at the International Conference on Massive Storage Systems and Technology (MSST 2026) in Santa Clara, California. Lattice is launched in partnership with the Linux Foundation to accelerate community innovation around scalable AI and HPC storage infrastructure.
Gary Grider, Head of the High-Performance Computing Division at LANL, stated that pNFS-Lattice is unique because it is an open-source, userspace, scalable pNFS metadata server built from the ground up, leveraging the concept of separating pNFS metadata services from metadata storage (directories). Because the service is decoupled from persistent metadata and runs in userspace, it is ideally suited as an elastic service that can be dynamically resized. Its open-source and userspace nature lowers the barrier for community participation. Performance tests conducted during the collaboration showed improvements from 70 GB/s to 400 GB/s. On LANL's existing production hardware, standard Linux NFS configurations achieved throughput between 3 GB/s and 7 GB/s, while the pNFS Lattice architecture achieved 40 GB/s on the same servers. Additional testing with a top-tier technical university demonstrated performance improvements of over 300% for metadata-intensive workloads.
In standard metadata benchmarks such as MDtest, early tests showed up to a 10x improvement over standard Linux KNFSD. Lattice's advanced features delivered over 300% improvement in traditionally difficult metadata-intensive workloads. Combined with its elastic, ephemeral metadata scaling model—which allows dynamic addition of metadata services based on demand—Lattice breaks through the limitations of traditional high-performance data design. Roger Cummings, President and CEO of PEAK:AIO, stated that the AI infrastructure market is approaching an inflection point where simply scaling compute no longer yields significant efficiency gains. The collaboration with LANL is based on the premise that for AI infrastructure to scale efficiently, metadata must be elastic, distributed, and open. Lattice represents this shift. Additionally, PEAK:AIO will offer a commercially supported version, PEAK:AIO pNFS, which is a superset of Lattice designed for organizations requiring enterprise-grade SLAs and a full feature set without directly managing the open-source stack. The business model is analogous to the relationship between Lustre and its commercial distributions, while maintaining a fully open standards foundation.
Mark Klarzynski, Chief Strategy Officer and Co-founder of PEAK:AIO, noted that Lattice's key innovation lies in decomposing what was traditionally locked within a single metadata server into four distinct layers. This separation unlocks intelligent scaling capabilities in ways traditional storage architectures were never designed to support. Metadata and data services can now become distributed, elastic participants capable of scaling with workloads, failing over, and adapting. PEAK:AIO is a software-defined AI storage company headquartered in Manchester, UK, whose platform delivers high-performance AI storage from a single server to exabyte scale on any industry-standard hardware. It is deployed at Los Alamos National Laboratory, NHS AIDE, Oxford Robotics Institute, Carnegie Mellon University, University of Liverpool, University of Strathclyde MediForge Hub, and the Zoological Society of London. Los Alamos National Laboratory is a multi-program, federally funded research and development center under the U.S. Department of Energy's National Nuclear Security Administration, with primary responsibilities including nuclear weapon design, addressing nuclear threats, and conducting national security science, technology, and engineering work.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
