This content was previously published on Forbes.
Hammerspace unveiled its new high-performance NAS architecture, Hyperscale NAS, to cater to the growing demands of enterprise AI, machine learning, deep learning initiatives, and the increasing use of GPU computing both on-premises and in the cloud.
The new solution addresses the limitations of traditional NAS systems by offering enhanced performance and scalability for processing large datasets, a critical requirement for AI model training.
What is Hyperscale NAS?
Hammerspace’s new Hyperscale NAS combines the high performance and scalability of an HPC parallel file system with the standards-based connectivity and enterprise features of traditional NAS. It’s particularly suited for GPU computing at scale for AI training, inference, and other high throughput, low-latency parallel processing use cases.
The new solution leverages NFS 4.2 Flex Files in Linux to create a parallel file system using existing NFS-based NAS, addressing scalability and performance challenges in standard NFS storage for AI and HPC workloads. This offering is an addition to Hammerspace’s Global File System platform. It aims to simplify deploying a parallel file system while utilizing a standard NFS-based NAS with a Linux client.
Hyperscale NAS exploits the flexible file layout of NFS 4.2, separating metadata from the data path using an out-of-band metadata server managed by Hammerspace Anvil software. This separation streamlines the process required from the backend storage network, enabling a standard NAS solution to overcome traditional scalability limitations.
One of the key advantages of Hyperscale NAS is the ability to deliver file storage performance and scalability comparable to parallel scale-out file systems while leveraging existing capabilities within Linux and NFS 4.2. This contrasts with typical parallel file systems used in scalable HPC environments, which require custom clients and often entail greater complexity and resource demands.
This architecture also supports simultaneous data access for users and applications via NFS or SMB, ensuring correct permissions handling and locking between protocols. It supports a range of networking technologies and provides comprehensive enterprise data governance and management.
Compared to scale-out NAS architectures, Hyperscale NAS provides linear scalability to thousands of nodes while maintaining consistent throughput and IOPs performance, making it more efficient and suitable for large-scale GPU computing
Hammerspace has also completed the Nvidia GPUDirect Storage Self-Certification for Hyperscale NAS. The certification means that deploying the solution with existing storage systems can present any storage system as GPUDirect Storage, ensuring high throughput and low latency performance for Nvidia GPUs.
Analysis
Hammerspace’s new Hyperscale NAS addresses the complex demands of modern high-performance computing, particularly in AI, machine learning, and GPU-intensive workloads.
Its ability to deliver consistent performance across a range of mixed I/O workloads and to scale linearly while maintaining this performance is particularly impressive. It’s a departure from the limitations of scale-out NAS architectures, which often struggle with performance bottlenecks as they expand.
The proven capability of Hyperscale NAS in AI model training, supporting thousands of storage nodes and GPUs, underscores its potential as a leading solution in this space. It’s not just the performance that makes Hyperscale NAS stand out; it’s also the architecture’s ability to improve the performance of existing NAS systems without requiring modifications.
The new offering expands Hammerspace’s addressable market while also bringing the company a fiercely competitive environment. Hyperscale NAS, while differing architecturally, competes directly against Weka.io’s Data Platform, IBM Spectrum Scale, and solutions from VAST Data.
The new solutions build on an impressive period of growth for Hammerspace. The privately held company recently released details of its overall performance. It has grown its sales pipeline by more than 300%, increased its partner channel by 65%, and closed a $56.7 million institutional funding round. The company has doubled its sales force and added an impressive roster of executives to its leadership rank.
The introduction of Hyperscale NAS is timely, given the growing reliance on AI and machine learning in enterprise operations. As the concerns of IT organizations shift from storage systems to understanding and managing data, Hammerspace finds itself at the right place at the right time. Its new Hyperscale NAS solution raises the bar, positioning Hammerspace to solve today’s enterprises’ increasingly complex data challenges.
