Search the docs

Introduction and Solutions

A Hammerspace Global File System (GFS) presents a single logical file system that spans multiple locations, regardless of geographic distance. Users and applications at every site see the same files and directories, with the same names, permissions, and metadata, even though the underlying data may physically reside at a different site or in cloud object storage. The goal is to make data a global resource for the enterprise instead of confining files to a single location.

This is a different model from traditional storage, where the only way to give a remote site access to data is to copy files — producing copy sprawl, manual staging and de-staging, and management complexity. Hammerspace replicates the presentation layer (metadata) continuously across sites and moves file data only when policy or access requires it.

What a Global File System Is — and Is Not

A Hammerspace GFS is one file system and one namespace shared by two or more Hammerspace clusters, each of which is a site. Every site serves its own local users and applications directly, accepts writes, and keeps working on its own if the other sites are unreachable. Sites stay aligned by continuously replicating metadata to one another, and move file data only when policy or access requires it.

What a GFS is not is a real-time collaborative editor. It does not merge simultaneous edits to the same file the way a shared online document does. Each site operates on its own instances of files and the system reconciles changes using an eventually consistent model. Where two sites change the same file at the same time, GFS retains a recoverable copy of the superseded version rather than silently discarding it (see How conflicts are handled for the mechanics).

Two properties define the model:

Active-active

Every site can serve data and accept writes. Global accessibility allows active-active write patterns, but those patterns require planning and discipline (see Protecting Data: Snapshots, Versioning, and Undelete and the unsupported use cases in Prerequisites, Sizing, and Networking).

Eventually consistent

Changes made at one site propagate to the others within a short, bounded interval (5 seconds by default for metadata). Hammerspace deliberately does not use cross-site file locking, because synchronous locking over long distances degrades performance and makes all sites read-only if any one site becomes unavailable.

Solutions Enabled by GFS

The Global File System supports a range of data-management solutions:

  • Hybrid cloud — extend on-premises resources to cloud storage with global file access.

  • Burst-to-cloud — use compute in one or more cloud regions on demand.

  • Multi-site collaboration — enable follow-the-sun workflows.

  • Workflow automation — stage data non-disruptively to compute anywhere for processing.

  • Disaster recovery — file-granular, prioritized recovery on any site.

  • NAS to the edge — extend NAS to edge locations without dedicated infrastructure.

  • Cloud data migration — migrate data in the background without interrupting access.

  • Storage consolidation — consolidate NAS non-disruptively, and centralize edge backups to a core site.

Key Features

  • Multi-protocol data access — active-active across NFS 3, NFS 4.1, NFS 4.2, SMB 2.x/3.x, and S3, with no client software required.

  • Objective-based data policies driven by file-system and custom metadata (see Core concepts for what an objective is, and Objectives and data placement for how to build one).

  • Transparent cross-site data orchestration with integrated global deduplication and compression (see The shared Object Storage Volume in depth).

  • Encryption of the data payload in transit and at rest in cloud/object storage (see The shared Object Storage Volume in depth).

  • Global snapshots and global undelete (see Protecting data: snapshots, versioning, and undelete), and file-granular WORM.

  • Conflict resolution with cross-site ownership versioning (see How conflicts are handled).

  • Per-site data management, so sites cannot interfere with one another’s data mobility.

  • Heterogeneous, capacity-agnostic storage — any vendor, only enough capacity for active data.

  • Tolerant of high-latency links; qualified/supported for up to 16 sites per share (see Requirements and limits at a glance — this is a supported-configuration limit, not a hard product ceiling).

  • Support for high-performance AI and HPC use cases via Parallel NFSv4.2 with FlexFiles.