How AuriStor Enables Global Data Access Without Storage-Based Pricing

AuriStor's distributed file system enables global data access for research and financial institutions, offering predictable costs and eliminating storage-based pricing.

Data teams face a simple yet challenging problem: how to provide people with access to the data they need, wherever they are, without incurring significant costs or compromising security?

AuriStor built a distributed file system that addresses this. Their approach is practical rather than flashy. They took the Andrew File System, identified what didn't work, and spent a decade fixing it. The result is a commercial product with a pricing model that challenges industry norms.

The business case for distributed file systems

Organizations generate data in multiple locations. Research teams collaborate across institutions. Financial firms run compute workloads globally. Scientists need to access experimental data from anywhere.

The traditional solution is copying files between systems or using VPNs. Both create problems. Copying leads to version control issues and wasted storage. VPNs add latency and complexity.

AuriStor provides a global namespace accessible via a simple path: /afs/organization.edu/dataset. Files stored anywhere appear in the same directory structure for all users. The system handles replication, caching, and synchronization automatically. Zero client configuration required—install the client, and DNS handles service discovery.

Use cases that make sense

The U.S. Geological Survey uses AuriStorFS to deliver natural hazard information during emergencies. When hurricanes or earthquakes strike, reliable access to current data is crucial. The system replicates content across three AWS regions, taking snapshots every 10 minutes and replicating them.

Their content updates constantly in real-time. Clients automatically fetch from the nearest region. If that region fails, they seamlessly access data from the next closest location.

SLAC National Accelerator Laboratory runs experiments that generate massive datasets. Researchers worldwide need access to the same files. AuriStorFS lets them read data directly from compute nodes or access it remotely from their laptops. No intermediate copying required. They can monitor experiment output in real-time—a common use case that's impossible with traditional HPC storage.

One large financial institution (under NDA but identifiable from public presentations) deployed AuriStorFS for software distribution at scale. They serve 175,000 clients across 80 cells with 300 servers managing 1.5 million volumes. The infrastructure spans 180 to 200 regional cells globally across multiple cloud providers, including AWS, Google Cloud, and Oracle Cloud.

They hit OpenAFS limits at 15,000 compute cores and 20,000 client nodes. Performance degraded and reliability suffered. After migrating to AuriStorFS, they expanded to handle ten times the client load while reducing server count. One misconfigured AWS deployment accidentally sent 15,000 cloud clients to an internal data center cell instead of the intended cloud cell. Nobody noticed for months because AuriStorFS handled the load without impacting performance.

Pricing that scales differently

Most storage solutions charge based on capacity or data volume. Add another terabyte and your bill goes up. Scale to petabytes and costs become unpredictable.

AuriStor charges per server and per user/machine identity. The base price is $21,000 annually for a cell with up to 4 servers and 1,000 identities. Add more servers at $1,000 to $2,500 each, depending on quantity. User identities follow a tiered structure—1,000 to 2,500 identities cost $1,375 total, while 50,000 to 100,000 identities cost $30,000.

Storage capacity doesn't affect pricing. As CEO Jeffery Altman explained, "We give you the first 100 exabytes for free." Store 100TB or 500TB on the same server and pay the same amount. For organizations with large datasets but controlled user populations, this model makes budgeting straightforward.

The perpetual license adds leverage. Stop renewing and you keep using the last version you licensed. Security updates continue for two years. This protects against vendor lock-in and sudden price increases.

For the financial customer, the initial deal was performance-based: pay us what monthly outages cost you if we can prevent them. AuriStorFS eliminated the outages. The customer calculated the savings, and that became the license fee.

Volume-based policy management

The unit of management in AuriStorFS is a volume—a policy container for a dataset. A volume might be a home directory, a machine learning dataset, or a specific version of Java, such as 3.17.63, for a particular application.

Volumes carry policies that follow the data:

• Security requirements (encryption level, authentication strength)

• Replication strategy (how many copies, where they're located)

• Geographic restrictions (data cannot leave EU, must be on physical hardware behind locked doors)

• Maximum access controls (system administrators can prevent users from granting overly permissive access)

This matters for AI workloads where you don't want training data escaping your perimeter. A volume can specify that it must never be hosted outside Switzerland, or only on servers labeled as being in specific jurisdictions.

Security for multi-tenant environments

Combined identity authentication verifies both the user and the machine. You're not just "you"—you're "you on your phone," "you on your company laptop," "you running an application in AWS." Each combination can have different permissions.

A researcher may access a sensitive dataset, but only from designated workstations within a secure facility. The same user from a coffee shop gets denied. This granular control helps organizations share data externally while maintaining compliance.

The system logs every remote procedure call with authentication details and network endpoints. For regulated industries, this creates a comprehensive audit trail without the need for add-on products.

File servers can enforce security policies. You can require that all connections to a file server use AES-256 encryption and strong authentication. Volumes can only be attached to or copied to file servers with matching or stronger security policies.

Performance metrics that matter

One customer reported that shutdown times for a file server with 1.7 million volumes dropped from a 30-minute timeout (which often caused the process to be killed) to 4 seconds. This means maintenance windows shrink dramatically. You can restart servers during business hours without impacting users.

File transfers improved substantially. A recent customer test showed that a 1GB file copy, which took 3 minutes 11 seconds with OpenAFS, was reduced to 1 minute with AuriStor's 2021 release, and then to 30 seconds with the latest version, incorporating path MTU discovery and proportional rate reduction.

For data-intensive workloads, this compounds. Hundreds of file operations per analysis job add up. The system handles 500,000 simultaneous client connections on production servers. Database operations scale to 32 write transactions per second and 40,000 read transactions per second across the cluster.

The financial customer deploys new software releases continuously; updates are pushed out every few minutes. Traditional container image distribution can't keep up with this pace. AuriStorFS replicates only incremental changes, making this deployment velocity possible.

Real distributed locking

Unlike copy-sync solutions, AuriStorFS implements true distributed file system semantics. If a user in France, a user in New York, and a user in Australia all open the same Excel spreadsheet, they can collaborate in real-time.

When one person edits a cell, Excel locks that cell. The lock is immediately visible to all users. As soon as changes are written and the lock drops, updates appear everywhere. It's not three local copies with last-write-wins synchronization—it's a single source of truth with cache coherency.

This is particularly important for high-performance computing workloads, where multiple processes write to different regions of large files. Clients can work on specific portions without interfering with one another. The file server fetches only the regions' applications that are actually accessed, not entire files.

Integration challenges remain

AuriStorFS doesn't replace every storage system. It's not designed for databases or VM images because it lacks server-side byte range locks. That feature is planned but won't arrive before 2026.

The system works best for read-heavy workloads with collaborative access patterns. Research data, software distribution, configuration management, and shared project directories fit well. Transaction processing and database backends don't.

Organizations running OpenAFS can migrate incrementally. The protocols are compatible, allowing you to run mixed environments during the transition. However, you lose some features until you complete the migration—you're limited to the lowest common denominator.

What analytics teams should consider

If your team works with large datasets across multiple geographic regions, evaluating distributed file systems is worth considering. AuriStorFS offers predictable costs, strong security, and proven performance at scale.

The perpetual licensing and fixed-price model work well for organizations with stable infrastructure teams and long planning horizons. Startups that move fast might prefer cloud-native alternatives, which come with different trade-offs.

The source code is closed, but it is available to licensees who wish to participate in its development and maintenance. This creates a walled garden development model—the opposite of open source, but chosen deliberately after watching OpenAFS struggle with funding. The company is profitable with sub-$2 million revenue and a core team of five developers.

For research institutions, government agencies, and enterprises with complex data sharing requirements across multiple locations, AuriStor provides a viable option. The technology works. The pricing is transparent. The perpetual license protects against vendor lock-in.

The question is whether a distributed file system fits your data architecture. If you require global data access without storage-based pricing and can tolerate the current limitations, it's worth evaluating.