Hammerspace Hits Scale Up Phase For Its Global Filesystem Solution

Data orchestration vendor Hammerspace has taken on a hefty round of funding to scale up its global filesystem solution.

The USD $56.7 million raised is the first external funding the company has taken on. That would be a large series A round, but since its founding in 2018, Hammerspace has enjoyed financial support from founder and CEO David Flynn after previous success with Fusion-io which was acquired by SanDisk in 2014 for around $1.1 billion. This funding round is therefore better thought of as a Series B round, with internal funding providing the function Series A funding normally would.

Hammerspace makes scale-out storage software for unstructured data that provides global access to files wherever they’re stored. It is ideal for real-time collaborative work such as data analysis, media asset design and production, and AI/ML workloads.

“It’s all about placing the data where you need it to meet business objectives,” said Molly Presley, SVP of Marketing at Hammerspace. “It could be latency to an application, or cost, or performance. Hammerspace will place the data in the appropriate location to meet your business SLA requirements.”

As subscribers to The Crux are already aware, Hammerspace started out as the software components from a previous storage company Primary Data (also founded by Flynn) that had interesting technology, but struggled to get traction with its hardware/software combo.

{{}}

Hammerspace is also the name given to the virtual realm that cartoon characters use to pull objects out of thin air. The reference is intentional.

How Hammerspace Works

Hammerspace uses some important features of the NFS v4.1 standard to work its magic. Specifically: Parallel NFS, or pNFS. The separation of data from metadata is very clearly defined in NFS v4.1 and is what enables Hammerspace to provide high-speed access to metadata at a variety of locations, all within a single namespace.

pNFS has an inbuilt ability to manage data on the fly, which turns out to be quite useful. Hammerspace uses it for data tiering, and to separate the replication of metadata from moving actual data around. Metadata tends to be much more compact than data, and is therefore easier to copy quickly.

Hammerspace is also able to abstract the underlying data storage mechanism from the data layout, which helps it insert itself into existing environments. Using a method it calls FlexFiles, Hammerspace can use pNFS to provide client access to data while actually storing the data on older NFS v3 storage.

NFS v3 (and v2 before it) are stateless protocols. This created challenges as the world moved to very large numbers of storage clients all trying to access a small number of servers. Servers can be more efficient if they’re able to store some state about clients—pre-fetching of data based on common data access patterns, for example—but Hammerspace turns this limitation into an advantage.

If you’re interested in learning more about the history of pNFS and how Hammerspace works, CTO Trond Myklebust goes into detail in this Hammerspace podcast. Myklebust is also the maintainer and lead developer of the Linux kernel NFS client.

Hammerspace in the Enterprise

The separation of metadata from data in pNFS means Hammerspace can store the actual data on existing NFS systems that enterprise customers already have. The metadata, meanwhile, can be stored in Hammerspace on its metadata servers. This makes adding Hammerspace into an enterprise environment much easier, because customers don’t need to completely replace existing systems with an entirely new approach.

“If you have six different arrays, Hammerspace can be the unifying layer,” said Molly Presley, SVP of Marketing at Hammerspace.

This separation also helps Hammerspace partner with other storage vendors that are trusted by enterprises, such as Pure Storage (a PivotNine client). Customers can get the benefits of a data orchestration layer while keeping the storage they already know and value.

Hammerspace is still working out the ideal buyer for its solution, as customers are rapidly changing how they think about data management. The hype around large-language models and similar SALAMI use cases is obscuring some of the more valuable data processing workloads that Hammerspace is also well-suited to. Some of the more advanced customers are creating roles such as Chief Data Officer and Data Architect to focus on the challenges specific to large-scale data management.

“Once you connect the user and application to Hammerspace you never have to reconnect it elsewhere,” said Presley. “If data is in flight, if you’re moving data to a new location, it’s not disruptive because users and applications are interacting with metadata.” This frees customers from low-level questions about storage infrastructure so they can instead focus on more holistic, whole of organisation data management questions.

“I think it’s going to be called data orchestration as an industry. We’re starting to hear more usage of that term,” Presley added.

We can certainly see the value in separating out storage infrastructure questions from data management questions. This separation has been stubbornly difficult to achieve in practice, but this seems mostly due to limitations in what storage technology has been able to provide.

Hammerspace seems to be one of those companies that needed a range of technologies to become mature enough, and widespread enough, that they became viable. NFS v4 has taken over a decade to mature to the point that it can be reliably deployed; pNFS has been similarly constrained. However, the widespread availability of Linux clients with solid pNFS support has done a lot to make options like Hammerspace easy enough for the enterprise to adopt.

The challenge for Hammerspace now is demonstrating that there is a compelling enough set of problems in real-world enterprise customers, and helping them find and adopt Hammerspace as the solution. Customer traction has been solid thus far, from what Hammerspace tells us, but to pay back the substantial money invested in the company will require finding a much larger group of people willing to buy its offering.

PivotNine doesn’t see any unique hurdles for the company in this regard and we look forward to tracking their progress.