OpenSFS, the Intel-backed sponsors of the main community version of Lustre, announced the latest release of the file system, which comes with a number of features designed to push Lustre further past the halls of HPC and into a much more diverse pool of potential users.
During a conversation about this update with Galen Shipman, Chairman at OpenSFS, we discussed how the 2.5 release of the Lustre file system is taking the requests and requirements of a growing set of potential enterprise Lustre users to heart. Among the key feature sweeps with this release are the addition of hierarchical storage management (HSM) capabilities, which directly addresses demands from the commercial community.
Shipman says that Lustre has finally hit a point in its technical maturity, resiliency and manageability where users outside of HPC are able to consider it as an option to support their growing data-heavy infrastructure. “Enterprise and traditional HPC are being confronted with many of the same data challenges that used to be limited to HPC sites,” he says. “It’s those forces—one being increasing data volume and the need to be able to process those volumes in a short amount of time using scalable algorithms (running on Hadoop or in MPI environments, for example) but that is mature and easy enough to use that it can be managed outside of that small cadre of HPC experts that most companies don’t have access to.”
He points to his work at Oak Ridge National Lab (his other role is directing ORNL’s compute and data environment for scientific projects) as an excellent example of this shift. Although it’s certainly an HPC-oriented center, Shipman says the team there is now using Lustre where they would have traditionally gone with scale-out NFS in support of workloads that look a lot more like “big data” enterprise workloads than HPC. The fact that it’s now much easier to manage, he says, is one of the keys, not to mention its reliability for their purposes.
Shipman says they’re using Lustre on these enterprise-like workloads to store high-value datasets that have to be available around the clock with full confidence that they can grab their data when they need it from within Lustre itself—a notable feature that the newest update offers.
Feature enhancements like the core one for the 2.5 release of Lustre, the addition of hierarchical storage management (HSM) resolve one of the key pain points for Lustre users—especially those on the enterprise side, many whom have made this a requirement for their adoption, says Shipman.
In essence, the value here is that from directly within Lustre, the update allows the files that are available in large-scale tape (or even other NFS storage systems) integrated within the same Lustre namespace users aren’t burdened with the constant concern of where their files are (on low cost/performance storage, tape, high performance storage). They are able now to manage in an integrated way on the same namespace and can automatically migrate from Lustre into, say for example, a tape archive or large-scale scalable NFS in a manner that’s transparent to the user. When the user comes back to access those files, the policy that they’ve set (whether it’s two months and then into tape, etc.) users can pull that data from wherever it is directly from within Lustre, it will stage those datasets into the Lustre file system and the processing lifecycle can begin anew.
Shipman says the addition of HSM is the direct result of their annual gathering of end users on both sides of the fence (academia and commercial) where new requirements and needs are shared. These meetings have produced other features that have bolstered Lustre’s wider appeal. Shipman says other improvements that have gone before include a solution to the problem of Lustre’s inability to scale the namespace and metadata performance. The fix here was the more recently-added DNE (distributed namespace—an OpenSFS-funded project) that takes aim at offering a more solid solution for big data enterprise users. Other features, including the online file system checker have responded to other enterprise-driven requests.
At the end of the day, however, Shipman says what is really driving Lustre’s lengthened reach into a more diverse range of environments is the vendor ecosystem’s commitment. The addition of new features are further enhanced by a large number of vendors who pull from the OpenSFS community code and create turnkey Lustre-based solutions that come ready to run on day one—a far-flung dream for those interested in Lustre even a few years ago.
As the OpenSFS community seeks to add an increasing set of features, including some around boosting single-client performance to round out other enhancements (metadata scability, data replication, online file system checking and now HSM), it will be interesting to see what commercial areas seem most drawn in by the shine of Lustre.