At the SpectraSummit event in Boulder, Colorado, earlier this month, HPCwire learned of a developing partnership between storage vendor Spectra Logic and research software-as-a-service provider Globus to integrate Spectra’s BlackPearl Gateway storage into the Globus Cloud, creating an endpoint for BlackPearl that effectively enables an archive tape system to be usable by the Globus community.
Spectra Logic’s BlackPearl Deep Storage Gateway, introduced in 2013, sits in front of the deep storage tape libraries as well as the recently launched ArticBlue disk-based nearline system. An intelligent controller, BlackPearl uses an extension of the S3 cloud storage interface to enable a RESTful, object-based interface to deep storage. The appliance was designed for very large data sets.
Steve Tuecke, project lead and CEO at Globus.org and deputy director at The University of Chicago’s Computation Institute, explained that when Globus Connect service is dropped onto BlackPearl it creates an end point in Globus. Now all the functionality of Globus – the transfer, sharing, publication and discovery – “just works,” he said, and it becomes another storage tier within the user’s ecosystem.
Spectra and Globus have identified several primary benefits that come from this integration, including:
+ Consistent, familiar user interface/experience spanning the storage hierarchy.
+ Reliable, secure “fire-and-forget” transfer of massive data sets, optimized for tape.
+ Visibility into storage utilization.
+ Low administration costs.
Globus provides research data management software, delivered as a service (i.e., SaaS), directly from customers’ storage systems. The primary market for Globus is the non-profit research and education community — end users from universities and federal agencies, resource providers at campus computing centers and national computing facilities, and developers of research apps — however it also open to commercial use.
According to Tuecke, Spectra and Globus were drawn together on account of shared university customers who wanted to use Black Pearl as their tape storage management system and wanted Globus as the user interface. With the launch of ArcticBlue, there was a broadening of the opportunity by bringing near-line disk into the equation, Tuecke explained, this 10-cents/GB option to deploy storage with a long lifecycle.
Globus is hearing from customers that they have these mandates coming down telling them they need to have data available, but there is a real struggle with where to put all the data. “This is the gap that we are trying to plug,” said Tuecke. “We’ve got the software to do that, and what we’ve been looking for is the hardware to do that, something that’s just trivial to install and maintain, turnkey system, with really good cost points that scale nicely.”
“The online archival element is a really interesting opportunity for that mid-market HPC to have a solution that still meets their needs, i.e., inexpensive, meaning not a premium enterprise solution, and designed with cost in mind both to deploy and administer. As a combined solution, we’ve got the user interface and Spectra has the system that provides the backend.”
“The campuses are looking for that sub-hundred-dollars-per-terabyte-per-year cost to manage this long-term archive data and this is the first thing I’ve seen that will do that out of the box,” Tuecke continued. “It was already attractive (with BlackPearl) and now it got really interesting (with ArticBlue).”
As for market opportunity, Tuecke reported that his mission is the research world, which can be distilled down to 110 universities plus the national labs, the NASA facilities, and forth and to a big swath of that, he expects this “mid-market solution” to be very attractive.
SpectraLogic’s Brian Grainger was effusive in his description of Globus, calling it “one heck of a technology.” He joked that anyone in university research that is not looking at them must have unlimited budget. Further, the National Center for Supercomputing Applications and Argonne National Labs are both joint customers who are actively involved in the development of the BlackPearl client. Grainger offered this as a stamp of approval for everyone else to look at this new approach to managing data.
Teucke described how Globus’ work with some of the largest HPC sites, such as NCSA’s Blue Waters system, had them dealing with the really high-end tape-based HPSS systems and how this prepared them for the collaboration with Spectra. “We already had the optimizations in place to allow for things like waiting for tape to spin up and doing the tape ordering on the requests so we can optimize the tapes better,” said Teucke. “So it’s not just a dumb interface to a file system; we’re able to leverage and knowledge the fact that this is off-line media and ensure that big transfers in and out proceed smoothly. That will all be there day one with our BlackPearl integration.”
Teucke added that they are still finalizing the integration, but he said it’s really just a matter of installing BlackPearl and dropping in the Globus Connect software — “the little piece of software” that connects the storage system to the Globus cloud so that Globus can perform its management functions. This is the part they are finishing right now, I/O integration against the BlackPearl S3 APIs, which is trivial to install, according to Teucke.
More on Globus
Globus’ roots go back two decades to the 1995 Supercomputing Conference and an iconic grid computing demonstration led by Ian Foster. Foster was joined by Carl Kesselman and Steve Tuecke and the Globus Toolkit was born. The software has become a vital infrastructure component for many resource owners, HPC facilities, and research projects, including eXtreme Science and Engineering Discovery Environment (XSEDE), Open Science Grid (OSG), and many others.
Two years ago, Globus began building up a subscription model in-line with its mission to become a sustainable non-profit that was no longer subect to the inconsistency of having to rely on grants and other funding. While Globus still offers endpoint-to-endpoint file transfer at no cost to non-profit researchers, subscriptions unlock advanced features. For end users, this includes file sharing and data publication; and for admins, this includes a management console and usage report.
“Charging for high-value services allows us to fund the operations that support our users, and to continue delivering useful data management capabilities to the non-profit research and education community,” explains Globus on its website. Commercial Provider plans are also available.
Over the last year, Globus has also begun rolling out data publication capabilities in addition to the file transfer and file sharing components. In this era of ever-expanding data volumes, Globus recognized that new methods would be required for communicating, discovering, and accessing scientific data. Furthermore, agencies like NSF and the DOE are starting to mandate that data products associated with publications be published to support accessibility, verification, reproducibility and extension. Traditional systems, such as libraries, were sufficient for handling PDF files for publication, but a terabyte data set has different needs. Globus has been focusing on how to bring data publication capabilities to the community, noted Tuecke, so that when a scientist has a dataset they need to publish, the process is “trivial.”
The post Old School Grid Meets New School Storage appeared first on HPCwire.
