High-end clustering users contemplate snapshots
Some users and analysts say the line between high-performance NAS clusters and enterprise NAS systems continues to blur as clustering vendors begin to include enterprise features.
Traditionally, systems like Panasas' ActiveScale clusters have been squarely targeted at the niche market of high-performance computing (HPC), such as university research, and oil and gas exploration. Many of Panasas' customers remain in that market space and use the product for "run and done" high-volume processing applications.
But according to Larry Jones, vice president of marketing for Panasas, the company is also seeing users with new applications, like engineering and even stock trading, wanting to use the system for not only high-performance processing, but for re-evaluation of data after initial processing and storage for the future.
![]() |
||||
|
![]() |
|||
![]() |
Not all of Panasas' existing customers said they'll necessarily go for the data protection features. According to Gary Grider, group leader of high-performance computing systems integration at Los Alamos National Labs, his shop will stick with its current backup method -- letting each scientist using space on the system extract files for backup elsewhere.
"We do not know what data of the hundreds of terabytes (TB) per day that are generated by our simulations should be protected, and we also don't change much data -- we write new information," Grider said. "Almost all simulation oriented supercomputer sites use the archive method -- snapshots are a reasonable solution for some things, just not our use."
Meanwhile, however, another Panasas HPC beta tester said he was interested in purchasing the system and using the snapshot capability for at least one project.
According to Wesley Emeneker, graduate research assistant at Arizona State University, the 7 TB beta system he used for testing would most likely be put to use in production for processing and studying high-resolution photos of the moon. Over the next few years, the system could grow to over 1 petabyte (PB), but the data would also need to be stored and re-evaluated over the course of the study.
"I think there is starting to be some convergence of application needs between the large enterprise and high-performance computing," Emeneker said. "Stable storage is becoming more popular in HPC shops, and intensive data mining is popping up more and more within big corporations."
"There are some high-end NAS customers becoming interested in our products, for sure," Jones said, though he emphasized that Panasas is not actively trying to make a move down market. Still, he admitted, "As we talk to larger companies in, say, the energy delivery market, or large banks and telephone service providers, we're seeing that they are thinking about their data centers as big factories, similar to HPC users."
Right now, however, something none of the NAS vendors do -- at least not very well -- is support the NDMP, which separates the data path from the control path, allowing for local backups alongside centralized management in NAS systems. But Isilon has promised the capability will be updated for certification with major backup vendors. Jones said Panasas currently prefers to use the multiplexing features in the major backup software products instead of NDMP, but that the capability is also on his company's roadmap. Jones called NDMP "a checklist item for NAS systems" -- raising the question once again of his company's intentions toward the enterprise NAS space.
"We're still mainly marketing to HPC organizations and have no immediate plans to change that," Jones said. "But who knows what we'll see a couple of years from now?"
Analysts weigh in
Meanwhile, analysts said the recent announcements from Panasas and Isilon (with more to follow -- both Ibrix Inc. and PolyServe Inc. confirm snapshots are on their roadmaps -- show the lines between HPC and enterprise NAS blurring, whether the vendors are officially targeting different markets yet or not.
"Customers that have storage products from [these] vendors … want to use these systems for applications that are more mainstream, and these storage systems are also finding new markets. The vendors are getting themselves closer to more transaction-oriented applications," said Tony Asaro, analyst with the Enterprise Strategy Group. "This is certainly true for Panansas since they have architected their system to handle all I/O types, including streaming, large files and smaller transaction-oriented data."
Moreover, experts said, even as HPC systems add enterprise features, on the flip side, HPC systems are already working with features that enterprise systems are struggling to make possible; chief among them object-based storage. Panasas' system today is object based, splitting the file system at the directory-tree level across many storage blades, supervised by a metadata director.
"The line will definitely continue to blur," said Joseph Martins. "In addition to recent announcements, such as Panasas snapshots for HPC, one need only look at the high-end NAS offerings and announcements to see advancements in clustering, parallel processing, provisioning and load balancing. NAS players want a bigger piece of the high-performance market as well."