Hewlett-Packard Co. is boosting its storage grid architecture by enabling applications to see a single file system image across connected servers and devices within Linux environments.
The Palo Alto, Calif., company will unveil the HP StorageWorks Scalable File Share (SFS) system Wednesday at the International Supercomputer Conference in Heidelberg, Germany. Due in the third quarter this year, the self-contained file server is capable of distributing files in parallel while sharing bandwidth from dozens to hundreds of clustered servers.
HP SFS is built on the company’s Proliant servers and StorageWorks disk arrays and contains the open-source Lustre protocol, with new code from HP to make the standard more resilient and automated, officials said.
Lustre reports out how to partition file servicing into different areas to reduce I/O bottlenecks and enable scalability. HP “smart cells” are then used as building blocks to create immense storage grids by adding modular chunks of storage capacity or bandwidth to the client on the fly without impacting the system, officials said.
HP SFS could be a boon to customers running into a brick wall using the Network File System (NFS) protocol, which is incapable of supporting Linux cluster scalability requirements, said analysts. NFS tops out at about 10 terabytes compared with SFS’ ability to scale to 200TB or 300TB and use of Lustre to add bandwidth of several hundred megabytes per second.
Running HP SFS since last August to help administer a 53TB file system at the heart of a supercomputer featuring 2,000 processors, Scott Studham, associate director of advanced computing at Richland, Wash.-based Pacific Northwest National Laboratory, said the tool’s performance and scalability is “unmatched.”
“We have 1,000 clients all writing to it at the same time. NFS has problems if you try to do that,” Studham said. “[Plus] it’s open source, so it gives us the assurance and ability to see the code and make modifications.”
Studham said his organization is starting all large-scale file systems with Lustre for anything over 10 terabytes.
Eventually, HP SFS will plug into HP’s Reference Information Storage System (RISS) for archive services, HP officials said. Next up, HP is building a block serving component, similar to a traditional array controller, to round out its storage grid vision.
SFS was developed by HP, the U.S. Department of Energy and Cluster File Systems Inc., which served as the principle developer of the Lustre architecture.