HP executives describe the StorageWorks 9100 Exteme Data Storage System as a large appliance consisting of HP DL460 blade servers and external JBOD storage, tied together with
The ExDS9100 is aimed at Web 2.0 companies -- firms managing large amounts of storage, usually for Web-based services or applications, such as photo, video or data hosting. These types of organizations have been getting lots of attention lately from storage vendors, as this influx of data stored online places more emphasis on file-based systems.
Meanwhile, IBM is also planning a block-based clustered offering based on its January acquisition of Israeli clustered-block storage startup XIV Ltd.
HP plans to go after those Web 2.0 shops with a system based on commodity components from its own product lines, which it claims will be less expensive than putting whitebox components together like Google's PC-based computing farm. Other vendors in this space include NetApp, Isilon and Data Domain Networks.
Ian Duncan, director of marketing for HP's scalable NAS division, estimates that organizations typically spend between $4 and $5 per gigabyte building out their own Internet-scale data centers using whitebox servers. He said the ExDS9100 will price out at less than $2 per gigabyte. "Infrastructure costs are critical in a pay-per-use market," he added.
The ExDS9100 is meant to scale to hundreds of petabytes. The minimum configuration is a cluster of four blades and 246 TB of storage, for a total of "under $500,000," Duncan said.
From that minimum, customers can add storage in 82 TB blocks. These blocks are groups of SAS-connected JBODs that pack in as many 3.5-inch 1 TB midline SAS drives as possible into each 1U subunit. The PolyServe software gives the blade servers parallel access to the storage blocks. The blades can either control the storage or run applications. An additional layer of management software will allow customers to add storage blocks without disruption and are intended to mask the complexity of the underlying system.
Duncan said customers can respond to spikes in performance demands by adding new blades on the fly. HP claims the performance of the system will scale linearly, from 800 MBps in the four-blade configuration to 3.2 GBps in the 16-blade configuration. By comparison, Isilon claims that its most recent X Series clustered NAS system offers up to 10 GBps throughput.
So far, the HP system does not offer quality of service for individual data streams or embedded transcoding for video applications. Other clustered storage vendors, such as Omneon, offer embedded transcoding, but Duncan said HP has never received a request for it. Storage blocks can be added while the system is running, but data is not restriped across the entire new capacity -- at least not in its first release. Instead, the blade servers access "a big gloop of homogeneous storage," according to Duncan.
The HP system employs an architecture that is physically different from many of today's clustered storage systems, which typically pair processing power and storage in individual building blocks.
However, according to Schulz, it's a common misconception that storage and processing power have to be paired in modular hardware to have a clustered system. "The trick to making something clustered, particularly for NAS, and that's what this is, is the software and how it presents itself to the application servers," he said. "HP is using off-the-shelf components to accomplish the same thing Isilon does."
There is some debate about this definition of a cluster. "You could argue that because each blade, when added, becomes part of one system, that it's actually not a cluster, if you see a cluster as being made up of separate systems," said analyst Mark Peters, Enterprise Strategy Group. "It's a very fine semantic line."
Another semantic dividing line questioned by Peters is the distinction drawn between Web 2.0 and typical commercial businesses, including HP's conversations around ExDS. Peters said this type of product could also find its way into many types of data centers, if recent trends continue. "Everyone is capturing more digitized information, no matter what business they're in," he said. "Unstructured file data is growing dramatically everywhere."
IDC estimates that just over half of the disks shipped in storage systems this year are deployed to support some file-related data. By 2012, more than 80% will be used for files.
"It's great to see HP develop a more packaged route-to-market for PolyServe," said IDC analyst Brad Nisbet. "While many of the pure content-driven organizations are willing to take on solutions that entail custom assembly of parts, many are looking for prepackaged solutions. This is even more for those enterprise organizations that have a blend of needs between legacy block/database data and growing volumes of file content."