Imagine a storage device that uses its own horsepower to manage data, requires no manual settings for security and doesn't care if the client server speaks in blocks or files. That's the promise of object-based storage. Object-based storage technologies shield the application or operating system from the low-level details of managing file storage. In one method, intelligence is added to the storage device in order to offload low-level storage management tasks traditionally handled by the operating system, such as mapping files to actual storage blocks on the disk drive and managing file attributes and other associated metadata.
Although widespread use of object-based storage is still some years away, the technology could result in storage systems that are more scalable, reliable, secure and manageable.
The T10 Technical Committee, which is part of the Washington-based InterNational Committee for Information Technology Standards and the Mountain View, Calif.-based Storage Networking Industry Association (SNIA), is working on a specification for object-based storage, called Object-Based Storage Devices (OSD). OSD turns files, directories and storage-related elements into objects that storage management software accesses using an extended SCSI-3 command set.
"But SCSI is just one component of what we're doing," says Michael Mesnier, a storage architect at Intel Corp. and co-chairman of the SNIA OSD Technical Work Group. "We're also looking at a more general-purpose definition of object-based storage which is irrespective of the transport, which means you can run it over SCSI, you could run it over Fibre Channel . . . over TCP/IP or whatever. To me, that's a much stronger impact."
By putting some of the intelligence for accessing objects into the storage array instead of the application server, networks could be infinitely scalable because servers would no longer have to eat up bandwidth searching for and accessing files or blocks of data.
"Just like you could plug a different hard drive into your PC, you could add another server to a storage system in the same way," says Scott A. Brandt, an assistant professor at the Storage Systems Research Center at the University of California, Santa Cruz (UCSC).
UCSC's Jack Baskin School of Engineering is designing a high-performance storage network, based on commodity hardware, that can store up to 2 petabytes of data based on the proposed OSD model.
"While you're still dealing with blocks, they're hidden from the file system," Brandt says. "As you add more storage, you're adding more smarts. What might have been prohibitive details added to a large system are now details handled by the storage device itself."
Moving the object metadata and attributes out of the file system also eliminates the file server as a scalability bottleneck, Brandt says.
OSD makes for a much more efficient I/O configuration, says Mark Bradley, a technology strategist in Computer Associates International Inc.'s BrightStor unit. "You're no longer having to pass all these low-level read/write blocks back and forth between a file system and a device," he says.
For example, Bradley says, a file system could say, " ëI need foo.bar,' and foo.bar comes. Therefore, your communications over the interface, whether it's a network interface or not, becomes much less complicated and takes up less bandwidth, and in turn creates far fewer errors."
In April, EMC Corp. introduced what it calls content-addressed storage, based on a new storage server called Centera. Experts believe Centera is one of the first true object-based storage arrays.
Centera is an array that also handles all storage management issues by assigning each stored file a unique file object identifier that it passes back to the application. The application then requests that identifier to obtain the file, and the appliance takes care of the details of where and how the file is actually stored.
Mesnier emphasizes that the proposed standard is not a completely new model. "It's just taking [Network File System], in a sense, and applying it to what used to be considered dumb peripheral devices," he explains. "Imagine a world where two different file systems agree on data, then go directly to the same storage device and share the same data."
He says that the biggest benefits of object-based storage will come in the form of increased, more granular security and data-sharing capability.
"If I look at a block-based device today, an individual block cannot be protected," Mesnier says. "Once you get access to the entire device, you can read or write to any block you want to. You could format the device if you wanted to. With object-based storage, you can ascribe a security domain to each and every object."