Innovation is a concept in danger of being permanently devalued by overuse and relegated to the realm of marketingspeak. As our Innovative Technologies special report in the following pages shows, what IT managers see as innovation has nothing to do with hype. It's all about making their companies run more effectively and efficiently.
And sometimes innovation is based on something as prosaic as a new standard, like the one under consideration right now that could literally become a blockbuster for enterprise storage. The T10 Technical Committee, part of the International Committee on Information Technology Standards, is working on the Object-Based Storage Device Commands (OSD) standard. OSD will turn files, records, directories and other storage elements into objects that storage management software can access using an extended SCSI-3 command set.
Storage managers (think file systems like New Technology File System, or NTFS, and databases like Oracle Corp.) need no longer write and manage physical data blocks. That job will move into the SCSI storage device itself, along with the metadata and attributes required for the device to internally manage those stored objects.
What's in it for you? By breaking away those block-level and metadata management chores, OSD reduces the storage manager's job to one of simply mapping files and file structures to objects. That should lead to better interoperability and set the stage for convergence of storage-area network (SAN) and network-attached storage (NAS) technologies. OSD-compliant storage managers will be able to share data on the same device. Theoretically, backups initiated on a Windows file system could even be restored to a Solaris file system (in the unlikely event that both could agree on a common, shared attribute set) because the data and its attributes have been effectively separated from the operating system.
So OSD makes it possible for vendors to create a universal file system (which operating system vendors will surely resist), and brings a cross-platform capability that today's proprietary SANs lack. OSD also logically separates control and management information from the data path, which means applications don't suffer the latency penalties associated with going through an intermediate file system, such as a NAS filer.
OSD essentially combines the direct-write performance of SANs and the cross-platform benefits of NAS. And moving object metadata and attributes out of the file system also allows for more scalable storage, eliminating the file server or NAS filer head as a scalability choke point. The argument over whether to serve up NAS-style files or SAN-style blocks goes away: Object-based storage should drive convergence from the ground up.
Another potential benefit of OSD is its ability to improve data security. Today, disk-level security is limited to crude tools such as LUN masking and zoning. OSD will make this more granular. "Because objects are self-describing, you can ascribe a security domain to each one," says Michael Mesnier, a storage architect at Intel Corp. and co-chairman of the Storage Networking Industry Association's OSD Technical Working Group, which developed the specification. With OSD, the user (called an initiator) must present a key to the storage device before it grants access to the requested data. The file system determines the user authorization and key distribution methodology; the OSD-based device enforces t.
While OSD will drive convergence, it won't make the SAN/NAS debate go away anytime soon. That's because the specification doesn't address how data should be transported. Both architectures will continue to evolve as object-based storage emerges, but distinctions between them will blur.
IT isn't likely to see the benefits of object-based storage until a new generation of intelligent (and more expensive) OSD-complaint SCSI drives and file managers appear. That will take time. For example, Microsoft Corp.'s long-promised object-based file system, code-named Longhorn, will morph NTFS into an object database. But it won't arrive until at least 2006, and its support for OSD is unclear.
Nonetheless, Mesnier predicts a base OSD standard by year's end. Intel is building an open-source reference implementation, which Mesnier says it will release this fall. The question that remains is whether vendors will run with it.