Sun Offers More Storage to OpenSolaris
Page 1 of 1
Sun will continue to pursue internal object storage development for the block-based ZFS and object-based Lustre file systems and the ADM hierarchical storage management system (HSM), but Lustre doesn't use the standard ANSI T10 OSD interface.
The move reflects Sun's growing focus on ZFS as part of its open storage efforts.
Block-based SAM-QFS development efforts will continue inside Sun. "Development will continue on the block-based file systems and archive software for enterprise users in our core markets," said Plant.
Sun's shift in strategy also appears to be a departure from the T10 OSD vision.
Plant noted that OSD "is a proposal which provides object storage through, in essence, a more intelligent storage device, such as a disk that supports more advanced T10 commands. Sun's open storage strategy is fundamentally based around a different set of principles, wherein customers can take advantage of a set of inexpensive, simple, commodity technologies to build a physical storage infrastructure, and then intelligence and differentiation is provided through open storage software. The result is that customers experience lower barriers to entry and exit, and the cost of physical storage is minimized."
Plan cited ZFS as an example, which he said offers a complete set of RAID functionality in software using general purpose compute resources, "thereby eliminating the need for proprietary RAID cards that are today less efficient than the RAID provided by ZFS and increase the cost of an overall storage solution for customers.
"ZFS is able to innovate across layers of software in ways that would not be possible if storage intelligence existed only below the device layer," he said. "These principles apply across our Open Storage products, including the Sun Storage 7x10 series, and our core ZFS, Lustre and QFS assets."
Plant noted that while OSD protocol commands could be virtualized and delivered over virtual block protocols like iSCSI, "our sense is that this is an inappropriate model because it corresponds to a hardware device model that we not believe is in the interests of best storage economics. And we believe it is for these reasons that OSD proposals, which date back more than ten years now, have not gained any market adoption."
Slow Uptake for Object Storage
OSD has long been viewed as a significant move beyond the communication, scalability and security limits of block and SCSI technology, but has yet to catch on beyond high-performance storage environments.
Brent Welch, director of architecture at Panasas, which has been a pioneer in both OSD and pNFS development, noted that "introducing new storage standards is a slow process. Even something like iSCSI that was only a slight variation on an existing standard took a long time. OSD is a fairly radical standard by comparison, and we expect widespread adoption to take time."
Panasas has used OSD and pNFS to develop a highly scalable parallel file system. "Sun doesn't have products in this area, and in fact it acquired Lustre (CFS), which is also an object-based file system," said Welch. "However, Lustre doesn't use the standard OSD interface, but an 'object-like' interface of its own design."
Panasas has also invested in open source Linux implementations of OSD targets and initiators, and a freestanding file system called exofs that runs over OSD. "Our goal is to foster more research and experimentation in this standard interface," Welch said.
Indeed, the OpenSolaris community could help spur OSD development but then Sun would need to include the community code in enterprise editions of SAM-QFS. The SAM-QFS OSD effort could thus prove to be a good test of the flexibility of Sun's open storage vision.
pNFS also includes block, object and file access as part of its back-end protocols. "Things are slowly crystallizing around some new storage interfaces," said Welch. "But storage is so fundamental that we expect these things to take time."
John Lohmeyer of LSI (NYSE: LSI), chair of the T10 INCITS SCSI Storage Interfaces Technical Committee, said that OSD "seems to offer a level of abstraction that should be compelling for some users. However, I think most OS vendors do not have much of an incentive to adopt OSD; they would be delegating an important part of their operating system to other vendors.
"Also, since OSD has been developed mostly in an academic arena, it seems to allow a lot of options," said Lohmeyer. "This can make it difficult to find the common ground for a standard interface. I think OSD will remain mostly of academic interest until some class of users with clout demands this interface and pushes vendors for a specific subset of features."
StorageIO founder and senior analyst Greg Schulz said T10 OSD enjoys "a passionate following in the HPC, academic and extreme computing environments, but it still remains elusive for general purpose and commercial computing environments. ... Given time, somewhere in the future, OSD may finally live up to its expectations, however, for now, object-based storage systems such as those that support access via various application object interfaces, including DICOM for medical, REST, SOAP and XML, among others, continue to gain in commercial adoption due to their ability to support common interfaces with existing underlying technologies."
SAM-QFS was developed by LSC and acquired by Sun in 2001. Sun released the SAM-QFS source code to the OpenSolaris community a year ago. QFS can scale from 1 to 128 nodes and has "virtually has no limit to the amount of information managed," according to a Sun blog posting.
This article is courtesy of Enterprise Storage Forum.