The Honeycomb technology, which will go on sale later this year, has two goals: to speed data-retrieval tasks such as searching and to make disk failure a nearly ignorable event, said Mark Canepa, executive vice president of Sun's storage group.
Key to the search-related promise of Honeycomb is descriptive information called metadata. When a computer stores a file, it includes some basic metadata, such as the time the file was written or modified, but Sun is among those pushing for vastly more elaborate--and therefore useful--metadata.
Undaunted by past flubs, Sun Microsystems is working on a new storage technology, "Honeycomb," that it hopes will speed data-retrieval tasks such as searching and make disk failure a nearly ignorable event.
With recent partnerships and acquisitions, Sun is in a better position in the storage field than it has been in the past, and Honeycomb's approach to metadata and information retrieval could further improve the firm's standing. Still, Honeycomb faces competition from giant EMC, Hewlett-Packard, StorageTek and start-up Permabit.
The disk failure side of the equation has to do with storing data across hundreds or large numbers of disks, far more than are found in currently popular RAID, or redundant array of inexpensive disks, systems. The idea behind both approaches: Don't put all your eggs in one basket.
At the end of 2004, Sun transferred the Honeycomb team from Sun Labs to its product group. Honeycomb technology will be brought to market both as new standalone systems and as an improvement to existing midrange models, Canepa said.
Sun and storage
Sun, which primarily sells servers and server software, sees the world somewhat differently from storage specialists such as EMC. Sun has argued for years that it's better to buy an integrated system rather than individual components that must be interconnected by hand.
But the company has had a hard time selling that vision when it comes to storage. A year ago, Chief Executive Scott McNealy said the storage attach rate--the storage revenue as a fraction of server revenue--. Yet from the last quarters of 2003 to 2004, that figure declined from 24 percent to 22 percent.
For some time, but it's had problems, with failing. In response, the company inked a deal to systems, for midrange systems and . In addition, Sun licensed technology and bought engineering services from storage specialist Procom Technologies in 2004.
With storage hardware, "They're in better shape than they have been for a long time," said Data Mobility Group analyst John Webster.
And metadata tools could be a very useful addition, especially for customers interested in processing data that's retrieved in real time, Webster added. For example, a retailer might want to link a stream of data logged bysystems that track product sales with a stream of video that records customers' actions in stores. Those two linked data streams, indexed by metadata that records the time the information was gathered, could be used to find out customer traits retailers want to know.
Not first to market
Sun isn't alone with Honeycomb. Its chief competitor is storage specialist EMC, which not only beat Sun to market with a competing product, Centera, but also whose software provides a necessary interface to make metadata useful.
EMC had more than 1,200 Centera customers in 2004, has lured software companies to build Centera support into 165 applications, and has sold a total of 30 petabytes of Centera storage capacity so far. "It's clear from our results that Centera has real momentum and is being embraced in the market. Customers are struggling with unstructured data and looking for help," spokesman Greg Eden said.
Sun believes it has an edge, though. Where EMC relies on a database housed on a separate server to store and process metadata, Honeycomb builds that function directly into the storage system, Davis said. Sun's approach is cheaper, simpler and doesn't require separate administrators, he said.
In addition, Sun once again will play its openness card. It will provide an open interface for metadata and work to standardize it so software companies and customers won't have to worry about tailoring products for different storage systems or for having data locked into a single suppliers' storage system, Canepa said.
Another competitor in the market for this so-called content addressable storage, or CAS, is start-up Permabit, whose technology is used in Storage Technology's Lifecycle Fixed Content Manager 100 system, introduced Monday. And Hewlett-Packard offers its HP StorageWorks Reference Information Storage System.
As currently planned, each Honeycomb system is a rack-mounted module 5.25 inches tall. More and more modules can be added to increase storage capacity, but that expansion doesn't require more administrative staff, Davis said.
The systems will userunning on Advanced Micro Devices' Opteron processor, he added. Opteron's fast input-output capability is useful for the task, Davis said.
Honeycomb will be incorporated later into the company's StorEdge 5210 and 5310 systems. Those systems use Intel's Xeon processors today, but "before the end of the year, those will ship with Opteron technology," Canepa said.
Thousands of disks
High data reliability is the other half of the Honeycomb promise. Today, RAID technology can store data across a group of drives such that no data is lost even if one drive fails. Administrators replace the failed drive, and a RAID system can reconstruct the data.
But with drives now reaching 500GB in size, "the rebuild process can end up taking days," during which there is a serious possibility of a second drive failure and irretrievable data loss, Canepa said.
Honeycomb aims to sidestep this by storing data across many drives; the system can withstand multiple failures without blinking, Canepa said. The result might eventually be a more relaxed and less expensive approach to storage system maintenance.
"If you think of having hundreds or thousands of disks, you may accumulate failures for two or three months, then go out on a periodic basis and replace those," Canepa said. "Or you may have enough capacity that you never replace them. After five years, you just migrate the data off of it."