Flash memory is catching the notice of IT storage pros, whether from dealing with the increased use of USB thumb drives by the mobile work force or pondering the speedups promised for Windows Vista. However, recently the flash talk has taken a twist: RAID.
At the recent meeting of the San Francisco SNUG (storage networking user group), the combination of flash and RAID cropped up. The presenter, Diamond Lauffin, senior vice president at Nexsan Technologies, described a hardware project he had put together: a 2GB array made from USB thumb drives.
“If you had some files that you wanted really protected, you only need buy a couple of the [memory] sticks, put them into a little USB hub, and stripe a RAID process. Now, you have a nonvolatile flash memory RAID device—for only $100,” he said.
Beyond the tinkering interest, Lauffin held up flash-based RAID as an example of how the storage industry is quick to label a new technology solution as “impossible.” He pointed to the time (not so distant) when big companies and analysts pooh-poohed the economics of D2D (disk-to-disk) backup over tape. Of course, D2D is Nexsans business. (Lauffin later in the meeting discussed the integration of D2D, compliance and archive management into a single appliance.)
He noted the recent shipments of flash-based drives for notebook computers, adding that these products are way more expensive than ordinary hard disks. And they have about half the capacity at that higher price. So, the drives appear are at a disadvantage when measured by price per gigabyte (or terabyte).
Still, Lauffin said the march of technology and manufacturing efficiencies can work wonders to such “impossible” economics.
“Do we think that within two, three or four years we will not have flash memory in all of our RAID boxes and have it cost less than current moving disks? Do we think that thats impossible?” he questioned the crowd.
At the same time, Lauffin neglected to mention that there are some vendors offering currently solid-state RAID and drives, such as BitMicro Networks. The company offers a range of “E-Disk” solid-state drives with various server-centric interfaces such as Fibre Channel and Ultra320 SCSI. (There are some additional angles to these drives that we will explore in an upcoming column.)
Still, the RAID flash proposition isnt as simple as Lauffins ad hoc USB thumb drive array. Flash is a much different technology than rotating media and the reliability metrics of a particular piece of flash memory is very dependent upon vendor implementation and the expected usage of the device.
For example, at Augusts Flash Memory Summit conference in San Jose, Calif., Intel scientist Knut Grimsrud, director of storage architecture for the Intel Storage Technologies Group, expressed concern over the varying quality of wear-leveling algorithms used in flash memory devices and the performance of multilayered flash memory in enterprise applications.
Flash memory is susceptible wear and flash vendors employ “leveling” algorithms to spread out erase cycles across all the cells on the flash memory as well as managing a set of spares for use as the device ages.
Grimsrud said that the effective wear rates for flash memory his group had tested were half that of the rates listed on the manufacturers spec sheets. He said that Intel tested various memory modules when developing its Robson logic board caching technology due out in mid-2007.
“The rated number of erase cycles … missed by a mile. This [result] was not an isolated case and [the memory] wasnt in thermal stress either. Were not assuming that any data thats on the spec sheet is right,” he said.
This shorter life span wasnt such a problem with applications such as MP3 audio or digital photography, where the actual number of read-write cycles is low, Grimsrud noted. And the importance of the data is not so important for an individual song or image.
However, with enterprise applications on hosts or servers, the integrity of data is paramount, he said. (Yes, this is a very debatable thesis—some folks say that consumer storage needs to be more reliable since the expectation is greater.)
Of course, Intel is a NAND flash manufacturer. Vendors at the conference said that it was no coincidence that Intel decided to certify only its own flash modules for use with the Robson technology and found that all the rest of the industry came up short.
Still, Grimsruds point is valid: wear-leveling algorithms vary from vendor to vendor and the usage model for enterprise storage is different than the one for consumer applications.
This is something we all might want to keep in mind when building that thumb drive RAID project and running it for any length of time. Not all USB flash dongles are alike. (Actually, the whole thing sounds like a user accident waiting to happen. I can just hear someone saying: “You had so many of them, I borrowed a couple.” Doh!)
Next Page: Flash standards missing in action.
Flash Standards Missing in
Action”>
Grimsrud also said the current state of flash standardization is troublesome for hardware companies that want to extend its use outside of consumer applications. He reported on Intels involvement with standards groups to fix the problem.
“Right now, the use of flash is a nightmare. Everyones flash behaves a bit differently. Even with flash from the same manufacturer, the next [generation of] flash behaves differently,” he said.
According to Grimsrud, the development times for systems is longer than that of simple storage devices, such as the USB keys we all use. By the time a design is actually moved into production, the expected flash component may be superseded by a newer product with slightly different characteristics. He called this “design-time pre-association.”
“By the time you come out with a compute platform, the flash that was available when the design was done is obsolete and not fully relevant any more. Theres good consistency among manufacturers, but its not like horseshoes,” Grimsrud said.
Intel is supporting the Open NAND Flash Interface Working Group, which aims to create a standard interface for flash components.
This is different than a card standard; this is to improve the design and integration process for hardware and software components. Its the way the rest of the storage industry has operated for some 15 years. He said a draft standard was expected in the fall.
Despite the near-term concerns, Grimsrud was upbeat on flash in the enterprise. He predicted a “new golden age of computing where flash is the next big thing.”
Again, its good to remind ourselves that this prediction was offered to a room of flash manufacturers and vendors. But as Lauffin said, the difference between impossible and possible (reality) is usually something on the order of four to five years.
Yet, one thing that shouldnt happen in the flash market is the move to make some flash brand names generic in the way that Coke and Xerox are sometimes used to describe a cola beverage and paper copy, respectively.
I was shocked to see that the Wikipedia appears to acknowledge some generic-ization of JumpDrive, meaning a USB flash drive. JumpDrive is a brand of Lexar Media.
TechTargets Whatis.com also presents this term.
Its an outrage. In all my years, I have never heard anyone, professional or user, use “jump drive” as a generic term. I have heard USB flash drive, thumb drive, keychain drive and key drive. But “jump drive?” Nope.
Lets hope this brand-grab stops now.
What do you think? Is flash coming to a RAID near you? And what should we call USB flash drives? Let us know here.
Check out eWEEK.coms for the latest news, reviews and analysis on enterprise and small business storage hardware and software.