No sooner have we got used to the idea of SCSI hard drives being replaced in enterprise arrays by serial versions of the once PC-focused ATA spec, than it's time to change back to SCSI - though once again, in a serial-attached (SAS) flavour.
So says Mark Ward, the CEO of storage array developer Copan Systems. Drives will also get physically smaller, he adds.
"The balance [in enterprise arrays] has gone from 30 percent SATA to 70 percent in three years," he says, "But looking forward, the industry will probably make SAS a replacement for SATA, as the vendors believe that SAS will eventually be cheaper to manufacture.
"And 2.5 inch will replace 3.5 inch - at the moment 3.5 is still around twice as efficient, but they will cross over. We're also looking at solid-state - in three to five years it could be cheaper than hard disk."
Copan's speciality is low-power long-term online storage. It was one of the pioneers of MAID (massive arrays of idle disk), which keeps data online but turns off hard drives to save power when they're not being accessed. (If a drive's not been accessed in 30 days, it is powered up for a check.)
MAID to WORI?
This kind of storage is ideal for "persistent data - that's write once, read infrequently," says Ward. It's data which you won't access very often, but when you do need to get at it, you'll want to get to it easily and reasonably quickly. In the past, this has been pushed off to secondary tiers of storage, often in an optical disc or tape library.
"Tape was invented to address the problems of 30 years ago. Those problems have been superseded now," says Ward. "People tried tiers, but now they're all tiered out - all they're doing is adding complexity.
"Yet everyone agrees there's transactional data and there's persistent data. For example, a major US bank lets you view your cheque images online - so it needs storage like this.
"We have large customers in social networking too - there is incredible data availability demand, and that's persistent data. Also there's things like satellite imaging where you might need to pull up an image in a hurry - we have 7.5PB at the CIA for that."
Other examples of applications that involve persistent data include e-discovery and compliance, and persistent data might still need replication for disaster recovery and de-duplication to save space, Ward adds. But he says that while there's different ways of addressing it, the basic challenge is the same for everyone.
"We have the same drivers as everyone else - we haven't placed ourselves in a niche. It's a purpose-built system for copies of data. It has to be online so you can get it quickly when it's requested. Analysts say 70 to 80 percent of data is in this category."
He argues that where Copan has scored is in focusing very tightly on the characteristics and access patterns of that 70 to 80 percent - and in discarding the baggage associated with the 20 to 30 percent of transactional data.
"For random I/O, call EMC," he smiles. "For persistent data, we use one-sixth as many racks as EMC's densest Clariion. Our storage also runs at 10 percent of the power consumed because of MAID and the access patterns of persistent data."
The infrequent access advantage
Those access patterns are why the array density can be far higher. Ward says that Copan has canister technology that protects its hard drives from heat and vibration, enabling them to last several times longer, plus it has developed a proprietary ASIC that allows it to have 896 drives in an array.
However, none of that would be enough if it actually tried to run all 896 drives at the same time. The key is that it never needs to, because it knows that they will only be read infrequently, so it can adjust its array designs accordingly.
"We only ever allow a maximum 25 percent of our system to be powered up," Ward explains. "A customer might ask: 'What if we want 26 percent?' The answer is if you do, then it's the wrong data and you need to go back to your persistency definitions."
He adds that, as a veteran of EMC from the days when it still supplied compatible memory boards for minicomputers such as the DEC PDP-11 and VAX, he now finds the storage market as challenging and exciting as ever.
"It's déjà vu all over again - it's just like it was with EMC 15 years ago," he says, though he adds that some things have changed for the better. For instance, he jokes that in EMC there was an unofficial game among salesmen to see who could sell the most storage by persuading customers to store more copies of their data - in some cases as many as eight, nine, 10 or even 12 copies.
Now, cutting back on storage is the aim. "We are a growing company in a market that wants to save power and space and get more for its buck," he says.