Current Hard Drives Hitting Limits, Change Forecast
That's the word from one of the storage industry's leading technologists who this week discussed storage and data center trends with a roomful of storage administrators and CIOs at Nth Generation's Summer 2006 Technology Symposium, held this week in Anaheim, Calif.
Nth Generation is a San Diego-based solution provider and one of Hewlett-Packard's biggest partners in the Western United States.
Richie Lary, an independent consultant who once helped define Digital Equipment's VAX computer architecture and the storage line that was eventually acquired by Compaq and then HP, looked at expected hard drive development over the next 10 years and said he expects the typical 3.5-inch drive to have a capacity of 3 Tbytes to 7 Tbytes.
To get to that capacity, however, will require some rethinking of hard drive technology because of how complicated hard drive technology has become.
"I like to compare hard drive technology to a Boeing 747," Lary said. "If you scale a hard drive to the size of a 747, it's like flying the 747 1/100 of an inch off the ground at 2,500,000 miles per hours, with the pilot able to read the spots on a set of dominos."
The main problem with improving hard drive technology is that areal density, which is the number of bits of data that can be stored in a given space, is hitting a wall thanks to the superparamagnetic limit. That is the point at which the bits are so close that turning a bit on or off can affect the state of a neighboring bit.
The industry has been pushing the superparamagnetic limit higher for years by using "squarer" bits to increase density, but density boosts from that technology, which depend on shrinking the size of certain ICs, are leveling off even as IC miniaturization trends level off, Lary said.
The latest technology in hard drives, vertical recording, under which the bits are lined up vertically on the media instead of horizontally, helped push density up last year, Lary said. "It's been the technology of the future for 15 years," he said. "In 2005, manufacturers started shipping drives with it. But it's only a one-time jump in density."
Hard drive vendors are also looking at Heat Assisted Magnetic Recording (HAMR), Lary said. With HAMR, the drives will have a laser to heat up a tiny spot on a new type of hardened media in order to let a bit of data be written before that spot cools down. Lary said Seagate estimates that technology will be available in 2010. But then he looked at Steve Sicola, a friend of his and the vice president of advanced storage architecture at Seagate, and said, "Steve said that's a little iffy for 2010."
Another key fact about disk drives is that increases in capacity have outstripped increases in performance and will continue to do so in the future, Lary said.
That is pushing the hard drive industry to move to 2.5-inch hard drives over time from the more typical 3.5-inch hard drive seen today, as the smaller size platter, while having less capacity, allows greater performance. "3.5-inch hard drives shipping today already are coming with 2.5-inch platters," he said.
Lary said that vendors are also developing 30,000 rpm hard drives, which double the speed of today's fastest drives. However, he said that the power requirements to get that speed are huge, and so vendors may settle for 22,500 rpm drives for now.
Other new technologies still being developed to improve the hard drive of the future include vertical MEMS (Micro-Electrical-Mechanical Systems) and termomechanical storage. With vertical MEMs, the single hard drive read-write head moving across the spinning platter is replaced by thousands of tiny heads on a single chip that moves very little. Thermomechanical storage allows the writing of data at the atomic level, as demonstrated last year by IBM.
Lary said holographic optical storage also shows promise and may be commercially available this year, but performance is very low.
While hard drive vendors are looking at new ways to improve their drives, others are looking at ways to improve the performance and data protection characteristics of RAID, Lary said.
The main problem now is that RAID technology was developed at a time when a 1-Gbyte drive was a big drive. RAID protects data by making sure that the data is written across several drives so that if one drive fails, the data is available and can be used to rebuild the failed drive.
Unfortunately, as hard drive capacities improve, the chance of double failures, where two drives fail at the same time, increases because of the time needed to rebuild the new large-capacity drives. "Bigger disks mean larger rebuilt times after each failure," Lary said. "And one hard drive error stops the rebuild."
That becomes a major concern in enterprise data centers, where the large number of high-capacity drives is causing the mean time to data loss (MTTDL), which Lary said is a better measure of reliability than mean time between failure (MTBF), to continually fall.
This is especially true as enterprise data centers increase their adoption of SATA drives, which since they are based on desktop drives, emphasize capacity and low cost over performance and reliability, Lary said. "If you have one petabyte of desktop drives with RAID 5, you could lose data twice a year," he said.
The latest RAID technology to help counter that problem is RAID 6, under which two hard drives are kept as hot spares so that the loss of two drives at the same time should not disrupt operations. However, RAID 6 has the side effect of increasing the number of reads and writes for a drive by 50 percent, which increases the wear over time.
Other technology to supplant RAID 6, such as WEAVER codes, is also being developed, but can result in a doubling of the amount of capacity used in order to protect data without impacting performance, he said.
While hard drives typically come with five-year warranties, Lary said, "nobody keeps hard drives for five years. But in the future, as hard drives don't become obsolete as quickly as now, people will keep them longer, and so the failure rate will be higher."
The other problem with RAID, Lary said, is loss of sleep. "If 6 percent of hard drives fail, you are going to lose sleep until the rebuild is done," he said.
John Randall, business development manager at Nth Generation, who took the stage after Lary, said that as a marketing guy, he understands what Lary is talking about when he talks about RAID 6 and WEAVER problems.
"I understand that that if you want RAID 6 for 1 petabyte of storage, you need 2 petabytes," Randall said. "And remember, you can get that from Nth Generation."