Drive suppliers hit capacity increase difficulties
PMR or HAMR for your platter?
Hard disk drive suppliers are looking to add platters to increase capacity because of the expensive and difficult transition to next-generation recording technology.
There are two candidates to replace the current Perpendicular Magnetic Recording (PMR) technology. The first is Heat-Assisted Magnetic Recording (HAMR) and the second Bit-Patterned Media. PMR becomes unstable once the magnetised areas become very small - they are subject to influence from neighbouring bits, among other interference. HAMR uses a media formulation in which heat is needed to write data, thus reducing such influence. BPM has a ring of insulating material around each bit to deal with the same problem.
Conversations with various industry-people, including Xyratex CEO Steve Barber, create a picture in which PMR has to be kept in use for four to five years because of the difficulty and expense of moving to either BPM or HAMR.
We have heard from Isilon and Buffalo spokespeople that hard disk drive (HDD) manufacturers have been talking about 4TB 3.5-inch drives arriving in the middle of 2011. We have also heard that there are 3TB evaluation drives out amongst OEMs.
Currently we have 2TB 3.5-inch SATA drives which have four double-sided platters inside them, with a read/write head per platter surface. Seagate has announced a 3TB FreeAGent GoFlex external drive but has not said what sort of disk drive is inside it. Our understanding is that it is a 5-platter 3.5-inch unit.
There are two ways to increase HDD capacity: increase the areal density per platter or increase the number of platters. We rule out increasing the size of the platters as that would increase the vibration at their edge and make the maintenance of the correct read/write head height more difficult. A platter size increase is also against the grain of HDD evolution and we are currently seeing a transition to 2.5-inch units. These employ a glass platter which is stiffer than the aluminium ones seen in 3.5-inch format drive.
Xyratex supplies drive array subsystems to IBM, Dell, EMC and NetApp. It buys an large number of disk drives and knows what is going on in the HDD industry. Barber says one HDD manufacturer is telling Xyratex that 3.5-inch drives are dead, with 2.5-inch the future, due to rebuild times: "2TB drive rebuild times are heading towards a week."
Technology transition costs
Barber says the fact the that HDD manufacturers are discussing 3TB drives with their customers indicates that they can't double densities as they have in the past. The costs and technical difficulties of a move to BPM or HAMR technology are huge: "The next-gen technologies work in the labs but are extraordinarily difficult to move into manufacturing. BPM manufacturing costs are off the scale; they'll need 12 to 14 per cent of [HDD manufacturer's] CAPEX [capital expenditure]." The current CAPEX percentage devoted to PMR manufacturing is rather less.
What will drive manufacturers do? End-users are growing data at a 50 per cent compound annual rate and the world needs disk drives to store it. "HDD suppliers will put more heads and platters into drives or ship more drives."
But datacentre floor space and power supplies are both limited. You can't simply double the number of storage arrays. The conclusion is that we're going to see more platters and heads inside disk drives. He thinks we could see three or four platters inside 2.5-inch drives, possibly even five. The laptop 2.5-inch drive bay slot needn't constrain enterprise arrays. We could see one, two or three platters with 3.5-inch drives.
That means that drive slots inside storage arrays will have to become larger. It's conceivable that you could add just one more platter and maybe shrink the overall height of the plaster stack to keep the drive enclosure depth unaltered, but add another two or three platters and a deeper clamshell case is needed. Increased platter-count drives will also draw more power.
Barber reckons an interim capacity boost can come from 'shingle writing', which means overlapping tracks slightly, like shingle tiles on a roof, to cram more of them onto a platter surface. This will be done in conjunction with a larger sector size of 4K, but there will be a performance degradation because rewriting data means rewriting the whole track, not just a sector. Shingle writing is used "in external drives today … It's good for long large writes; not so good for transaction type environment with many short writes." You can't use the technique for enterprise drives because of the performance impact. The Seagate 3TB FreeAgent possibly uses the technique.
"This will extend areal densities towards 1Tbit/in2, maybe more." But then you are forced to go to a next-generation technology: "BPM is four to five years out, we think, because the lithographics are far from where they are needed to be today." The current supplier of such equipment to the HDD suppliers, ASML of the Netherlands, "have no machines capable of doing this year". Barber also mentioned that "Samsung has basically ordered every machine they can make this year".
"A couple of startups have sold one or two pieces of BPM equipment into labs for evaluation. It's a long way from being put into high-volume production." He is referring to Molecular Imprints and Obducat, and has doubts about their ability to survive for the next four to five years without orders from the HDD suppliers. "They have to spend a fortune on research and development. Customers won't buy in volume for at least five years." Obducat has recently undergone a reverse stock split - often a sign of financial tension.
The overall take-home from this is that we will see future capacity increases coming from an increased platter count in hard drives, both 3.5-inch and 2.5-inch, with a slowing down in areal density growth, and less frequent drive refreshes. External drives will diverge from desktop and enterprise drives because shingle writing will be used in the external drives but not the internal ones. We are going to see the life of PMR technology extended out to 2014 or 2015, and maybe HAMR will come in because it will be a less expensive next-generation technology to use than BPM. ®