Amazon may be using disk drives with hot-swappable components Amazon's won a patent for “Hard disk drive assembly with field-separable mechanical module and drive control”. In Amazon's world disk drives have two distinct systems. There's a “drive mechanical module” comprising a drive's arms, heads, spindle, motor and platters. And there's a “control module”, basically the drive's …
Jaz: disk in cartridge, motor, heads and controller in drive. REV: disk and motor in cartridge, heads and controller in drive. Pre-USB floppies: No intelligence in drive at all, just transistors to amplify control signals enough to drive stepper motors. The disk controller chip (on an add-in card, on the mother board or integrated into the chipset) had just enough brains to match the sector/track fields in the sector header. Everything else was handled by the CPU. Early hard disks connected in a similar way to floppies, with the controller (which could handle multiple drives) separate from the mechanism.
Drives gained intelligence because intelligence became cheaper, the market size increased giving economies of scale, and the reduced latency (between mechanism and controller) made a big difference to performance. For a while, it was often practical to swap the controller cards on a pair of similar drives. This became more difficult because the modern interconnect is more difficult to swap and drives are cheaper than the time required to fix them.
Where to place the divisions between media, mechanics, electronics and intelligence have always been selected by the market forces, not because only a single person on the entire planet had enough brains to spot that a component had become cheap enough to bundle with the medium. The market is going through another big shift: consumers are switching from mechanical to solid state and data centres are becoming the major/only customer that needs the huge capacity of modern drives.
It is about the right time to change which bit goes where again. Patent offices all over the world are calling this an invention worthy of protection - without any details of a standard interconnect or how to implement the standard. Patents were supposed to be about rewarding publishing research so that everyone else did not have to repeat the research. In real life, patents are a way to punish people who do research because a patent holder who contributed nothing can demand royalties from the people who do the actual work.
Not putting all the electronics into the same box with the mechanics? Who could have thought of it?
Sigh.
Wonder if the original floppy drives would be prior art? They had very few smarts, the microcomputer had to control them in great detail, which was one reason why in the old times, floppies from different microcomputers usually had different incompatible physical formats.
I think you will find that the ST506 interface (ie what preceded ATA) was exactly this. The drive brought out all signals (stepper motor and data in/out) at TTL levels, that the hardware did the rest. of course, this was when H/Ds were 5MB or 10MB.
I should know: I designed disk controllers of this type for Apple, BBC Micro, and several minicomputers with discrete components.
Pertec (and other) tape drive interfaces were the same (only different :-) for 1/2" tape - yes, I did several of those too (ICL, CDC, etc).
ATA was just the interface to a Western Digital chip that drove ST506 brought out on a 40 pin connector.
Yeah, but ... Ever try to hop swap an ST-506?
As I remember it, you could not even move it to a different PC while cold, and expect to read the old data. (Tried it with my first two PCs). Some data was visible, but most was unreadable. The controllers evidenly controlled the drive in slightly different ways.
I have a much better idea: let's make every component removable and replaceable, including the heads, platters, bits of electronics, and the housing. I'd apply for a patent, but sadly my parent's generation has robbed me of this idea: DEC RP04 pack.
I have to agree with FK above that patent system today would have been funny if it weren't doing quite so much damage to the companies and people who actually make and invent things.
I read about something similar last week when researching data recovery solutions.
If a hard drive is unaccessable opening and replacing the controller MB (with another of the same model) is one option. Since you're not messing with the platters a clean room isn't required.
I had a hard drive that sounded fine, but was completely non accessible.
You'd still need to recover the data (using software) but that's the easy part.
Warning - disassembly is a b1tch
Removing the motherboard from a hard disk is an absolute doddle these days as well. They're already assembled in the manner Amazon are suggesting. It's usually about 4 hex screws, and the motherboard falls off. It's usually electronically connected by a ZIF style panel, no disconnection of ribbon cables anymore.
As other people have noted, it's a fairly straightforward approach to data recovery. If you're competent enough to remove a hard disk from a computer, you're competent enough to do the transplant.
I recall doing similar with a bios transplant on a PC motherboard. It complains a lot, but can often get you past a failed POST. I doubt it works in today's UEFI world though :-(
Back in the early/mid 2000s had a drive that I hadn't backed up too recently that suffered controller failure, so I went on eBay and bought a used drive of the exact same model and swapped the controller board off the eBay drive into mine. Worked like a charm.
I've heard that's no longer possible with modern drives, that the way drives are formatted with such high densities these days there's some drive-specific formatting data saved in the controller.
Maybe that's not true, or only true of certain modern drives, or Amazon has some way around it. But honestly I don't see the point in trying to "fix" an individual hard drive. Amazon will have plenty of redundancy, I don't see how it could be worth sending a human to swap a controller board to save a $100 drive! Wasn't Google abandoning servers when anything failed in them, and abandoning entire racks when too many servers in the rack failed?
I did this once too - in my case, it was the other way round, and to a disk drive instead of a hard drive.
The mechanical side of the drive had failed, and as it was from an XBox 360, a replacement drive wouldn't work as it was not registered. The solution was to take the electronics from the broken drive and pair them with the working drive.
Jobsagoodun.
Tried that about 6 years ago - when a disk motor ran but the PC couldn't detect the disk. The same model disks in my spares had controller boards with various version numbers - but managed to find one that matched. Had high hopes - but it didn't solve the problem.
I wonder how the various possible failures stack up in order of probability?
Okay, let's fix this straight away; if anyone has NOT ever repaired or recovered a drive and the data by swapping a known good drive controller from another HD of the same flavor please leave the site immediately until you do so. Thank you.
I'm glad you other folks did as well. I just looked at the dead drive, the admin for the dev group explained there were no backups, and I just asked for a fresh drive of the same type. It's not rocket science. It's barely computer science. Still, makes a great story. Until it doesn't. Like my personal 2TB Seagate desktop drive with the goofy power/USB adapter on the bottom. I got another of the same drive, but no chance, English bed-wetting type. No data for you. Still, only some iTunes files and a couple of recoverable other items were lost. Nothing important, like porn.
This post has been deleted by its author
A quite naive idea, because all current drives require drive-specific calibration data, especially since analogue "speaker coil" head positioning was introduced!
I doubt that they will save any money and it will probably cost more because an integrated analogue/digital SoC is probably a lot cheaper that trying to separate off the digital only part with some kind of fast enough and reliable linking interface to the digital <-> analogue layer. They maybe able to separate off the higher digital layers like caching and queuing, but trying to do the lower digital layers too will probably just make it slower and less reliable.
HP still commands a high price for their modest capacity ( <1 TB) SAS drives compared to the high-capacity consumer SATA drives, so this MIGHT be useful. But like Yarr said, what about the SMART stuff and the bad-sector map?
Then too, it depends on how the component pricing breaks out - if a $300 SAS drive has a mechanical portion that costs $200 and the electronics board costs $100, then maybe at Amazon-scale it makes sense to put a new board on an existing mechanical drive and save the other $200. Especially if you're doing that thousands or tens-of-thousands of times a year.
But, eh, in MY datacenter if there's any whiff of "badness" to a drive, it's gonna get swapped completely with a new drive.
Then all our drives would be like Jaz drives today, instead of the typical form; interface/controller/spinny_bits all as one unit. I had some Jaz drives, and these:
https://en.wikipedia.org/wiki/SyQuest_Technology
And guess what? All the drives were dirt-cheap crap, and expensive, so you could not get enough of them to keep the carts around long enough to be suitable for long term storage. Tapes did that better, and CD-ROM made the 44MB carts look as useless as floppies. Now I know better. I just print out all my data to punch tape, photoscan those to image files, then save those to old AOL promotional floppy disks, and Bob's your uncle. Safe as the crown jewels, me old son.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
