@"an awesome 5TB capacity"
So how long before we get a 5TB drive. :)
It's done it; Seagate has achieved a 625Gbit/in2 areal density enabling it to produce 3TB Barracuda drives on just three platters instead of the current five. It says it is on track for the channel to receive product around the middle of the year with 1TB, 1.5TB, 2TB and 3TB capacity points. These are 3.5-inch form factor …
"El Reg would like to point out that, if Seagate stayed with its 5-platter design, it could produce a single drive with an awesome 5TB capacity."
We won't see such a drive until they push out a 4-platter 4TB drive in a few months (once other drive manufacturers can [and do] put out 4TB drives). Then they'll likely trickle a 4.5TB and perhaps the 5TB drive. Right now, they'll make enough of a premium pitching the same 3TB capacity but at slightly less cost than competitors, and make a tidy profit.
"and, its statisticians having given up, "virtually countless hours of music"."
2.88MB per 3min song (128Kb/s bitrate), and assuming 5TB and not 5TiB:
5000000 / 2.88 = 1736111 (1.7 million) songs, or * 3 (minutes) = 5208333.33~ minutes, 86805.55 hours, ~3616.9 days, or ~9.93 years worth of audio. Hence why their stat people just gave up. It's not worth listing the number of MP3s anymore.
How much space is this in Libraries of Congress?
... 128kbps mp3's are sooooooo 2001...
These days for me it's either 256kbps VBR or better if I'm stuck with mp3, but I'd prefer lossless (flac or ape).
Anyway, a more interesting metric (from my point of view) would be : how many 10mpix canon raw files can be stored on one of those?
(answer: probably more than my beloved 400D could shoot during its lifetime)
I would thing that in this case, the mass of the head assembly would come into play during seeking.
With a 5 platter head assembly, there is more mass to move so in order to have accurate positioning of that mass you need to move the assembly slower because there is more mass to start and stop. The faster it moves the less accurate it is.
With a 3 platter head assembly there is not as much mass to move so you can seek faster while still holding on to your positional accuracy.
That depends on your usage pattern. If you are reading huge sequential files stored without significant fragmentation, then the 3-platter drive may have to do a one-cylinder seek 5/3 times as often as the five-platter one. A one-cylinder (minimum) seek is much faster than a random seek. It could also be optimised, if the start of each cylinder is rotated with respect to the previous one by an amount of rotation with a latency just slightly bigger than the one-cylinder seek time.
If your typical file is smaller than one three-platter cylinder, both drives will be pretty much identical.
While thinking about performance, isn't this technology crying out for use in a small-platter drive spinning at 15000 rpm or faster? With twice the competition's areal density, they could make an ultra-fast drive twice as capacious. The last gasp of the sub-Terabyte hard disk, before SSDs take over?
Seek time should be the same - it is defined by the rotational speed of the drive, the diameter of the platter, and the speed of the actuator motor (plus a few other minor things). Since all these factors will be the same, the seek time will be the same (give or take a tenth of a millisecond).
What may change is the peak/average read/write speeds. That depends on the areal density and the number of platters/heads.
Question to those who know more about this than me...
Why do hard drives only have one head assembly (i.e. one head per surface)?
If it was possible to fit two or more independently-movable sets of heads within a standard-sized case, surely this would make things approaching twice (or three times, etc.) as fast (interface permitting)?
Obviously there are physical limitations, both fitting them all in the case and also ensuring that they wouldn't get in the way of each other. But it seems strange that I've never heard of anyone considering this before.
Cost is probably the answer. Plus, for real performance gains, everyone is now looking at SSDs; the future of HDDs is definitely bulk storage, with an increasing trend towards SSDs acting as a fast cache to provide the performance in both enterprise and end-user applications (e.g. Intel's forthcoming Z68 motherboard chipset includes 'SSD cache' controllers).
More years ago than I care to remember, we fitted fixed head 10" disks to early computers with a massive (!) 640k storage. There were multiple static heads and a spinning disk. It was blisteringly fast (for the time) but fiendishly expensive to produce and hence only used by those with buckets of dosh and a need for fast access.
It was used mainly for bringing things from main storage for sorting operations and other things that required fast access.
But it is insanely expensive. Platters are cheap, heads are not. Having multiple heads in a system will only reduce the the life time of the platter, and cause interference with the other heads. There are also issue involved when you have multiple head writing to the same sector, requiring a lot more engineering.
What you end up with is just two independent HDDs that happen to share the same platter and require some sort of master controller to ensure the heads don't overwrite the others' data and won;t run into each other. And with less reliability than a single disk. In the long term, it just makes sense to go with a RAID 1 with a decent controller.
the headstacks would need to be on opposite corners of the drive assembly. in the current drive case this is not doable. you would need a physically larger case.
the headstack also counts for more than 50% against the cost of the drive. the platters and control electronics are peanuts.
The reliability would take a tremendous hit as well. the more heads the lower the mtbf. on modern drives more than 90% of catastrophic failures are cause by blown heads due to static discharge...
Conner Peripherals (acquired by Seagate in 1996) tried this; it didn't work as well as expected.
At one time they did build (for a very short time) a hard disk drive with two HDAs, the Conner "Chinook" series. It had a couple of problems, namely:
1. It had an obsolete form-factor: Internally it used platters sized for 3.5-inch drives, but to hold the extra HDA it had exterior dimensions equal to those of a 5.25-inch hard drive.
2. The vibrations from one HDA could make cylinder seeking difficult for the other HDA: This wasn't a problem when transferring large, unfragmented files, because the HDAs didn't need to bounce around much looking for data. But if the drive was performing a lot of random seeks for small files (or fragmented large ones), seek times could actually increase and degrade performance.
Seagate and Verbatim for the win.
(That's what FTW means isn't it? eh?)
No, actually I like replacing hard drives and thrive on watching them die out of the box and as for floppy drives, the stake through the heart that was the USB thingee was not enough to satisfy me.
I am going to burn a floppy this weekend in remembrance of lost data.
Then I will bury it with a garlic clove.
Guess the brand name.
I've heard it said that one head actuator seeking inside an HDA would create so much vibration as to prevent any other heads from being stable enough to read or write. If you couldn't do seek and read/write concurrently, there wouldn't be very much point in multiple heads.
Reportedly, this can even be a problem between disk drives in a badly-designed multi-drive cage: if all drives are engaged in seek-intensive activity, the vibration mechanically coupled between them can degrade each other's performance. Or so the manufacturers of heavy-guage server towers with rubber drive suspension bushes claim. OTOH it might work on the same basis that wearing a paper bag over your head in the UK keeps the elephants away.
No comment from us about the (lack of) reliability of the Constellation ES range, having said that, in reply to an earlier post, the worst performing (reliability wise), drives for us over the past 15 years or so have been .......WD.
Forget the increased capacity - I'd be happy with the same sized platter if I got more reliability from what is admittedly now a mass market product.........with a price to match.
Annoyingly, the original 7200.11s worked great in the SCSI->SATA enclosure I bought them for. Then they "fixed" the firmware and screwed them up. I got something like 40MB/sec on RAID0. Had to finally bite the bullet, and by the Constellations as listed on the HCL, and suddenly went up to 180MB/sec.
I liked the rushed 7200.11s. Though in honesty, they've almost all failed where I still have some 7200.10s performing sterling service.
Fail - because that's what most of them have done by now...
I recently spent some time to buy a new internal drive.
The thing I couldn't help but notice were the number of RMA's associated with drives over 500gig.
Then I started to about how many of these drives are showing up in "wonderful" external backup systems.
The tech part is always interesting, but it's not everything.
Have you heard about ECC RAM? RAM sticks will flip bits randomly because of current spikes, cosmic radiation, etc.
The same thing happens to hard disks, they flip bits randomly. CERN did a study on this and found lots of flipped bits. This is called silent corruption. A modern disk devotes 25% of its surface to error correcting codes, and during usage, lots of errors occur all the time. But are corrected on the fly. But some errors are not corrected. If you look at the spec sheet of a SAS enterprise disk, it says "1 irrecoverable error on every 10^16 bits".
This Silent Corruption is a big problem, and hardware raid does not help. You need checksums to track changes in the data. HW raid does not compute checksums, they compute parity. ext3, JFS, XFS, ReiserFS, NTFS - are all vulnerable to this because they dont compute checksums.
Here are some research on this Silent Corruption, and how to solve this problem:
http://en.wikipedia.org/wiki/ZFS#Data_Integrity