
Look at the date
The post is required, and must contain letters.
In a final attempt to break out of the I/O density gap crippling hard disk drive I/O, Seagate is introducing drives with three heads per platter that will dramatically cut latency and seek times. Chairman and CEO Stephen Luczo said: "Latency and seek time delays are the open door through which flash drives are pouring in to …
Have been waiting for someone to do this for a couple of decades now, I suppose with really smart firmware it has the potential to more than treble random i/o since you can pick the head that is closest to a given request. However, with more moving parts - reliability may be less than current single head designs.
With the much larger HD capacities, this should keep hard drives in the game with flash for a lot longer.
If* this is the april fools (the press statements do sound a bit wanky) then why has it not been done before? I assume it's something to do with head angles/geometry. Also there's only room for 2 heads unless the platters are 2.5" size platters in a 3.5 enclosure.
As for reliability, the remaining 2 heads could pick up the slack and the firmware can send an SOS to the admin for a swapout.
* "Triple X" ??? Hmmm...
They stated there's 3 read/write heads on a single arm (well, more inferred, but still obvious). The 3 heads will still be bound by data that's near them. If they're a fairly large-spread array (about 1/3 the radius of the disk wide), it would potentially cut the seek time by a good quarter (most of the latency is likely due to initial movement/stopping at that point, rather than travel time). However, I would have proposed a dual-head solution (granted 3 heads might be just as simple/complex) combined with a dual arm setup. Then you'd have 2 independent arms with two heads per arm. You could read/write up to 4 tracks at once, but not be constrained by the physical location of the data, whereas a single arm with 3 heads would only be useful if other to-be-read-simultaneously data falls under one of the other heads. Two arms will act likely as quick as a RAID1 setup with an intelligent controller, where the two read requests are handled simultaneously, rather than speeding up one read, then both moving to the other read.
Granted, two arms = worse MTBF since there's more moving parts, but it's likely the best way to get better random I/O. Parallelism is why SSDs get such great data rates (yes, and access/"seek" time, but that's practically impossible to eliminate on spindle drives). HDDs will have to think in terms of parallelism before they can start to approach SSDs in performance. 3 heads is a good start, but it will only help in a limited number of situations. The largest benefit will be the (modestly) reduced seek times and concurrent read/write ops (assuming there's a blank track below one of the 3 heads).
Western digital are already working on the next-gen Globe storage devices. Essentially a series of very thin, hollow spheres, one inside the other, not unlike the Russian Babushka dolls, WD believes this contraption to be the next big thing in local storage.
Sharing a common center, with 3 axes of rotation, and no less than 12 heads per sphere (6 inside, 6 outside) this futuristic device will hold 4 times the volume of current storage devices, while cutting the seek time and multiplying throughput by as much as 1200%.
This post has been deleted by its author
Ah, but if they had gone for two heads it could also have been reported that the replacement for the Barracuda series would be the new Beeblebrox series (also providing the opportunity for endless fun making up a suitable puff quote over the licensing deal from the BBC).
A missed opportunity I feel.
Yep it is likely an April Fools gag but obviously one that has a lot of people interested in it being true. Even though the cost would increase to support and build such a design would go up it would certainly reduce the lead that SSDs have over HDDs in performance and help Seagate keep up their market share and volume.
I wrote about this back in the mid 90's when doing CompSci at uni - I wrote a essay which included some algorithms for a cdrom drive using multiple heads, I've always thought it was an obvious idea.
The trouble is the failure rate would be significantly higher than a standard drive and heat generated would also be increased too and cost is a big factor too.
There was an Israeli company that sent out a press release about a multiple head cdrom drive shortly after I wrote the essay but I don't think it was ever released.
Now thanks to the wonders of Google I've just found out that Conner actually made a scsi hdd called the Chinook at the start of the 90's that had multiple heads.
Seagate bought Conner in the mid 90's so this idea has been sitting on the shelf for a long time, I want to see the performance of these new drives purely to see if the idea really is as cool as what I though it was all those years ago!
This post has been deleted by its author
This post has been deleted by its author
even if it is an april fools ( and it is... )
here's the flagrant mistakes in the article that make this unbelievable
1) it is 3 heads per recording surface which would mean 6 per platter. ( platter has top and bottom. so 2 recording surfaces per platter. Mistake in title...
2) if it were possible to put 3 heads on an arm and switch electronically between them ( the switching is not a problem. Drives have been doing this for years. there is onyl one preamp with up to 8 heads attached. Each head consist of a read TMR or GMR element (Magnetoresistor), a write head ( inductive loop ) and a heating element to control flying height ( Bernoulli's law governs the gap between head and platter for a given speed. other factors that control flying height are temperature and pressure. Pressure doesn't change. Temperature does. Writing heats up the head , so when switching to reading that temperature must be maintained. Look up 'pole tip extrusion' (don't snigger! that's what it is called). right where was i ... ah yes 3 heads per arm : since the disc is round this would mean that the angular velocity of data flying under a head close to the hub is larger than that of data passing under a head on the rim. And that is not true. Can't do that.
So, to make this a believable April fools story : get the error's out of it. You can leave the pole tip extended.
Rubbish. Drive channels simply use the same Viterbi detectors/iterative architecture that the cell phone business has used for years. It's just a question of cost, power, and data rate. Servo is just glorified control theory. Make the components stiff, pack in as many turns in the VCM as possible, stick microactuators on everything, put in windstrippers to control windage and notch all of the resonant modes. Design of the mechanics themselves i.e. torsion, sway, etc. as well as accounting for the trace heating is far more complicated as is the contact detect using the heater. Opti and its iterations of write current, overshoot, and duration at different BPI's and TPI's is even more complicated than that.
BTW it's called pole trip protrusion, not extrusion. Squeezing a pole tip out would not be good. I also didn't understand your bit about angular velocity. The head cares about linear velocity, not angular, and it cares about its skew relative to that velocity. The fact that each head on the arm would have its own heater and would perform its own contact detect and fly control make it perfectly possible. However, you would never put multiple heads on the same arm as they wouldn't be able to access the entire surface. A true implementation would have independent E-blocks which could sweep the entire surface to give you full random access and full performance. But none of this would ever happen because it would drive up power and cost, and it would break the drive form factor.
Conner Peripherals (later Seagate) Produced the Chinook, a drive with two heads per surface.. the key to this was not two heads per actuator but two actuators, capable of reading or writing to two places on the same surface simultaneously.
With more modern control algorithms this technology could make a huge difference today, but it wont because nobodies that bothered, and nobody wants to pay for it.
Sequential write and random read no problem.. revolution latency ~50% to any head.
A two headed drive (two actuators) would be the equivalent of a double speed drive. twice the write twice the read and half the latency.
Conner used to make dual actuator disks. The model was called "Chinook".
http://en.wikipedia.org/wiki/Conner_Peripherals#Performance_issues_and_the_.22Chinook.22_dual-actuator_drive
Unfortunately, tripling the actuator counts would also almost triple the cost of a disk and destroy a non-trivial chunk cost advantage that mechanical disks have over SSDs.
Conner only made Chinooks because their technology wasn't good enough to keep up with spindle speed increases.
Several companies have tried multi-actuator drives. The cost usually was 3 to 4x the standard drive!
So, if you want an HDD that gives 300 IOPS for the same price as an SSD drive, send Mr. Luczo at Seagateban email requesting a free sample. He'll remember this April 1!
Good reporting, Chris!!!
In ancient times, late 1980's, we had what was named "compact disk" of about 40 MB made of about 6 10.5" platters. Interface was SMB. Believe the drive was Fujitsu. Had one moving head assembly the the usual amount of heads, one per platter side. But a region of 16 tracks on one platter side had a fixed head assembly which never moved. The drive manual suggested one put the most updated data in this region, filesystem metadata or swap. Said the drive will transparently use the stationary heads without seeking the movable heads.
Then a few years later Apple introduced the Lisa. Before the 3.5" hardshell floppy used on the Mac the Lisa used a 5-1/4 with two disk openings rather than just one. I don't remember if each head assembly on this floppy drive was double-sided, but do remember that each moved independently of the other. IIRC this was called a Twiggy Drive.
Should have gone like this;
"...scrapped the concept of read/write head and all of its mechanical failings with a rigid sensor bar spanning the full radius of the disc,
The sensor bar supports a lead just above the disc surface, through which a series of high frequency pulses are fed that can be switched between two discreet power levels.
Neither power level is high enough to affect or sense the data on the disc on their own, until Phase modulated to create a constructive peak at any precise point along the lead's length, coinciding with the desired read/write operation.
This effect is caused by the phase modulated pulse interacting with the previous pulse as it's reflected back along the lead's length, causing constructive interference.
The lower power peak would be used to read data from the disc , while the higher level peak would write to the disc.
In effect the the phase generated peak acts as virtual read/write head that can be instantly moved along the sensor bar's length."
There. that's just plausible enough to be believed.*
*Actually, if it turns out that this actually COULD work, please bookmark this post to show prior art when some asswipe tries to patent it! Hell, if it is feasible, the only stipulation I put on its use is that it must be called "DaFt Drive"!