Some IBM research showed that disk response time is largely down to moving the head/disk assembly (HDA). Once the HDA is positioned at a track then the disk's 10 or 15,000rpm gets the data read, or written, pretty quickly. The thing is there are two things to move: the HDA which is a start:stop device; and the disk which spins constantly. The IBM researchers reasoned that the way to get a faster disk response is to move the HDA less by making the data more sequential in nature; concentrated in one place on the disk rather than being placed in several different places.
That's okay with a large file but pretty difficult with database records. And anyway, if we are doing some blue sky thinking, it's not a good answer. Why not have more than one HDA?
If there were more than one - two say - then we would double the data rate off the disk and shorten the access time to the data. Certainly so, but the mechanical reliability of the device would decrease. HDAs break. The power need would go up which means the heat generated would increase and, in turn, disk packing density would go down.
So, let's say that multiple read/write capability is needed but having multiple HDAs is not the right way to achieve it. An HDA is basically, we could say, a signal detector and signal writer. Are there examples around of multiple signal detectors in a small space?
There is in the signal reading domain. Digital cameras have CCDs - charge-coupled arrays - which consist of large numbers of light detectors. A 4 megapixel camera has four million of them in an array which is created pretty much like a DRAM chip or a microprocessor chip or, indeed, an LCD screen. Could we have something like that positioned over a storage surface?
Suppose we had a storage surface like a hard drive's disk platter. Over it we positioned some panel of detectors that responded to changes in magnetic fields. What would happen?
It's pretty obvious when you think about it. A digital camera's CCD responds to light changes when the shutter is open. It detects signal strength change rather than just the signal strength. So in our magnetised recording surface example we'd either need to switch the magnetic fields on or off or move something; either the detector panel or the recording surface.
Movement is mechanical, which means more power and also means positioning. But switching fields on and off would be complicated. The movement wouldn't have to be much, just enough to move a data element out of one detector's field of view and into another.
How could the detectors be made? Current disk drive read/write heads are made by fabricating wafers and then cutting the wafers up into individual pieces, one per HDA. We could - this is back of the envelope theorising - cut the wafer to produce an array of read/write units suitable for our purposes. So now we have signal detectors (readers) and could also have signal creators (writers).
We get an array of these things, position them over a magnetised surface and then move one or the other. The detector array needs wires to the outside world and it needs some form of addressing, row and column wires should do it. Any movement of this could stress and break the connecting wires. Better to move the recording surface which doesn't need connecting wires. Like a disk platter it isn't electrically connected to anything.
How to move it?
It only needs moving slightly; a back and forth motion would do it. In fact it would be more akin to vibration than movement, as long as the vibration was sufficient to move data elements into and out of focus of the detectors. Magnetic force could be used, working against the edge of the thing. A small mechanical cam could be used, but that could be unreliable over time and would also cause wear. Inkjet printers move ink particles using piezo-electric force to cause ink droplets to move. Let's just say it's a solvable problem.
Why bother though?
The answer to this is response speed. The entire area of the disk surface under the detector panel becomes readable, or writable, in one operation, and that operation requires ansolutely no mechanical startup time to move a disk. Instead, the recording media surface vibrates many times a second, hundreds even. Then we might entertain the idea of effective bandwidths onto and off such a device of a hundred gigabytes a second or more with response time factors of magnitude faster than spinning disk.
Let's suppose that Intel/Seagate-style manufacturing smarts were used and we might even suppose that such a device could approach the response time of RAM. That sounds fantastic. It would be faster or as fast as flash memory without that medium's limited life due to its inherently restricted limit of write cycles (something like 5 million). It might also be a darn site cheaper.
If recorded data access speeds increased so much then the vast aeons of time CPUs wait for disk I/O, aeons at the GHz speeds of microprocessors compared to the milliseconds needed for dosk I/O, would vanish. Our computers would become massively more productive. Perhaps the power needs of such 'hard drives' would become much less, so they could be packed closer together and data storage density would rise as well.
Woudn't it be good if we could do this? Disks have remained stubbornly slow whilsts CPU speeds and PC bus speeds and network speeds have, in relative terms, rocketed. Computing progress is being held back by this, when you think about it, primitive need to move and position disk drive HDAs. Digital cameras don't have to move film behind the lens any more. Wouldn't it be wonderful if hard drives became much more digital and the moving HDA went the way of the punch card and paper tape?
Is this feasible or am I dreaming?