On Tue, 17 May 2005, Jed Rothwell wrote:
[snip]
I think many people are beginning to reach the limits of disk storage requirements. I have not filled out more than a third of a disk in several years. If I were a manufacturer trying to compete in the hard disk business, rather than increasing storage capacity I would try to differentiate my product by giving it unique characteristics. I would revive a long obsolete hard disk design I saw on a Data General MV 8000 back in the 1980s. It was a 20 MB disk with a row of multiple immobile heads on each platter. I mean the heads did not move: track to track seek time was zero.
Almost every mainframe manufacturer had fixed head disks, starting (I think) in the late sixties/early seventies. Most PC's still can't match the I/O capacity of a fast twenty-year-old mainframe. For that, PC manufacturers would have to add more capable backplanes/buses and external I/O processors, as well as fixed head disks etc etc.
To bring that idea up to date, you would develop an array of microscopic read/write heads that covered the entire surface of the disk, with one head per track. From the Maxtor.com site, I gather that a modern disk has 16,000 tracks. If you could make an array of 1,600 heads (spread in a staggered array across the entire top surface), that would give you one-tenth the capacity of a modern disk, or perhaps 25 GB, but way faster. It could be used for Windows swap space or something like that. Average seek time for a modern disk is ~9 ms. It would be zero instead, leaving only rotational latency, around 4.2 ms. You could activate all read-write heads at the same time and format the entire disk (or erase it) in the time it takes to rotate once (8.4 ms, or 60 seconds/7,200 rpm). With a new interface you could even write data to hundreds or maybe thousands of read-write head directly from RAM with DMA. That's what the DG computer did, as I recall.
- Jed
I would love to see this development in disk technology!
/Mathias
