The Disk Drive at 50: Still Spinning
Published: September 18, 2006
by Timothy Prickett Morgan
This month marks the 50th anniversary of the announcement of an innovative storage technology that IBM created and launched for its early computers--a device that predates its vaunted System/360 mainframe by eight and a half years. Back in September 1956, it was just called the IBM 305 RAMAC, and it wasn't a disk drive so much as computing system attached to a giant disk drive. That device, the great-great-granddaddy of the spinning disk in your computer and those in use at your company, is arguably one of the most persistent pieces of technology ever created.
Early phonographs created by Thomas Edison stored data on a rotating drum. Edison made scratches in the drum to record sound, and then rotated the drum to recreate the vibrations that in turn caused a horn to vibrate with the sound that was recorded--with a much lower fidelity, to be sure. Eventually, records were turned into platters and mass-produced, creating a music-listening industry where once there had been a sheet music business and a world of half-rate musicians--our relatives and friends.
By the late 1950s, information had been encoded on punch cards for decades and had more recently be put on magnetic tape. The genius of the RAMAC drive was its speed. But storing data on the top and bottom of metal disks using magnetic material is not an IBM invention--in fact, it was created in 1952 by an engineer named Jacob Rabinow, who worked at the U.S. Army's Ballistic Research Laboratory on its Aberdeen, Maryland, proving grounds. IBM's researchers initially chose a drum for RAMAC instead of a disk platter (which was easier to manufacture) because it was stiffer and therefore it was easier to maintain a consistent space between the drum and the head that read data off the drum or magnetically encoded it on the drum. But the RAMAC 305 actually consisted of 50 metal platters, spinning at 1,200 RPM and a recording density of about 100 bits per inch. RAMAC was capable of storing between 5 million and 20 million "words" of data. Yes, this predates bits and bytes as we know them.
RAMAC, which was invented in IBM's San Jose, California, disk labs, was noteworthy not because of its data density, but because it was the first movable-head data-storage device. (RAMAC is short for Random Access Method of Accounting and Control.) By making the read-and-write heads movable, the heads could be programmed to find data by accounting for disk spin and the physical position of sectors of data in relation to the arms. This significantly increased data access. You didn't have to--as with tape--start at the beginning and read serially to the end to find your data. You just jumped right to the spot. If there is a reason why computers are not based on tape--and no, you Atari and Commodore whiz kids from the late 1970s and early 1980s, you did not have a real computer by my standards because you did not have random access storage--speed is it.
Disk drives of the early 1960s--take the IBM 1311, for example--had a capacity of 3 MB stored on 50 tracks, with an average access time of 150 milliseconds and a spin rate of 1,500 RPM. The disks had a recording density of 1,100 bits per inch, and could move data out of the drive at 77 kilobytes per second.
The 3340 "Winchester" disk drive was launched in March 1973 and the disks we have today are, despite all the innovation in the disk market, still related to the Winchester. The Winchesters, which were created for the System/370 kickers to the System/360s, were originally designed to have two 30 MB disk spindles--a 30-30, so to speak, and hence the nickname based on a popular brand of double-barreled shotgun. The Winchester had two heads instead of one--always a better thing, so I hear--with an average access time of 25 milliseconds and it also featured fixed heads for frequently used data that could be accessed in 5 milliseconds. The Winchester spindles spun at 4,000 RPM, and the unit could push data at 885 kilobytes per second. In many ways, they are very much like the disk in your PC, except that they weigh a ton (well, not really), and the amount of kinetic energy in a Winchester drive made it a truly dangerous device. They used to "walk" across the floors of data centers, like a lop-sided washing machine, when they got out of alignment. In fact, they were the size of a washing machine, too. The Winchesters were also sealed drives, unlike prior generations, which allowed disk spindles to be removed from the drive and replaced with other spindles, much as tape drives do to this day. Removable spindles could be destroyed by dust and fingerprints, so IBM sealed them.
Fast forward to 2006. Disks used in conjunction with servers have shrunk from 24-inch platters in the RAMAC, to 14-inch platters in the 1960s, to 10-inch and 8-inch platters in the 1970s, to 5.25-inch platters in the 1980s, to 3.5-inch platters in the 1990s, to 2.5-inch platters in the 2000s--all while radically increasing the density of data stored on the platters. Mainframe disk arrays are still using 3.5-inch SCSI disk drives, but Serial Attached SCSI (SAS) drives in small 2.5-inch form factors are now common in rack-mounted servers and various disk arrays that are typically attached to them.
Take, for example, the Hitachi Ultrastar 15K147, which would have been an IBM disk drive had Big Blue not sold off its disk drive business nearly four years ago. The 15K147 is a 2.5-inch Ultra320 or SAS drive that spins at 15K RPM. It comes in 36 GB, 73 GB, and 147 GB flavors (that's with 2, 3, or 5 platters). This disk has an average seek time of 3.3 milliseconds (how long it takes to find a track with the disk arm) and an average latency of 2 milliseconds (how long, on average, it takes the disk platter to spin under the head with the right data). It has a sustained data transfer rate of 93.3 megabytes per second. With the Ultra320 SCSI interface, the 15K147 costs a mere $450 or so on the street.
The next big advance in disks is what is called perpendicular recording, which means taking the bits of data, which are really tiny magnets, and standing them up and down on a platter instead of end-to-end flat on the platter. By going perpendicular, disk drives can do what cities have done--go vertical, and pack a lot more material into a given space. In a city, the material is people, while in a disk, it is data. Hitachi has already demonstrated a perpendicular recording method for disks that can deliver a recording density of 230 gigabits per square inch, which is about double the best density available with the current longitudinal techniques. That means disks with multiple terabytes of capacity are on the horizon.
But, maybe instead of packing more data into a space, the industry will instead focus on making disk drives that have about the same capacity and consume a lot less energy and make a lot less noise. For many parts of the IT industry, this would be far more useful than convincing companies to store more and more data electronically. Toshiba has already created a disk drive that has a platter size that is only 0.85 inches, which has a 2 GB or 4 GB capacity. The disk platter is about the same size as the face found on dice, to get the visual image. The MK4001MTD MicroDrive has a single disk with two heads; it can move data at 12.5 megabytes per second, and has an average seek time of 16 milliseconds, an average latency of 8.33 milliseconds, and a rotational speed of 3,600 RPM. (No, that is not fast.) In many ways, this disk has the characteristics of a 3.5-inch disk from a decade ago--except that it is ridiculously small by comparison. And the MK4001MTD only consumes 0.12 watts when idle and only 0.6 watts when reading and writing data.
Imagine a RAID array made of these. Imagine how little it would cost to buy and use. Imagine what they will think of next in the disk business, as they have for five decades.
Frustrated by Losses and Glitches, IBM to Exit Hard Drive Biz