rcd@ico.isc.com (Dick Dunn) (09/28/90)
As to why the old folk keep babbling on about the bizarre old disks--hey, they were interesting bits of engineering and design, and they sure wrung a lot of performance from relatively unsophisticated technology! They also show some impressive numbers here and there (like the spindle drive motor on the 6603, or the number of heads on the 6638). But fear not, you young 'uns; we don't have very good memories, so pretty soon we'll forget what we were talking about and the thread will die out. In the meantime, here's more detail than you wanted about the 6603 (the really strange one) plus enough about the 6638 to distinguish it, with a few anecdotes sprinkled in to keep it from getting too dry. I've included as many numbers as I had handy; you can use them to calculate out whatever you want to know about these dinosaurs. If you're looking for still more numbers or details, send me a note--I worked with the '03 and I'm in touch with someone who did a lot with the '38. 6603: ===== This is the bizarre Bryant disk with the huge platters, horizontal spindle, and hydraulic-fluid-actuated positioner. The platters were indeed 39" in diameter, about 1/4" thick, and a careful measurement of weight and volume says that they are pretty close to pure magnesium. One platter plus the steel collar used to hold it to the spindle weigh out at about 28 lb. There were 14 platters on a spindle, so adding in the spindle itself, the rotating disk assembly weighed > 400 lb. You might ask "how the hell did they turn it?" The spindle drive motor was 8 hp. It drove the spindle with a set of V-belts (five, if memory serves). Fun fact: The spindle hole in a platter is about 5.5" in diameter, so you can drop an entire modern disk drive which is faster and higher-capacity than the 6603 through the spindle hole. The drive, in cabinet and all, weighed just a bit under two tons. ||||||| ||||||| ||||||| ||||||| ================= spindle supported, driven at center ||||||| ||||||| ||||||| ||||||| Both sides of all platters were used--28 surfaces. But two of the surfaces were for clock tracks and two surfaces (one per "side" of the drive) were "spares" that could replace a damaged surface. Data transfer was to/from a set of 12 heads in parallel; this helped the transfer rate considerably. There were 4 heads per surface. Outer heads had 128 sectors; inner had 100. Sector size was actually variable, but all the systems I'm aware of used a fixed sector size of 64 CM words or 320 12-bit (PP) words, out of the maximum of 348 data words. Rotational speed was supposed to be 900 rpm, though I remember seeing our mod-IIs running at 960. Anyway, that's 15 or 16 rps, so we're in the range of 60-70 ms for a rotation. Because of the need to bounce the data thru a PP to get to main memory, the normal organization was to interleave sectors 1:1. This gave a useful transfer of 64 sectors per rev, which if you cal- culate it out and tweak to today's norm of 8 bits/byte, is a bit under half a megabyte (8-bit bytes) per second. Positioning was via a hydraulic (fluid) servo, with 7 bits controlled by 7 solenoid-actuated valves (128 cylinders). The seek time was originally independent of distance (because of settling considerations), although it was influenced by rotational position--you had to wait for index after getting on track. One of the changes between the original and mod-II drives was that the mod-II allowed reading during a seek. The point of this was that you could start a seek, wait a while, then start reading. As soon as you'd read 3 sectors in a row correctly, you were on track! (The number 3 came from the mechanical characteristics of the positioner--it assured that you'd have watched through a complete oscillation as it settled.) Typical seek time was 250 ms on the original, 200 ms on the mod-II. The useful capacity of the drive, using the usual fixed sectors, was: 24 surfaces * 128 cylinders * (2 heads @ 128 sectors) + (2 heads @ 100 sectors) * 320 bits/sector (remember: parallel transfer) / 60 bits/word or about 7 mega-words. Translating from those words (60 bits) to contem- porary measure of 8-bit-bytes, it was effectively a 53-Mb drive. That says a single coffee-table-sized platter held about four and a half meg. About the platters catching fire: Has anyone ever tried to ignite a large chunk of magnesium? It's not easy. In fact, it's very difficult because it conducts heat pretty well; moreover, the metal tends to skin over if given a chance--you can manage to melt magnesium with a torch without igniting it. Thin pieces of metal or shavings are a different story, of course. But I suspect the stories of a fire in a 6603 are apocryphal. 6638: ===== The 6638 had smaller platters (26 inch, and much thinner) in a more conventional vertical-spindle arrangement. There were 72 platters in all, mounted on two spindles with separate motors. Each spindle had 18 platters above and below the motor. Head-arms and positioners were between the spindles. There were two positioners which could operate independently. The 18-platter groups were further divided into pairs of 9's (so that the positioner mechanism sat between upper and lower parts, making it symmetric). 9 platters == 18 surfaces, of which 16 were used for data. ---|--- ---|--- | <-Positioner-> | Each ---|--- ---|--- ---|--- represents 9 platters Motor Motor ---|--- ---|--- | <-Positioner-> | ---|--- ---|--- There were 6 heads per surface. Now, stop a minute and multiply that out: 6 heads * 16 surfaces/group * 2 groups per quadrant * 4 quadrants == 768 read/write heads! The geometry here was different: The transfer was still 12 bit parallel, but used 3 heads/surface on each of 4 surfaces for the 12-bit transfer. There were 32 "cylinders" (head positions), which gave the 38 a better seek time than the 03. As with the 03, sector size is variable, but with a simplistic layout you'd have 100 sectors/track, and the total capacity becomes: 768 heads * 32 head positions * 100 sectors * 320 bits/sector / 60 bits/word or about 13 mega-words. (With a more sophisticated layout, the 38 had just about twice the capacity of the 03. At maximum capacity, this behemoth was the equivalent of today's 128 Mb in 8-bit-bytes.) The 6638 turned a little faster--just under 1200 rpm. Seek was much faster--110 ms max. Data transfer was faster--1.18 usec per 12-bit PP word (vs 1.4 / 1.8 for the 03). -- Dick Dunn rcd@ico.isc.com -or- ico!rcd Boulder, CO (303)449-2870 ...Worst-case analysis must never begin with "No one would ever want..."