As someone that worked on drum memories at the end of their era, the first thing I want to say is I hope you have the maintenance manuals with detailed instructions on drum setup and adjustment. That is something that if you get wrong, and don't follow the step by step procedures pretty much exactly, you can/will destroy the drum and/or the heads. Ideally you would have the factory setup manual, but you almost certainly don't. The Burroughs 205 had an adjustment procedure called "ticking the heads". You spun the drum up, gave it an hour or so for the temperatures to stabilize in the drum and surrounds. Then you _carefully_ adjusted each head inward until you just started to hear a faint ticking sound, as the head contacted the high spots on the oxide of the drum. Then you backed the adjustment off by some fraction of a turn and locked it down. This gave optimum read amplitude and write saturation of the drum coating. Knowing about that procedure, where the heads were deliberately crashed during adjustment, makes me a lot less worried than you are about that scraped drum. It could be trash of course. But it _might not_ be trash. Bit density was really low in those days, track spacing was wide, tracks were wide. All of that gave a lot of latitude for the completely expected irregularities in the drum coating thickness. In fact the ticking procedure flattened the bumps off the coating, making it better. It would be decades before the mag tape and disk guys really got coating deposition down to a fine art. Now I saw one metallic scrape, so one track may be toast. And I'm concerned about the mildew-like stuff. It would be good to try to wipe it off, but I'm *really* concerned that the drum coating may go the way of the silk screening on the tubes, when you cleaned them. If I were doing it, I would rotate the drum manually, maybe even using a stethoscope, and listen for any ticking against the heads. If I didn't hear any, I'd put the cover back on, power it up, and give it an hour to stabilize. If it still didn't have bearing problems or obvious crashing noises, I'd try using it and see what happened. You might get a surprise. BTW, for mildew removal and general gentle cleaning, a spray can of Lysol works wonders. I wouldn't use it on the drum, but a little of it along with the soap and water (or just it and a rag) can do wonders.
I worked on repairing a G15-D when I was an undergraduate. There were very explicit warnings not to move the computer until the drum had coasted to a stop. There was a spare clock track in case the original was damaged. I think I remember that there was a paper tape cartridge that held a diagnostic that could with some fiddling with the drum heads, rewrite the clock track. In any case if you do recoat the drum, you will need to recreate the clock track.
That should not be hard as we have multiphase electric motors that can run at extremely precise speeds it is then just a matter or pulsing a write head in the correct location.
Your memory s pretty much spot on! There is indeed a read-only clock track and a spare clock track in case the first clock track gets accidentally wiped out. I haven't found any documentation outlining how to recreate the clock track if both get wiped out, but Lloyd's G15 has a good clock track on it that we can use as a foundation to build a new clock track when necessary.
@@CommodoreGregsince the drum is the source of the system clock, simply spinning the drum faster will result in a faster machine. A VFD should suffice. However, going to fast and the signals may not propagate correctly, and data may not write to the drum correctly. Since there's a manually tuned blocking oscillator and multivibrator, probably can get more than a 3-5% overclock without returning those.
@@UsagiElectric Most of those coupling capacitors with glass ends are probably good because they are hermetically sealed so that moisture can't get in. You could check for DC leakage with your voltmeter if you have any doubts but you probably have already done that. So the old computers are basic medium to high mu amplifier circuits?
Yes, this must be considered I think. The material and even thickness are already taken care of this way, and it's cheap. Spray-adhesive ought to fit the bill. In the end, only experimentation will tell if it's the way to go I think. It could be wound manually, slowly, with an usb microscope in place to align it really well I presume. At least 3" tapes are/were available! Then again, overlapping layers is maybe better anyway? I guess It can be argued that actual coating is more authentic.
That was pretty much my first thought. I could see the join being a problem. As it crosses past the heads it could introduce noise. You obviously can't get tape that wide so my next thought was to wind tape on as a spiral. That way you don't get a sharp join line crossing the heads. Any noise caused by the discontinuity between wraps is going to be much lower frequency than the data so it should hopefully be ignored by the read heads. It would also be a lot easier to apply than multiple parallel loops. You'd only need to clamp/glue each end of the tape.
as soon as you talked about head crash, I thought "that's not possible on a drum". Because the heads are static and the drum doesn't move either beyond spinning around, the heads cannot crash unless the drum bearings fail, but crud accumulating on the drum or in the gap between the heads and the drum, most certainly can damage both. The way these old 50s-60s style magnetic storage surfaces were made is however fairly primitive compared to modern storage, so it should be possible - aka relatively simple - to resurface them, as long as you can get the chemical materials required together. There are a couple IBM film (that can be found on youtube) which show the whole process, minus the chemical mixing, of how they made the RAMAC disks, and it always struck me as a "is that it?" moment, but since they did it basically by hand, it really shouldn't be entirely impossible to repeat the process
The drum memory (General Dynamics if my memory is correct) that we had in the #1 EAX telephone switch from Automatic Electric did indeed move not just rotate. The drums were vertical and conically narrower at the top. There was a centrifugal scissor mechanism that lifted the drum at a certain rpm on spin up. This allowed the heads to ride on a cushion of air when spinning at full speed and provided clearance so the heads wouldn't crash on spin down.
@@randymass6688 I'm guessing that this drum was from before flying heads, since the patent refers to an abrasion-resistant coating, which you wouldn't need on a non-contact device.
@@ke9tv I'm not sure what the surface was. All I remember was it was shiny perhaps nickel? We had scheduled routine maintenance every couple years to spin down the drums and clean the surface with lint free cotton and 99% isopropyl alcohol. We quit doing that as we usually had to change a head or two after doing that proceedure.
@@randymass6688Mostly, the drum substrate was aluminum. I know they could be recoated. You needed a big engine lathe to resurface them but the process wasn't too different from turning a brake drum and then painting a surface.
Dude, being able to re-coat a drum would be a massive game changer! I'm guessing you'll need to strip the old coating and re-grind/polish the surface to get rid of any imperfections before re-coating. Then clean and re-adjust the heads. It's an insane undertaking, but if you succeed, it could be the start of bringing an entire generation of dead machines back to life! And if you are successful, the next obvious question would be is it possible to re-coat a Hawk drive platter. I'm guessing the answer to that would be no, since you couldn't just grind the platter thinner to get rid of the divots from the head crash. It might be possible to do something even more insane like aluminium spray coating to build the thickness back up before grinding and polishing, but the level of precision would need to be way higher, and the risk of destroying more precious heads might just be too high to make it worthwhile. ... unless you can figure out how to manufacture new heads.
The initial plan for stripping the drum would actually be machining. Chuck it in the lathe and machine off the existing coating to expose clean aluminum for the new stuff to adhere to. As for getting stuff to adhere to the aluminum, I was thinking and appropriate primer sprayed thick enough and then machined to the perfect runout. Finally, the new magnetic coating would have to be precisely sprayed and hopefully self-level to the level of precision we need. Recooating a hawk platter is a whole other level of precision though. The Hawk has around a 50x tighter clearance between the head and the magnetic surface. That's a level of precision I may not be able to achieve. Fortunately, the Hawk platters are the same as the DEC RL02 platters (which is an IBM standard), and there's a ton of those still floating around out there. It's the Hawk heads that are the hard part, and rewiring some of those is... not going to be easy, haha.
@@UsagiElectric Cool! Best of luck! Yeah, repairing or making replacement Hawk drive heads would be amazing, but if it were easy, I'm sure someone would've already attempted it by now.
@@UsagiElectric I'm still watching through the series, so I don't know how far along you've gotten, but I would not attempt on a lathe. I think an OD grinder would be a much more appropriate machine.
as a Hammond organ collector/repair guy who got into building analogue synthesisers, read "The Story of Mel" in The Jargon File back in high school, and is now replacing microcontroller firmware from 2008 with CMOS/transistor/diode logic, i'm really enjoying this video, and yes CLEAN THE TUBES. :)
The gunk that got on the tubes might be of the same origin as the gunk that got on the drum. I doubt the gunk would hurt the tubes, but the unit sure looks better with clean tubes, and being able to read the tube number is another safeguard against putting the wrong tube in the wrong place. (I believe the tube numbers were etched on the glass surface so cannot be washed or rubbed off by any reasonable process.)
Hawk stores more than 100 times as much information on ~comparable surface area so 100 times as much precision is needed. Coating a 16KB drum might be marginally possible to do by hand (although I personally doubt even that) but a 20MB hard drive platter is highly unlikely to work without extremely expensive precision machines of which none remain to this day. Furthermore, we have enough non-crashed platters to store all software and non-accounting data we will ever have, the problem is the lack of working non-crashed heads - and hand-coating the platters will increase the risk of further crashes and head damage.
@@mwwhited They do use vapor deposition for the precision coating of astronomical mirrors, which also get recoated eventually, but the coating is an element (aluminum), not a compound like iron oxide. Maybe a compound would decompose?
@@kennethflorek8532I know that you can coat ITO (indium tin oxide) on glass via evaporation and titanium oxide through plasma coating with oxygen as an impurity gas. So maybe coating iron oxide would be possible.
@@ToasterWithFur Amazing. Coating telescope mirrors is not really far out. Lots of people grind their own mirrors for telescopes and send them to a place to have them coated. It's a routine thing. For iron oxide, maybe not.
@@lauram5905 One would certainly want to evaluate the properties of the existing coating compound in order to figure out how to recreate it. It would probably beat a lot of random trials and waste less time and material.
Back in the later 1970's I bought a drum memory system that had been part of a NASA computer system that was used with I recall was called the "Astronomical Orbiting Observatory". I was able to get it working as drum storage on my Motorola MC6800 home computer at the time. It had a "massive" 256K of storage for me. I could read the entire drum in 8 seconds. It used 128 heads on a large solenoid that would engage the heads once the drum was up to speed. It was made by the Vermont Research Corporation. I still have the drum assembly but had to ditch the cabinet and the transistorized circuit boards. It's probably only good as scrap aluminum now. (I'd remove NASA label(s) first.) The same computer system had a paper tape reader and punch that I also had running on my 6800 system back then.
Hi. The “blocking oscillator” is probably nothing but a fine-tuned bandpass filter. You can tell, because there’s an inductor in series with the signal path from the 2 microsecond multivibrator to the blocking oscillator; this’ll stop frequencies higher than the bandwidth, and the capacitor will block DC and frequencies lower than the bandwidth. So only frequencies in the passband will make it through. In this case, you can also see that someone put a sketch of the output waveform off the output, which shows a ringing pulse with a DC offset of -20 volts. This comes from the output of T6 (pins 5 & 6), which is driven by the plate (pin 6) of V4 to the -100 volt supply. It’s likely the designers were trying to tame the odd harmonics if the square wave pulse from the 2 usec MV.
17:07 Machines using drum main memory tended not to be sequential in executing instructions. If the address of each instruction was just N+1 you’d have to wait for a minimum of one drum rotation per instruction, while the ALU etc. is orders of magnitude faster. So the instruction format includes a value N (7 bits) which indicates where the next instruction is to be found on the drum. So allowing for the execution time of each type of instruction you could line up the next instruction to be read without any latency. This was termed ‘optimal programming’. Alan Turing who conceived this idea for his ACE computer, on which the G15 is loosely based, had the idea that an assembler program would keep track of the timings automatically to avoid this being a huge pain for the programmer.
Indeed, this was the job of the SOAP assembler for the IBM 650 which was also drum-based. It did a pretty good job of optimizing the next instruction address.
I wonder if ATR Magnetics in PA could help out with this. They are the only manufacturer of analog audio tape in the US. If not the actual application process, I'll bet they could supply a bucket of the magnetic coating.
@@ReinaldoRauch Ha! For sure. Way above my head. To me magnetic goo is magnetic goo. I imagine there were PhDs done on the subject at one point regarding grain size, exact chemical composition, additives, yadda yadda. My thought process would be that this is old enough, the density is low enough, you're dealing with binary, that there'd be a lot of wiggle room and if you could get it to stick and at the right thickness you'd have a nice shot starting with an existing liquid that's good enough for tape, although it might cost thousands of dollars for the quantity needed for the necessary experimentation. Going the other direction and trying to get random goo to sound like a high quality analog audio recording definitely seems impossible. Such an interesting challenge. May have to scrape some rust off my truck and experiment myself.
@@barcodenosebleed5485 I'd think one could get powder; the goo part is matched to whatever it has to cling to. I'd be thinking epoxy here. There will be pretty substantial centrifugal forces involved that aren't involved with any kind of tape. Perhaps application of an excess followed by vacuum curing (to outgas any entrapped air) followed by machining, applying a bit more epoxy minus the powder, and polishing would be the process needed.
In 1962, my high school obtained a Monroe "Monrobot" desk sized computer. It had paper tape input and output for long term memory, an IBM non selectric typewriter also useable as I/O, and used magnetic drum memory as the main memory for the computer. So it was much like the system you describe. This was my first opportunity to program a computer, and it was great fun!
The Monroe MONROBOT machines were a series of computers originally based on magnetic drum storage. The series had quite a range of models, some of which were rather large in scale that used drum as a high-speed storage unit and magnetic core for main memory. The earlier machines(1950's), used the drum as main memory. One MONROBOT drum-based machine was made famous by predicting the 1956 win of Dwight D. Eisenhower for the presidency of the US. Sadly, as far as I know, there aren't any examples of any of these machines left anywhere that I've been able to discover. If anyone knows of any Monroe MONROBOT machine around today, please, let the world know about it! Such a machine deserves a place in a museum along with the G-15, LGP-30, and IBM drum-based computer, and perhaps even to be restored to operational condition.
Thanks for your comments and description of the history of the Monroe Monrobot. Since it was the first computer I ever had a chance to use, it will always have a place in my heart!
Hi, the blocking oscillator is a type of transformer (inductance) coupled oscillator that works by "blocking" hence the name the current flow of the tube via the feedback path produced by the coil from plate to the grid! Also, the chokes on the multivibrator may be peaking coils, they were used in TV circuits to provide higher bandwidth and better specs, altough with a few downsides with I don't remember. As soon as I remember something else I'll update the comment
Also the tube designation can be used to recognize the brand that build it! Glass etched numerals are for GE, the little frame around the name is from RCA I think Usually small companies would produce their own tubes by just rebranding lots from various higher-volume factories
As an example, here in Italy we had (until the 70s I think) both FIVRE and ATES, both italian company names that, as acronyms meant "Italian radio electronic valve factory" -> "Fabbrica Italiana Valvole Radio Elettriche" and " Aquila Tubi Elettronici Semiconduttori" "Aquila (italian city) Electron Tubes and Semiconductors" Basically all the tubes we had were literal copies from the crates the US soldiers left here after the war, and then we started producing them under RCA license, keeping a similar box apperance and designations, altough a personal one (used only in italy) was introduced
Interestingly, the wires are all soldered instead of wire-wrapped. It's gonna be a nightmare to repair if any are broken, haha. But it is a thing of beauty!
@@UsagiElectric I meant that string that the wire bundles are wrapped in instead of the zip-ties that they'd use today. I got the name wrong. :/ I love the look of that. Everything meticulously planned out and organized. I saw a satellite backup at the Tellus Science Museum that was done the same way. The panels were removed and you could see inside it. Absolutely gorgeous.
Back in the early '80's, part of my BT apprenticeship was using lacing twine and learning the art of 'plate wiring'. Very satisfying when done nicely. 😊@@FirstLast-vr7es
The best way to get ones mind around how the drum works and is used is to recognize the Huskey's design was strongly influenced by his time at NPL, and in particular Turing's 5th ACE design. This is the design, (a simplified version) that was realized as the Pilot ACE (largely at Huskey's encouragement). The drum here essentially implements the same memory as the mecury delay lines in the Pilot ACE. If you are able to recoat the drum, the read-only clock track will have to be re-recorded. As for how the crashes might have taken place, IIRC the original procedures called for adjusting the head gap by screwing the head in util it just lightly touched the drum and then backing it out a certain amount. If that was done improperly, then the heads could dig into the surface. Of course, bad bearings could also lead to vibration in the drum causing it to hit the heads. In other conversations about G15 restorations, I have talked about building a drum emulator, which could still be good even if the drums at hand can all be restored.
Back in my early s/w dev days I worked on real-time systems that used drum memory, but not used as RAM (this was the 1980s). The drums were used as fast HDs, and with a fixed head per track there was no seek delay.
Early Hewlett Packard timeshared BASIC systems used a fixed-head magnetic drum as a swapping device. Since the heads were all fixed, there was no positioning time, and the moving-head hard disks of the time simply weren't fast enough to provide reasonable response time in a multi-user timeshared environment. These systems were used into the mid-1970's, by which time multiple moving-head disk drives with smart controllers that optimized access by sorting the order of disk seeks made swapping across multiple drives quick enough to eliminate the need for the drum. Back in the drum days of HP Timeshared BASIC, there were some operator-only commands (from the console terminal only) that would allow a program to be copied to the drum to allow it to load much faster than when loaded from the moving-head disk. I also remember touring the data center for a DEC PDP-10-based timesharing service in downtown Portland, OR where a HUGE (in a box about 10 feet long, four feet deep, and 5 feet tall) drum was using for the swap device on a timeshared system that hosted something like 200 simultaneous users. This was sometime around 1975 or 1976. I remember being told that it took almost 25 minutes for the drum to spin down to a stop when it was turned off (which was very rarely done) because of the sheer inertia of the drum. So, drum memory technology did live on, but not as main memory, but as a high-speed swap/paging space.
I am so sorry that Beau has crossed the rainbow bridge. I don't know your beliefs concerning these things, but in my belief system, I know he is there on the other side of the rainbow, happy and no longer dealing with age/disease, waiting to be reunited with you when the time comes. Losing a beloved animal companion is so very difficult, different in so many ways from losing a human loved one. No less difficult, though. My heart goes out to you and Ms. Usagi.
Please bear with me. Was in the AF 50 years ago and I was part of the maintenance department to service the ITT SPE (Stored Program Element), a par of them. Amongst the hardware that it used was drum memory for storage, 16 of them. From what I remember the heads were held off the drum by springs and need air to load the heads to the drums. When the spring broke, they would crash onto the drum surface. It looks just like what you are seeing here, only it was a volume set of heads (16?). We had a repair depo that did the repair/refurbishing of the drum units, and we did the final testing. We got very good at swapping out drum units, all 450lb. The SPE x 2 had about 65 cabinets, 16 drum units, 16 Potter computer tape units, card punch/reader and two printers. Next door was a transistorized version of a Sage. This was a transistorized computer made operation some time in 1962. It had 32,768 16-bit words with parity of core memory. The drums had four volumes at 4k word each. The interrupt system was controlled at each instruction. This drove the engineers from Honeywell batty and they would say the old girl would take her sweet time to acknowledge. LOL Hey, she was running in a 1mz clock, and the memory cycle took 4 cycles to do a memory operation! The low-end DEC that they were using as an interface must have been almost as fast as the SPE.
Fantastik was the cleaner that we used when I worked for a computer store back in the 80's when dirty computers came in for service. Always did a great job on both plastic and painted surfaces without damage and without needing huge amounts of elbow grease.. Removed tobacco smoke and all sorts of grunge very effectively and customers were always impressed that their machines looked new when they came back from the shop..
Just love the loom work, I learned the wax string looming as an apprentice and its so much nicer than tie wraps. Especially the ones that have been cut with side cutter, over flush cutter. 👍👍
Yes, this IS very fascinating for sure! It's always really interesting for me to see how they worked old technology because they hadn't yet developed what we have today! I've always been used to _storage_ needing moving parts until the years of flash, but never used to just a version of _memory_ needing its _own_ motor (even back when a stationary memory, core, was already also an option)!
I think you're right about the importance of having the real physical drum rather than an emulation, despite the idea of emulating such an odd thing having some appeal in itself. Drum is such a rare and unique technology that having another working example is pretty significant; I assume there are probably less than 10 functional drum memory systems left in the world at this point. Looking forward to seeing the drum swap and recoating efforts.
Cutting edge! We take so much for granted these days. I remember late 70s my Data General engineer brother brought home a broken drum memory. It was so very heavy.
considering it is a drum even if one or more heads would be diging into it, you could probably just replace the heads and point entire row to a different position, but given this is basically just ram, drum is possibly restorable if you could find a crazy enough person to recoat it
It might be possible to remove offending heads and rewire the head assembly to the working ones at the expense of losing some memory capacity. The downside is obviously that the existing software won't work. But if we don't have much software or won't be demonstrating it in the museum anyway - I think this is the way to go.
Depending on the configuration of the heads you wouldn't lose capacity but lose one bit of the each word. The #1 EAX that I worked on had a 24 bit word hence multiple rows of 26 heads. 24 data, 1 parity and one clock head. As the drum spun it would read one complete word per clock bits. All bits of a word read in parallel not serial as in modern day hard drives. I believe there also was an extra set of spare heads to be used in such a failure case. Not sure about that though as it was a looong time ago. @@jwhite5008
Thank you - that cleared up how drum memory works. With hard drives and floppy disks, we have sectoring with gaps to ensure existing data isn't overwritten (excepting Amiga drives that write whole tracks). Having the fast registers inside of the long registers clarified perfectly. For both those reasons, it is not so helpful to think of them as a differently shaped hard disk, but instead it is better to think of drums as being much more like mercury-delay line memory, only instead of audio pulses down a tube full of mercury, it is a sweep of magnetic coating.
I worked on a drum-based computer in the mid-1970s, the Philco AN/FYQ-9, which was used as a data processor for a portion of NORAD's Radar network. The tracks were non-volatile, so the system could be shut down and restarted without having to reload its software. In fact, it was a royal pain to reload the software, which was stored on IBM-style punched cards, one word per card, and cards had to be hand-loaded into the reader one at a time. This was intended for loading maintenance test routines, which consisted of a small number of instructions, but there were times when software updates were sent that required reloading the whole program. We also had to do this when one of the heads burned out, and we had to switch to one of the two spare tracks on the drum. I'm pretty sure this is how most drum computers worked. The fact that the Bendix G15 erases and re-writes every track as it goes (and especially the short line's use of only a fraction of a full track) tells me that its design was based on a delay-line-based computer, or even perhaps that it was initially meant to be a delay line computer. Delay lines have to be continuously re-written in order to retain data in the first place, and of course cannot retain data over a power shutdown. Drum memories were far less expensive and more reliable than equivalent delay line memory, and their use overlapped that of magnetic core memory. Good luck with your drum work, and with the rest of the restoration of this beast!
Best of luck, brother! I've complete confidence in your ability to bring the Bendix back to life after your astounding work (along with all your helpers!) in bringing multiple Centurion machines back to working order and beyond.
One thing I've wondered about is how much information one could store with a triode, transformer, and resistors wired as a blocking oscillator. For example, if a system had 1MHz master clock which was divided by ten to yield 100kHz master phase reference, then a 100kHz blocking oscillator which was fed the 1MHz clock would have ten different stable phase relationships with the master clock. Add a couple of semiconductor diodes, repeat the circuit a few hundred times, and one could have a somewhat-slow-to-write RAM (I think a single-triode oscillator might need to be driven a few cycles at the correct phase in order to settle in) using only one glass tube for each pair of decimal digits.
- Fascinating. - Thx for all your hard effort, and diligence/persistence. - Keep up the great work. - Looking forward to seeing the progress, eventual spinning of the drum, and finally, functioning of the entire system!...
Wow, great episode! Side bar, wanted to thank you for this great content over the years. I am a software guy, but love learning about how the hardware works and understanding the history of computing machines. After watching your videos, and other content from adjacent channels on TH-cam, I finally got the confidence to do some hardware stuff myself. Now no where near your level, I just took apart my old Atari 2600 that was acting a little flaky and recapped it and replacement soldered on some new connectors. It was immensely rewarding to see the old machine have new life breathed into it and work better than when I started. Just wanted to thank you for the great content that inspired me to take the first plunge into this world. Keep up the great work!
29 bits? Truly weird. And a clever way of partitioning the drum. I had a good laugh at the paperweighth part... and again, I hope I learned something. I wonder what it'd take to re-coat the drum, but I'm mighty rootin' for ya! :) Tube markings were sometimes etched with hydrofluoric acid, it's gonna hold under scrubbing.
the way they thought about word length was in decimal point precision in those days, so 29 bits is like precise to 8 decimal places or something like that.
@@tripplefives1402 Yes, I know. I think that getting the specific form of iron oxide required will be a bit more complicated than letting some iron filings rust a bit.
This reminds me, I've seen magnetic inks to be used in inkjet printers, e.g. for printing credit-card style strips... I wonder if something like that would be durable enough, or maybe could be baked-on... heh, or even printed onto plastic or paler and wrapped around the drum, hah!
Everyone (me included) is on the expectation of learning about the drum recoating process, but first things first, let's have that spare drum fitted and see how it goes. Check the bearings, tolerances, head calibration, lubrication, I'm sure you'll know better than me what to do anyway. Best of luck and thank you for yet another incredible video!
A drum that old may have been designed for contact heads, like the heads of an old audio-tape machine or VCR. Flying heads most likely came later, so 'head crash' is likely not a meaningful idea. But that water damage (or whatever that glop is) could easily have been abrasive and scraped the oxide away. IBM 650 drums and UNIVAC FASTRAND drums increased their bandwidth by using all tracks at once. Each read and write head corresponded with one bit of the word, and words were written in parallel. The 650 was a 1+1-address machine. Every instruction was a jump and held the address of the next instruction. The assembler would place instructions so that the next instruction would be arriving under the heads just in time for the next fetch. Stan Kelly-Bootle once described the FASTRAND drum as a storage device - for the storage of angular momentum. He continued that if more than three were in the same building, the installer had to consult Sperry's resident geophysicist for the correct latitude-dependent orientation of the drum axes. When I was an undergrad at Dartmouth, the computer center had a couple of FASTRAND drums that had been retired from NASA. There was a weird 'Apollo Support Device' (occupying an entire 19-inch rack) for interfacing the GE 635 computer to the Sperry drums. It wouldn't have been workable except for the fact that the GE and Univac machines both had the same 36-bit word size. The IBM 650 was a weird architecture. It used bi-quinary encoding - it was a decimal machine with two-out-of-five bits representing a digit. 400 digits of the recirculating memory (which worked more or less like the CPU registers on the Bendix) were devoted to the addition and multiplication tables. There was a truly bizarre bit of code in the bootloader, getting that memory initialized, when all the arithmetic it could do had to use already-initialized values! It was entirely possible for a rogue program to scribble on the arithmetic tables, and just TRY to debug that! The Royal-McBee LGP-30 was similar to the 650, and gave rise to the Story of Mel en.wikipedia.org/wiki/The_Story_of_Mel . I heard about that one directly from Ed Nather; I was friends with his daughter. The original tale at users.cs.utah.edu/~elb/folklore/mel-annotated/mel-annotated.html wraps up _so much_ computing history - check out the annotations!
i really like how you denote the parts of the circuit with red and blue to denote positive and ground potentials in the schematic circuit. great restoration!
When I mention that I like to read about computer history I often get..."I didn't know that computers had a history". I'm glad that I did all that reading because now I can somewhat understand this incredibly interesting project. Even the cleaning process is interesting. New to this channel, off to binge watch.
I'm a trained xerox tech but never did I imagine that from printing to computing, it's basically the same. Write, erase, write... Repeat. All these men/ woman are marvelous from then. My friends dad worked for lucent technologies in Naperville Illinois and bell labs. He helped with invention of call waiting. He made his computer on a piece of plywood and we would play pong on it. Then he made his own stereo that was amazing! I truly miss him and wish I was older to appreciate it more!
So glad to hear you're going to explore recoating that drum. Seems like a viable repair. After all, the tolerances on these old systems are a lot 'looser' than today's equipment.
CRC 2-26 is awesome for cleaning wrinkle paint. I use it on my old vintage test equipment and also brings back the sheen on black plastics as well. WD-40 works too but I find it too aggressive.
I reckon a camera setup on the electric meter on full power up if you get there will be a laugh 😆 I did a stint at IBM here in the UK back in the 80's. They were coating their own disk's there, quite interesting. It was all very high tech at the time. Do remember hearing a few head crashes on testing though, not that often fortunately!
I’m so excited that you get to experience this and have the opportunity to share it with us. Gorgeous machines like this set the foundation for our technological world today, and the people behind it had to figure all this out with barely anything to go on.
So your guess is that the nicotine found its way inside the drum enclosure, but not to all of the rest of the machine. Um, no. The drum is using an oxide coating like that used by magnetic tape and older disk drives.
@@BrightBlueJim: So at 4:24, that's what's called a "control panel". You can tell it's a control panel because it has controls on it and is not a magnetic drum. It is obviously not iron oxide, due to the presence of blue paint, of indicator lights, of control knobs, and the absence of voice coils. It is also conspicuously non-cylindrical.
On the paint, of course. The drum unit does not appear to be hermetically sealed, however, so even in that day there may have been "No smoking in the computer room" rules.
A minor correction: At 5:42 you describe the circuit involving V2 on the clock chassis as a monostable producing a precise pulse length. It's actually a Schmitt trigger, which produces a pulse of precise amplitude, with an output that snaps quickly between two voltage levels. Note that the small capacitor across the 56K resistor is not a timing capacitor - its' function is to compensate for the Miller capacitance arising from the grid-plate capacitance of the second triode, thus enabling a fast rise and fall time of the output pulse.
The high school I went to had an old computer that was donated to the school during the mid-1970's. The computer was made by 3M (the Scotch Tape (tm) people). It was designed in around '62 or '63, using discrete transistor logic. The CPU was considerably smaller than the G15 as a result, the whole CPU and the magnetic drum main memory fit in the rough equivalent of about 14U of rack space today. The machine had a 24-bit word, and was, like the G-15, completely bit-serial in architecture. It didn't store the working registers on the drum like the G15, instead opting for chains of flip flops organized as shift registers. It had a 24-bit accumulator, a 24-bit "B" register, a 5-bit opcode register, and some temporary registers for storing the operand address and next instruction address. Addresses were organized as block, track, and sector. Each block was 2K words, and there were 4 blocks, so the drum held a total of 8K words (by 24 bits, or the equivalent 24K bytes). Each block was divided into tracks which I can't remember the size of, and each word in a track was a sector. The previous life of the machine was as a process control computer for managing flow of natural gas in the distribution network of a gas utility. There was two complete CPUs, connected by a single 24-bit hardware-handshake register that allowed them to talk to each other. It was a master-slave arrangement, with a primary and secondary CPU. If the primary failed, the secondary had identical drum and register content to the primary (they were constantly synchronized through software) and could take over. The I/O system was extensive, with relay inputs and outputs, discrete logic inputs and outputs, counter/timers, and analog to digital and digital to analog converters. The I/O system was contained in a rack by itself that was 7ft. tall and stuffed with electronics. The instruction set was pretty basic, with just 24-bit sign-magnitude add/subtract, load & store, bitwise AND/OR/INVERT, test instructions, and a jump instruction. No index registers or indirect addressing. Any table accesses or computed branching involved self-modifying code. The instruction set documentation included calculations for operand and next instruction sector number based on the rotational speed of the drum. To optimize programming for maximum speed, you had to use these calculations to determine where to place an operand and next instruction. For example, for ADD, the operand sector needed to be the sector of the instruction plus three, and the next instruction sector was the sector of the instruction plus six. You didn't have to use these calculations, but if you didn't, the drum would likely have to make an extra revolution in order to find the operand sector, and yet another revolution to find the the next instruction to be executed. Needless to say, optimizing things like table lookups and computing branches were cumbersome. When I started at the high school, both CPUs ran just fine, although only the primary CPU had an interface to the 33ASR Teletype (110 baud) console, so any I/O the secondary machine wanted to do had to be routed through the primary CPU via the interconnect register. The machine also had a transistorized Parabam real time clock with BCD output that could be read all at once (HH:MM:SS), two banks of BCD thumbwheel switches that could be used for inputting into the accumulator (one was six digits with sign, the other was three digits without sign), and a very wide carriage IBM typebar typewriter that was used for logging output. I was never able to get that typewriter working, though. My crowning achievement on the machine was that I wrote a functional FOCAL (FORmula CALculator, a simple interactive computer language created by Digital Equipment Corp.(DEC) for the PDP-8) interpreter for the machine during my senior year. It would take about two seconds to take a simple line of a program entered from the Teletype, run basic syntax checking, and store it into program storage. Doing a direct command, such as "TYPE 2+2" took almost five seconds to execute and print out "= 4.0000" on the Teletype. The floating point math routines were particularly challenging to write to try to eke out as much speed as possible. The whole thing was painfully slow, but it worked. I wrote it to prove to myself it could be done, not as a practical tool. It was coded entirely in native machine code (8-digit Octal words). There was roughly room for about a 40 line program, and all of the rest of the space on the drum was filled by the interpreter. One day during my sophomore year, I was working at the machine, and suddenly this horrible shrieking noise filled the room. I immediately hit the master breaker and the noise lessened in frequency as the drums spun down. It turned out that a bearing had failed in the drum of the secondary CPU, and that caused a massive head crash. It was beyond repair, with deep grooves carved by the heads through the oxide layer into the aluminum substrate of the drum. The secondary CPU then became a parts donor to keep the primary CPU running. I was able to keep the primary CPU running through my senior year. I went back to the school the next year to visit, and found the machine was GONE! A new math director had taken over when the previous director moved into district administration, and the idiot determined that the machine was useless for anything even though it still ran, and had it junked. I was heartbroken. Very sorry to see the drum in the G15 was bad...but it looked nothing nearly as bad as the secondary CPU drum in the old 3M machine. I hope that the drum from System Source's other G-15 is good, and that the clock tracks are still OK. Without the clock tracks, the machine is just as useless as a one with a bad drum. If you re-coat the crashed drum, you'll have to figure out how to re-write the clock tracks properly. Most of the drum-based computers had the clock tracks written on the drum at the factory by a special fixture, and in many cases, the heads that wrote the clock tracks were only used to write them with the fixture, and were not user accessible. There obviously was no way to programmatically re-write the clock tracks, because it's a chicken-and-egg problem. Be aware that if you re-coat the drum, that some means will have to be found to make the aluminum drum surface completely smooth (no grooves), and that the coating will have to meet very exacting thickness and magnetic properties in order for the original heads to be able to properly pick up and alter the state of the magnetic field in the magnetic coating. Also make sure that there is no runout in the bearings, as that could be why the drum crashed in the first place. It'd suck to go to all the work to resurface the drum only to have it crash again on spin-up. Drum memory computers and accounting machines were quite common in the 1950s and early 1960's as it was less-expensive than magnetic core at the time. As automated methods became available for producing magnetic core arrays, the cost of core went down pretty dramatically, and by the mid-1960's, core had pretty much taken over even in the lower-end computers. Machines like the G-15, the LGP-30 (which was a genius design by Stanley Frankel using many tricks to reduce the number of tubes to a bare minimum), the IBM drum-memory machines, and this great old 3M machine (that I was so lucky to have been able to tinker with for four years...I learned so much from it) were wonderfully rube goldberg-ish ways of mixing moving parts with electronics to make low-cost computers possible. Interestingly, there were even two (only two) electronic calculators made in the mid-1960's that used spinning rust technology as their main storage...the Wyle Laboratories WS-01 (oldcalculatormuseum.com/w-wyle.html), and the Canon Canola 167 (oldcalculatormuseum.com/w-can167.html). Both of these calculators had short market lifetimes because of the issues with spinning media and head crashes, as well as being slow. Magnetostritive delay lines and magnetic core quickly eliminated spinning rust as a way to store working registers in electronic calculators.
I was excited to see a usagi electric upload and even more excited to see rotating drum memory. Can't wait to see it in action. I love seeing retro and vintage computing, but seeing really obscure and fascinating hardware like this just really is something else. Thank you usagi electric!
Wow.... it's magnificent! Always in awe of a good wire wrap.....Backside of the pre-amp is such beautiful construction. All the cards cleaned up look amazing. It's just such a damn cool looking machine!
Absolutely! They will have definitely picked up a large amount of oxide at the minimum. It's going to require quite a bit of effort, but I think if we approach it methodically, that we can bring it back, even if we have to make new heads
@@UsagiElectric Is there existing software that we would like to run on the system? If not I strongly suggest just not using the dead tracks, possibly rewiring the machine to make sure all the internal registers are on good tracks. I see only a minority of unusable tracks, the remaining ones should be more than enough for museum demo purposes if custom software is executed.
As another poster commented, the heads are (apparently) fixed and could not actually "crash." Something like loose dirt may have caused the scarring on the drum. That said though, the heads should be inspected and cleaned. I wonder if routine head cleaning was a required maintenance procedure. Is the drum in a contained housing with filtered air. Is there an air filter?
@4:54, I have never seen the kind of display that shows counter of " Total Hours". I believe it was manufactured by Veeder-Root Counter Company. Which is similar to Luggage locker with counter where you put the luggage in locker and put quarter in slot before closing the door and take the key with you.
Awesome tech! And I totally agree with you; the fun is in the complete package. My passion is in slightly newer computers (c64, c128, Amiga) but Original hardware, floppies, CRTs etc.
This is such an exciting project! I'm fascinated by the engineering that went into the design and manufacture of these devices. The quick access storage is reminiscent of the VT240 video RAM, where the horizontal blanking period of the raster scan is used to move bits into the RAM and shift lines of text when scrolling. You steal that little bit of leftover resource for something else! I'm glad there's a plan to replace the broken drum because I'm really looking forward to see this project succeed. Good luck!
Kinda excited to see this line of work, almost as much as the Centurion printer! A project I've kinda wanted to do for a while (but probably won't ever be able to) is to build a drum memory unit. Having a magnetic surface is quite the sticking point!
This video brings back some memories - of working on one of these units while I was at Rose-Hulman with Beta Iota Tau's founding chapter. *NOT* all fun memories - though the machine was "only" 25 years old more-or-less at the time....
Ok, if that recoating thing works out, that'll be *incredible!!* There are *so many* old machines that could benefit! Not only spinning-drum RAM, but also (perhaps, eventually) restoring *hard drive platters!* I realize the density of the material on the HDD platters would have to be higher than that on the spinning-drum RAM, but I'd imagine that if you can solve the spinning-drum problem, it's only another few steps to solve the HDD platter problem! How exciting!!
I watched some great videos on youtube of companies testing and making disk packs from back in the day. My personal fav. was them spinning the clean metal platter around and pouring on the rust solution on the platter. Found an article saying at that time was one of the best ways to coat the platter. Reminded me of one of those art paint sets from as seen on tv. Spin the paper around and apply paint. I figured it wouldn't be to hard to make new metal platters and coat the platters. Most of the time would be testing the formula to see things about thickness after coating and how much (I forget) to store the data and read it back. A good example also is Curious Marc. He had the old HP terminals using a tape drive. After replacing the belts on the drive. He remove of swapped out a resistor? to increase the write head output. Then he was able to use newer and more common DC tapes.
Cool. I used to design drum memories for Univac and Control Date. The rage at the time was "flying" heads to reduce the gap between the head and the drum. We argued with early disk developers about which was best. Clearly, they won.
I used drum memory for an old real-time aircraft training simulator. The computer memory had only 1024 words of memory and the code was loaded in strips into the core memory. Real-time was controlled but the rotation speed of the drum. Reason for drum memory was the astronomical cost of core memory! Interesting times!
Bringing these old systems back to life is amazing. I sometimes wonder if any of the old National Cash Register CRAM (Card Random Access Memory) units survived the scrapyard. Cardboard cards had a magnetic stripe with mechanism to retrieve cards for read/write.
Recoating hit me at approx. 24:00, and of course Usagi himself way before me. Reason is that I saw an old film showing how IBM engineers literally “painted” their first hard disk prototype platters. The fast rotation of the platter was an integral part of creating an even coat (very messy too), but how that process can apply perpendicular to the rotation (a drum, not a disk) is of course beyond me. The chemistry was indeed very secretive according to said film, but I would think that Usagis friends have all relevant knowledge today, 60+ years later. Best of luck! 👍
Disks can be spin coated. that process give very precise deposition of material. Maybe he can test magnetic paint on his crashed hard disks first. I'm not sure how to coat a drum
I'd think rotating the drum at a say a 45 degree angle would work. Let a combination of centrifugal force and gravity distribute the liquid from one end to the other. Or maybe they blew air along the length of the drum, though that might create ripples. Would be worth testing on surplus aluminum or cardboard tubes if the exact method isn't known.
I always thought drum memory was bigger, but maybe that's a nicely miniaturised version of the bigger stuff. If the recoating works on a drum, might it also work on those hard disk platters for the Centurion?
There is a G15 at the Computer History Museum in Silicon Valley (Mountain View, CA), which belonged to Harry Huskey, who designed the CPU architecture, this machine has a gold plaque on it with his name, and was a present to him when he left Bendix. AFAIK, it was fully functional when it was put away in his barn in Santa Cruz, where it sat for a few decades before being shipped to the CHM, I think in the mid 90s? (I helped them pack it up for shipment).
edit to above. Wait, the Huskey G15 was sent to BOSTON, not Silicon Valley. I know there is one on static display at the CHM in Mountain View, but its a different one.
These are fascinating machines. I worked in the computer Museum of the university of applied sciences in Kiel, Germany for a while. They have a couple of German Zuse computers. The Zuse Z22 is a tube computer that uses a drum as RAM too but I think it uses core memory for the registers. Unfortunately none of the computers there are in working condition and the museum was not interested in restoring them.
Same. But i'm certain it's possible. Don't forget, the sort of technology and equipment that the engineers were working with in the 1950's is exactly the sort of things a hobbyist machinist has in his garage today. I'm honestly surprised there haven't been many hobbyist RDM units manufactured.
Agree completely with the concept, also I love when people do modern software that runs on 100% original hardware as you have shown before. I say it's time to get a drum or recoat the one you have!
I really like this effort, and I love to have the methods back I case we lose the ability to make machines and have to recover from the end of the work. 8K ram on magnetic tape interleaving fast and slow memory genius….
When trying to describe computers to non technical people, "memory" being very short term is one of the hardest topics to get round... "memory" sounds like it should be long term. These old machines, where what we would call "memory" and "storage" getting kinda blurred together, make some sense of that "bad choice" of name.
First, give a cleaning to the drum, and coat it with a good layering of acrylic; as an Objet d'art it is its own beauty. Second, get a new cylinder of aluminum and start fresh. From the deep dive on TH-cam, the coatings (generalized) appear to be vapor deposited Nickle/Phosphorous layer and a final layer replace Phosphorous with Cobalt. Polish this as you go :)
I just gotta say that when I saw the power supply cord at the beginning of the video it really reminded me of the supply line that powered our entire mobile home. It would be interesting to know how much power the RDM requires VS the rest of the system.
From reading about the G15, the drum had a timing track that was written at manufacturing time and was never erased/re-written. If you were to recoat the drum, you would presumably need to find a way to re-write the timing track as well. All doable but certainly not trivial.
He has the advantage of modern computing equipment available to generate the required signals. Could probably program an Arduino to do it if information is available about what needs to go on the timing tracks.
That 'dual use' of the drum track for both long and short term use is ingenious. When I first saw "room for 108 29-bit words", I had assumed that was all the way around the drum. But now I understand those 108 words fit in about 7/8ths of the circumference and the other 1/8th is for the short term. But the 'short term' is being rewritten over a 'shifting' 1/8th of the drum constantly. Two uses for the same magnetic track is really bizarre!!!
There is a fascinating book called "How to build a working digital computer" by Alcosser, Phillips and Wolk. It gives instructions on how build a computer on blocks of wood with paper clips, lamps, batteries, etc. Not joking! It has a chapter on how to build drum memory out of a can!
When I was in high school we had a computer called an ECP-18 whose main memory was a drum. Very basic by modern standards, but at the time it was the coolest thing in the world.
With these micro-sprinkle like surfaces, I had good experience with a good IPA scrub so far. The only ones I haven't managed to get clean so far are the ones where someone used a marker to write stuff on it :(
Absolutely can not wait to see this thing going, I think it’s worth it to simulate the hard drive if there’s no other choice, especially if they want it running in the museum, just for reliability. It would be epic though if you can recoat that drum. It was made 70+ years ago, we must still have the knowledge, just wondering if the machines that did it still exist lol
At (roughly) 21:00 - Wow. I actually understood all of that. That is amazing! Looking forward to more of this computer that's three years older than me!
As someone that worked on drum memories at the end of their era, the first thing I want to say is I hope you have the maintenance manuals with detailed instructions on drum setup and adjustment. That is something that if you get wrong, and don't follow the step by step procedures pretty much exactly, you can/will destroy the drum and/or the heads. Ideally you would have the factory setup manual, but you almost certainly don't.
The Burroughs 205 had an adjustment procedure called "ticking the heads". You spun the drum up, gave it an hour or so for the temperatures to stabilize in the drum and surrounds. Then you _carefully_ adjusted each head inward until you just started to hear a faint ticking sound, as the head contacted the high spots on the oxide of the drum. Then you backed the adjustment off by some fraction of a turn and locked it down. This gave optimum read amplitude and write saturation of the drum coating.
Knowing about that procedure, where the heads were deliberately crashed during adjustment, makes me a lot less worried than you are about that scraped drum. It could be trash of course. But it _might not_ be trash. Bit density was really low in those days, track spacing was wide, tracks were wide. All of that gave a lot of latitude for the completely expected irregularities in the drum coating thickness. In fact the ticking procedure flattened the bumps off the coating, making it better. It would be decades before the mag tape and disk guys really got coating deposition down to a fine art.
Now I saw one metallic scrape, so one track may be toast. And I'm concerned about the mildew-like stuff. It would be good to try to wipe it off, but I'm *really* concerned that the drum coating may go the way of the silk screening on the tubes, when you cleaned them. If I were doing it, I would rotate the drum manually, maybe even using a stethoscope, and listen for any ticking against the heads. If I didn't hear any, I'd put the cover back on, power it up, and give it an hour to stabilize. If it still didn't have bearing problems or obvious crashing noises, I'd try using it and see what happened. You might get a surprise.
BTW, for mildew removal and general gentle cleaning, a spray can of Lysol works wonders. I wouldn't use it on the drum, but a little of it along with the soap and water (or just it and a rag) can do wonders.
Lysol consist primarily of alcohol so that explains that.
I worked on repairing a G15-D when I was an undergraduate. There were very explicit warnings not to move the computer until the drum had coasted to a stop. There was a spare clock track in case the original was damaged. I think I remember that there was a paper tape cartridge that held a diagnostic that could with some fiddling with the drum heads, rewrite the clock track. In any case if you do recoat the drum, you will need to recreate the clock track.
That should not be hard as we have multiphase electric motors that can run at extremely precise speeds it is then just a matter or pulsing a write head in the correct location.
Your memory s pretty much spot on!
There is indeed a read-only clock track and a spare clock track in case the first clock track gets accidentally wiped out. I haven't found any documentation outlining how to recreate the clock track if both get wiped out, but Lloyd's G15 has a good clock track on it that we can use as a foundation to build a new clock track when necessary.
We know you're secretly going to mod the track and overclock it by 0.000001 MIPS.
@@CommodoreGregsince the drum is the source of the system clock, simply spinning the drum faster will result in a faster machine. A VFD should suffice. However, going to fast and the signals may not propagate correctly, and data may not write to the drum correctly.
Since there's a manually tuned blocking oscillator and multivibrator, probably can get more than a 3-5% overclock without returning those.
@@UsagiElectric Most of those coupling capacitors with glass ends are probably good because they are hermetically sealed so that moisture can't get in. You could check for DC leakage with your voltmeter if you have any doubts but you probably have already done that. So the old computers are basic medium to high mu amplifier circuits?
I haven't been this excited since CuriousMarc and his team restored the Xerox Alto!
Oh the work on that Diablo 2.5 drive tho'! Watching them get the drive working
I wonder if you could just stick a turn of modern magnetic tape to the surface of the drum.
What rpm are we looking at tho? Interesting idea if the tape holds on.
Yes, this must be considered I think. The material and even thickness are already taken care of this way, and it's cheap.
Spray-adhesive ought to fit the bill.
In the end, only experimentation will tell if it's the way to go I think.
It could be wound manually, slowly, with an usb microscope in place to align it really well I presume.
At least 3" tapes are/were available! Then again, overlapping layers is maybe better anyway?
I guess It can be argued that actual coating is more authentic.
The adhesive mag tape method is nondestructive... Put the tape on and continue restoring and when you are ready, scrape the tape off and recoat.
That was pretty much my first thought. I could see the join being a problem. As it crosses past the heads it could introduce noise. You obviously can't get tape that wide so my next thought was to wind tape on as a spiral. That way you don't get a sharp join line crossing the heads. Any noise caused by the discontinuity between wraps is going to be much lower frequency than the data so it should hopefully be ignored by the read heads. It would also be a lot easier to apply than multiple parallel loops. You'd only need to clamp/glue each end of the tape.
as soon as you talked about head crash, I thought "that's not possible on a drum". Because the heads are static and the drum doesn't move either beyond spinning around, the heads cannot crash unless the drum bearings fail, but crud accumulating on the drum or in the gap between the heads and the drum, most certainly can damage both. The way these old 50s-60s style magnetic storage surfaces were made is however fairly primitive compared to modern storage, so it should be possible - aka relatively simple - to resurface them, as long as you can get the chemical materials required together. There are a couple IBM film (that can be found on youtube) which show the whole process, minus the chemical mixing, of how they made the RAMAC disks, and it always struck me as a "is that it?" moment, but since they did it basically by hand, it really shouldn't be entirely impossible to repeat the process
There were also machines which had flying heads.
The drum memory (General Dynamics if my memory is correct) that we had in the #1 EAX telephone switch from Automatic Electric did indeed move not just rotate. The drums were vertical and conically narrower at the top. There was a centrifugal scissor mechanism that lifted the drum at a certain rpm on spin up. This allowed the heads to ride on a cushion of air when spinning at full speed and provided clearance so the heads wouldn't crash on spin down.
@@randymass6688 I'm guessing that this drum was from before flying heads, since the patent refers to an abrasion-resistant coating, which you wouldn't need on a non-contact device.
@@ke9tv I'm not sure what the surface was. All I remember was it was shiny perhaps nickel? We had scheduled routine maintenance every couple years to spin down the drums and clean the surface with lint free cotton and 99% isopropyl alcohol. We quit doing that as we usually had to change a head or two after doing that proceedure.
@@randymass6688Mostly, the drum substrate was aluminum. I know they could be recoated. You needed a big engine lathe to resurface them but the process wasn't too different from turning a brake drum and then painting a surface.
Dude, being able to re-coat a drum would be a massive game changer! I'm guessing you'll need to strip the old coating and re-grind/polish the surface to get rid of any imperfections before re-coating. Then clean and re-adjust the heads. It's an insane undertaking, but if you succeed, it could be the start of bringing an entire generation of dead machines back to life!
And if you are successful, the next obvious question would be is it possible to re-coat a Hawk drive platter. I'm guessing the answer to that would be no, since you couldn't just grind the platter thinner to get rid of the divots from the head crash. It might be possible to do something even more insane like aluminium spray coating to build the thickness back up before grinding and polishing, but the level of precision would need to be way higher, and the risk of destroying more precious heads might just be too high to make it worthwhile.
... unless you can figure out how to manufacture new heads.
Something like an epoxy base might make sense here.
The initial plan for stripping the drum would actually be machining. Chuck it in the lathe and machine off the existing coating to expose clean aluminum for the new stuff to adhere to.
As for getting stuff to adhere to the aluminum, I was thinking and appropriate primer sprayed thick enough and then machined to the perfect runout. Finally, the new magnetic coating would have to be precisely sprayed and hopefully self-level to the level of precision we need.
Recooating a hawk platter is a whole other level of precision though. The Hawk has around a 50x tighter clearance between the head and the magnetic surface. That's a level of precision I may not be able to achieve. Fortunately, the Hawk platters are the same as the DEC RL02 platters (which is an IBM standard), and there's a ton of those still floating around out there. It's the Hawk heads that are the hard part, and rewiring some of those is... not going to be easy, haha.
@@UsagiElectric Cool! Best of luck!
Yeah, repairing or making replacement Hawk drive heads would be amazing, but if it were easy, I'm sure someone would've already attempted it by now.
@@UsagiElectric I'm still watching through the series, so I don't know how far along you've gotten, but I would not attempt on a lathe. I think an OD grinder would be a much more appropriate machine.
as a Hammond organ collector/repair guy who got into building analogue synthesisers, read "The Story of Mel" in The Jargon File back in high school, and is now replacing microcontroller firmware from 2008 with CMOS/transistor/diode logic, i'm really enjoying this video, and yes CLEAN THE TUBES. :)
The gunk that got on the tubes might be of the same origin as the gunk that got on the drum. I doubt the gunk would hurt the tubes, but the unit sure looks better with clean tubes, and being able to read the tube number is another safeguard against putting the wrong tube in the wrong place. (I believe the tube numbers were etched on the glass surface so cannot be washed or rubbed off by any reasonable process.)
pulled an old hammond analog gear drive oscillator stack recently.. was still perfect... a real beauty.
If the drum recoating works out, maybe you can recoat the hawk drive platterrs also? Might be worth a try :)
Hawk stores more than 100 times as much information on ~comparable surface area so 100 times as much precision is needed. Coating a 16KB drum might be marginally possible to do by hand (although I personally doubt even that) but a 20MB hard drive platter is highly unlikely to work without extremely expensive precision machines of which none remain to this day.
Furthermore, we have enough non-crashed platters to store all software and non-accounting data we will ever have, the problem is the lack of working non-crashed heads - and hand-coating the platters will increase the risk of further crashes and head damage.
Should work with vapor disposition.
@@mwwhited
They do use vapor deposition for the precision coating of astronomical mirrors, which also get recoated eventually, but the coating is an element (aluminum), not a compound like iron oxide. Maybe a compound would decompose?
@@kennethflorek8532I know that you can coat ITO (indium tin oxide) on glass via evaporation and titanium oxide through plasma coating with oxygen as an impurity gas. So maybe coating iron oxide would be possible.
@@ToasterWithFur Amazing. Coating telescope mirrors is not really far out. Lots of people grind their own mirrors for telescopes and send them to a place to have them coated. It's a routine thing. For iron oxide, maybe not.
Recoating the drum would be an awesome collab with Ben from Applied Science!
I was just about to say, the one person with all the equipment necessary to try multiple avenues and variations on the patent is Ben Krasnow
@@lauram5905 One would certainly want to evaluate the properties of the existing coating compound in order to figure out how to recreate it. It would probably beat a lot of random trials and waste less time and material.
Back in the later 1970's I bought a drum memory system that had been part of a NASA computer system that was used with I recall was called the "Astronomical Orbiting Observatory". I was able to get it working as drum storage on my Motorola MC6800 home computer at the time. It had a "massive" 256K of storage for me. I could read the entire drum in 8 seconds. It used 128 heads on a large solenoid that would engage the heads once the drum was up to speed. It was made by the Vermont Research Corporation. I still have the drum assembly but had to ditch the cabinet and the transistorized circuit boards. It's probably only good as scrap aluminum now. (I'd remove NASA label(s) first.) The same computer system had a paper tape reader and punch that I also had running on my 6800 system back then.
Hi. The “blocking oscillator” is probably nothing but a fine-tuned bandpass filter. You can tell, because there’s an inductor in series with the signal path from the 2 microsecond multivibrator to the blocking oscillator; this’ll stop frequencies higher than the bandwidth, and the capacitor will block DC and frequencies lower than the bandwidth. So only frequencies in the passband will make it through. In this case, you can also see that someone put a sketch of the output waveform off the output, which shows a ringing pulse with a DC offset of -20 volts. This comes from the output of T6 (pins 5 & 6), which is driven by the plate (pin 6) of V4 to the -100 volt supply. It’s likely the designers were trying to tame the odd harmonics if the square wave pulse from the 2 usec MV.
17:07 Machines using drum main memory tended not to be sequential in executing instructions. If the address of each instruction was just N+1 you’d have to wait for a minimum of one drum rotation per instruction, while the ALU etc. is orders of magnitude faster. So the instruction format includes a value N (7 bits) which indicates where the next instruction is to be found on the drum. So allowing for the execution time of each type of instruction you could line up the next instruction to be read without any latency. This was termed ‘optimal programming’. Alan Turing who conceived this idea for his ACE computer, on which the G15 is loosely based, had the idea that an assembler program would keep track of the timings automatically to avoid this being a huge pain for the programmer.
Indeed, this was the job of the SOAP assembler for the IBM 650 which was also drum-based. It did a pretty good job of optimizing the next instruction address.
I wonder if ATR Magnetics in PA could help out with this. They are the only manufacturer of analog audio tape in the US. If not the actual application process, I'll bet they could supply a bucket of the magnetic coating.
at least that would be a great video explaining why that wouldn´t work
@@ReinaldoRauch Ha! For sure. Way above my head. To me magnetic goo is magnetic goo. I imagine there were PhDs done on the subject at one point regarding grain size, exact chemical composition, additives, yadda yadda. My thought process would be that this is old enough, the density is low enough, you're dealing with binary, that there'd be a lot of wiggle room and if you could get it to stick and at the right thickness you'd have a nice shot starting with an existing liquid that's good enough for tape, although it might cost thousands of dollars for the quantity needed for the necessary experimentation. Going the other direction and trying to get random goo to sound like a high quality analog audio recording definitely seems impossible.
Such an interesting challenge. May have to scrape some rust off my truck and experiment myself.
Was wondering if they could simply paste down video tape to the drum and calibrate the head distance to match the slightly larger drum diameter.
@@barcodenosebleed5485 I'd think one could get powder; the goo part is matched to whatever it has to cling to. I'd be thinking epoxy here. There will be pretty substantial centrifugal forces involved that aren't involved with any kind of tape. Perhaps application of an excess followed by vacuum curing (to outgas any entrapped air) followed by machining, applying a bit more epoxy minus the powder, and polishing would be the process needed.
That would be epic if you can recoat that drum successfully.
That's all I was thinking about the whole time. It has to be possible since it was done in the 50's.
That would open the doors to recoating platters too.
We have a company in Belgium that still makes magnetic heads Am Belgium
I don't see any reason we can't manufacture and coat a brand new drum as well. Seems easy enough with a lathe.
@@josugambee3701 Probably a centerless grinder but yes - it's gotta be possible !
In 1962, my high school obtained a Monroe "Monrobot" desk sized computer. It had paper tape input and output for long term memory, an IBM non selectric typewriter also useable as I/O, and used magnetic drum memory as the main memory for the computer. So it was much like the system you describe.
This was my first opportunity to program a computer, and it was great fun!
The Monroe MONROBOT machines were a series of computers originally based on magnetic drum storage. The series had quite a range of models, some of which were rather large in scale that used drum as a high-speed storage unit and magnetic core for main memory. The earlier machines(1950's), used the drum as main memory. One MONROBOT drum-based machine was made famous by predicting the 1956 win of Dwight D. Eisenhower for the presidency of the US. Sadly, as far as I know, there aren't any examples of any of these machines left anywhere that I've been able to discover. If anyone knows of any Monroe MONROBOT machine around today, please, let the world know about it! Such a machine deserves a place in a museum along with the G-15, LGP-30, and IBM drum-based computer, and perhaps even to be restored to operational condition.
Thanks for your comments and description of the history of the Monroe Monrobot. Since it was the first computer I ever had a chance to use, it will always have a place in my heart!
Hi, the blocking oscillator is a type of transformer (inductance) coupled oscillator that works by "blocking" hence the name the current flow of the tube via the feedback path produced by the coil from plate to the grid!
Also, the chokes on the multivibrator may be peaking coils, they were used in TV circuits to provide higher bandwidth and better specs, altough with a few downsides with I don't remember. As soon as I remember something else I'll update the comment
Also the tube designation can be used to recognize the brand that build it!
Glass etched numerals are for GE, the little frame around the name is from RCA I think
Usually small companies would produce their own tubes by just rebranding lots from various higher-volume factories
As an example, here in Italy we had (until the 70s I think) both FIVRE and ATES, both italian company names that, as acronyms meant "Italian radio electronic valve factory" -> "Fabbrica Italiana Valvole Radio Elettriche" and " Aquila Tubi Elettronici Semiconduttori" "Aquila (italian city) Electron Tubes and Semiconductors"
Basically all the tubes we had were literal copies from the crates the US soldiers left here after the war, and then we started producing them under RCA license, keeping a similar box apperance and designations, altough a personal one (used only in italy) was introduced
Absolutely fascinating. That wire wrap and routing was a thing of beauty.
Interestingly, the wires are all soldered instead of wire-wrapped. It's gonna be a nightmare to repair if any are broken, haha.
But it is a thing of beauty!
@@UsagiElectric I meant that string that the wire bundles are wrapped in instead of the zip-ties that they'd use today. I got the name wrong. :/ I love the look of that. Everything meticulously planned out and organized. I saw a satellite backup at the Tellus Science Museum that was done the same way. The panels were removed and you could see inside it. Absolutely gorgeous.
Back in the early '80's, part of my BT apprenticeship was using lacing twine and learning the art of 'plate wiring'. Very satisfying when done nicely. 😊@@FirstLast-vr7es
The best way to get ones mind around how the drum works and is used is to recognize the Huskey's design was strongly influenced by his time at NPL, and in particular Turing's 5th ACE design. This is the design, (a simplified version) that was realized as the Pilot ACE (largely at Huskey's encouragement). The drum here essentially implements the same memory as the mecury delay lines in the Pilot ACE. If you are able to recoat the drum, the read-only clock track will have to be re-recorded. As for how the crashes might have taken place, IIRC the original procedures called for adjusting the head gap by screwing the head in util it just lightly touched the drum and then backing it out a certain amount. If that was done improperly, then the heads could dig into the surface. Of course, bad bearings could also lead to vibration in the drum causing it to hit the heads. In other conversations about G15 restorations, I have talked about building a drum emulator, which could still be good even if the drums at hand can all be restored.
Another positive side of the drum swap is that you’ll be able to free whoever, or whatever, is trapped inside that drum.
lol, great point
Back in my early s/w dev days I worked on real-time systems that used drum memory, but not used as RAM (this was the 1980s). The drums were used as fast HDs, and with a fixed head per track there was no seek delay.
Some early big 19“ rack style hard disks also sometimes had one platter with fixed heads for quick access.
Early Hewlett Packard timeshared BASIC systems used a fixed-head magnetic drum as a swapping device. Since the heads were all fixed, there was no positioning time, and the moving-head hard disks of the time simply weren't fast enough to provide reasonable response time in a multi-user timeshared environment. These systems were used into the mid-1970's, by which time multiple moving-head disk drives with smart controllers that optimized access by sorting the order of disk seeks made swapping across multiple drives quick enough to eliminate the need for the drum. Back in the drum days of HP Timeshared BASIC, there were some operator-only commands (from the console terminal only) that would allow a program to be copied to the drum to allow it to load much faster than when loaded from the moving-head disk. I also remember touring the data center for a DEC PDP-10-based timesharing service in downtown Portland, OR where a HUGE (in a box about 10 feet long, four feet deep, and 5 feet tall) drum was using for the swap device on a timeshared system that hosted something like 200 simultaneous users. This was sometime around 1975 or 1976. I remember being told that it took almost 25 minutes for the drum to spin down to a stop when it was turned off (which was very rarely done) because of the sheer inertia of the drum. So, drum memory technology did live on, but not as main memory, but as a high-speed swap/paging space.
amazing !!!
one of the most impressive things I have ever seen.
please more videos
I am so sorry that Beau has crossed the rainbow bridge. I don't know your beliefs concerning these things, but in my belief system, I know he is there on the other side of the rainbow, happy and no longer dealing with age/disease, waiting to be reunited with you when the time comes. Losing a beloved animal companion is so very difficult, different in so many ways from losing a human loved one. No less difficult, though. My heart goes out to you and Ms. Usagi.
Please bear with me. Was in the AF 50 years ago and I was part of the maintenance department to service the ITT SPE (Stored Program Element), a par of them. Amongst the hardware that it used was drum memory for storage, 16 of them. From what I remember the heads were held off the drum by springs and need air to load the heads to the drums. When the spring broke, they would crash onto the drum surface. It looks just like what you are seeing here, only it was a volume set of heads (16?). We had a repair depo that did the repair/refurbishing of the drum units, and we did the final testing. We got very good at swapping out drum units, all 450lb. The SPE x 2 had about 65 cabinets, 16 drum units, 16 Potter computer tape units, card punch/reader and two printers. Next door was a transistorized version of a Sage. This was a transistorized computer made operation some time in 1962. It had 32,768 16-bit words with parity of core memory. The drums had four volumes at 4k word each. The interrupt system was controlled at each instruction. This drove the engineers from Honeywell batty and they would say the old girl would take her sweet time to acknowledge. LOL Hey, she was running in a 1mz clock, and the memory cycle took 4 cycles to do a memory operation! The low-end DEC that they were using as an interface must have been almost as fast as the SPE.
I used to work for 3M Magnetic Media Division. You have a task ahead of you to be successful at recoating the oxide.
Fantastik was the cleaner that we used when I worked for a computer store back in the 80's when dirty computers came in for service. Always did a great job on both plastic and painted surfaces without damage and without needing huge amounts of elbow grease.. Removed tobacco smoke and all sorts of grunge very effectively and customers were always impressed that their machines looked new when they came back from the shop..
Fantastik was pretty much the same as Formula 409, which I think is still available.
29:42 I did actually network G115 machines - they were the transistorised successors - back in around 1982 using PDP11s as protocol converters!
Wow, what a Beauty! 👍👏
Just love the loom work, I learned the wax string looming as an apprentice and its so much nicer than tie wraps. Especially the ones that have been cut with side cutter, over flush cutter. 👍👍
"Help me, I'm inside the drum." Tech humor is universal across time and space...
Yes, this IS very fascinating for sure! It's always really interesting for me to see how they worked old technology because they hadn't yet developed what we have today! I've always been used to _storage_ needing moving parts until the years of flash, but never used to just a version of _memory_ needing its _own_ motor (even back when a stationary memory, core, was already also an option)!
I think you're right about the importance of having the real physical drum rather than an emulation, despite the idea of emulating such an odd thing having some appeal in itself. Drum is such a rare and unique technology that having another working example is pretty significant; I assume there are probably less than 10 functional drum memory systems left in the world at this point. Looking forward to seeing the drum swap and recoating efforts.
It sure looks like gunk got into the machine, perhaps when stored somewhere it should not have been.
Cutting edge! We take so much for granted these days. I remember late 70s my Data General engineer brother brought home a broken drum memory. It was so very heavy.
You are rapidly becoming the US expert in obscure tube computers.
considering it is a drum even if one or more heads would be diging into it, you could probably just replace the heads and point entire row to a different position, but given this is basically just ram, drum is possibly restorable if you could find a crazy enough person to recoat it
It might be possible to remove offending heads and rewire the head assembly to the working ones at the expense of losing some memory capacity. The downside is obviously that the existing software won't work. But if we don't have much software or won't be demonstrating it in the museum anyway - I think this is the way to go.
Depending on the configuration of the heads you wouldn't lose capacity but lose one bit of the each word. The #1 EAX that I worked on had a 24 bit word hence multiple rows of 26 heads. 24 data, 1 parity and one clock head. As the drum spun it would read one complete word per clock bits. All bits of a word read in parallel not serial as in modern day hard drives. I believe there also was an extra set of spare heads to be used in such a failure case. Not sure about that though as it was a looong time ago. @@jwhite5008
This is fast becoming my favourite channel on TH-cam. Great work!
Thank you!
Thank you - that cleared up how drum memory works. With hard drives and floppy disks, we have sectoring with gaps to ensure existing data isn't overwritten (excepting Amiga drives that write whole tracks). Having the fast registers inside of the long registers clarified perfectly. For both those reasons, it is not so helpful to think of them as a differently shaped hard disk, but instead it is better to think of drums as being much more like mercury-delay line memory, only instead of audio pulses down a tube full of mercury, it is a sweep of magnetic coating.
I worked on a drum-based computer in the mid-1970s, the Philco AN/FYQ-9, which was used as a data processor for a portion of NORAD's Radar network. The tracks were non-volatile, so the system could be shut down and restarted without having to reload its software. In fact, it was a royal pain to reload the software, which was stored on IBM-style punched cards, one word per card, and cards had to be hand-loaded into the reader one at a time. This was intended for loading maintenance test routines, which consisted of a small number of instructions, but there were times when software updates were sent that required reloading the whole program. We also had to do this when one of the heads burned out, and we had to switch to one of the two spare tracks on the drum.
I'm pretty sure this is how most drum computers worked. The fact that the Bendix G15 erases and re-writes every track as it goes (and especially the short line's use of only a fraction of a full track) tells me that its design was based on a delay-line-based computer, or even perhaps that it was initially meant to be a delay line computer. Delay lines have to be continuously re-written in order to retain data in the first place, and of course cannot retain data over a power shutdown. Drum memories were far less expensive and more reliable than equivalent delay line memory, and their use overlapped that of magnetic core memory.
Good luck with your drum work, and with the rest of the restoration of this beast!
Best of luck, brother! I've complete confidence in your ability to bring the Bendix back to life after your astounding work (along with all your helpers!) in bringing multiple Centurion machines back to working order and beyond.
Thank you!
Stubbornness has its benefits at times, haha.
As long as we don't quit on it, I'm sure we'll bring it up!
Blocking oscillators are used to provide narrow trigger pulses from an input clock. It's basically a pulse shaper.
One thing I've wondered about is how much information one could store with a triode, transformer, and resistors wired as a blocking oscillator. For example, if a system had 1MHz master clock which was divided by ten to yield 100kHz master phase reference, then a 100kHz blocking oscillator which was fed the 1MHz clock would have ten different stable phase relationships with the master clock. Add a couple of semiconductor diodes, repeat the circuit a few hundred times, and one could have a somewhat-slow-to-write RAM (I think a single-triode oscillator might need to be driven a few cycles at the correct phase in order to settle in) using only one glass tube for each pair of decimal digits.
- Fascinating.
- Thx for all your hard effort, and diligence/persistence.
- Keep up the great work.
- Looking forward to seeing the progress, eventual spinning of the drum, and finally, functioning of the entire system!...
Wow, great episode!
Side bar, wanted to thank you for this great content over the years. I am a software guy, but love learning about how the hardware works and understanding the history of computing machines.
After watching your videos, and other content from adjacent channels on TH-cam, I finally got the confidence to do some hardware stuff myself. Now no where near your level, I just took apart my old Atari 2600 that was acting a little flaky and recapped it and replacement soldered on some new connectors. It was immensely rewarding to see the old machine have new life breathed into it and work better than when I started.
Just wanted to thank you for the great content that inspired me to take the first plunge into this world. Keep up the great work!
29 bits? Truly weird. And a clever way of partitioning the drum. I had a good laugh at the paperweighth part... and again, I hope I learned something. I wonder what it'd take to re-coat the drum, but I'm mighty rootin' for ya! :)
Tube markings were sometimes etched with hydrofluoric acid, it's gonna hold under scrubbing.
@@tripplefives1402 The coating will most likely be ferric oxide powder, not iron filings. Also, metallic paints are made with aluminium flakes.
the way they thought about word length was in decimal point precision in those days, so 29 bits is like precise to 8 decimal places or something like that.
@@tripplefives1402 Yes, I know. I think that getting the specific form of iron oxide required will be a bit more complicated than letting some iron filings rust a bit.
@@tripplefives1402 that'd make sense, could be done in a jig, with hand drive or some motor drive turning very slowly.
This reminds me, I've seen magnetic inks to be used in inkjet printers, e.g. for printing credit-card style strips... I wonder if something like that would be durable enough, or maybe could be baked-on... heh, or even printed onto plastic or paler and wrapped around the drum, hah!
You gotta love TTL circuits... Tube-Tube-Logic 😀
Everyone (me included) is on the expectation of learning about the drum recoating process, but first things first, let's have that spare drum fitted and see how it goes. Check the bearings, tolerances, head calibration, lubrication, I'm sure you'll know better than me what to do anyway. Best of luck and thank you for yet another incredible video!
A drum that old may have been designed for contact heads, like the heads of an old audio-tape machine or VCR. Flying heads most likely came later, so 'head crash' is likely not a meaningful idea. But that water damage (or whatever that glop is) could easily have been abrasive and scraped the oxide away.
IBM 650 drums and UNIVAC FASTRAND drums increased their bandwidth by using all tracks at once. Each read and write head corresponded with one bit of the word, and words were written in parallel. The 650 was a 1+1-address machine. Every instruction was a jump and held the address of the next instruction. The assembler would place instructions so that the next instruction would be arriving under the heads just in time for the next fetch.
Stan Kelly-Bootle once described the FASTRAND drum as a storage device - for the storage of angular momentum. He continued that if more than three were in the same building, the installer had to consult Sperry's resident geophysicist for the correct latitude-dependent orientation of the drum axes.
When I was an undergrad at Dartmouth, the computer center had a couple of FASTRAND drums that had been retired from NASA. There was a weird 'Apollo Support Device' (occupying an entire 19-inch rack) for interfacing the GE 635 computer to the Sperry drums. It wouldn't have been workable except for the fact that the GE and Univac machines both had the same 36-bit word size.
The IBM 650 was a weird architecture. It used bi-quinary encoding - it was a decimal machine with two-out-of-five bits representing a digit. 400 digits of the recirculating memory (which worked more or less like the CPU registers on the Bendix) were devoted to the addition and multiplication tables. There was a truly bizarre bit of code in the bootloader, getting that memory initialized, when all the arithmetic it could do had to use already-initialized values! It was entirely possible for a rogue program to scribble on the arithmetic tables, and just TRY to debug that!
The Royal-McBee LGP-30 was similar to the 650, and gave rise to the Story of Mel en.wikipedia.org/wiki/The_Story_of_Mel . I heard about that one directly from Ed Nather; I was friends with his daughter. The original tale at users.cs.utah.edu/~elb/folklore/mel-annotated/mel-annotated.html wraps up _so much_ computing history - check out the annotations!
i really like how you denote the parts of the circuit with red and blue to denote positive and ground potentials in the schematic circuit. great restoration!
When I mention that I like to read about computer history I often get..."I didn't know that computers had a history". I'm glad that I did all that reading because now I can somewhat understand this incredibly interesting project. Even the cleaning process is interesting. New to this channel, off to binge watch.
Amazing work restoring such classic equipment. Love to see these marvels of engineering come back to life decades after they were last shut down.
I'm a trained xerox tech but never did I imagine that from printing to computing, it's basically the same. Write, erase, write... Repeat. All these men/ woman are marvelous from then. My friends dad worked for lucent technologies in Naperville Illinois and bell labs. He helped with invention of call waiting. He made his computer on a piece of plywood and we would play pong on it. Then he made his own stereo that was amazing! I truly miss him and wish I was older to appreciate it more!
The legends name is Dan Loftus. A true genius I knew and didn't know until later in life ..
So glad to hear you're going to explore recoating that drum. Seems like a viable repair. After all, the tolerances on these old systems are a lot 'looser' than today's equipment.
CRC 2-26 is awesome for cleaning wrinkle paint. I use it on my old vintage test equipment and also brings back the sheen on black plastics as well. WD-40 works too but I find it too aggressive.
I reckon a camera setup on the electric meter on full power up if you get there will be a laugh 😆
I did a stint at IBM here in the UK back in the 80's. They were coating their own disk's there, quite interesting. It was all very high tech at the time.
Do remember hearing a few head crashes on testing though, not that often fortunately!
I’m so excited that you get to experience this and have the opportunity to share it with us. Gorgeous machines like this set the foundation for our technological world today, and the people behind it had to figure all this out with barely anything to go on.
I would guess the brown residue is nicotine. It was used in the 1950s and 1960s, after all.
Edit: at 4:24, on the control panel.
So your guess is that the nicotine found its way inside the drum enclosure, but not to all of the rest of the machine. Um, no. The drum is using an oxide coating like that used by magnetic tape and older disk drives.
@@BrightBlueJim: So at 4:24, that's what's called a "control panel". You can tell it's a control panel because it has controls on it and is not a magnetic drum. It is obviously not iron oxide, due to the presence of blue paint, of indicator lights, of control knobs, and the absence of voice coils. It is also conspicuously non-cylindrical.
@@Jesse-qy6ur My mistake. My "brown residue, I thought you were referring to the brown stuff that is shown on the drum.
On the paint, of course. The drum unit does not appear to be hermetically sealed, however, so even in that day there may have been "No smoking in the computer room" rules.
@@Jesse-qy6ur Oh, I thought you were calling the drum memory "brown residue"
A minor correction: At 5:42 you describe the circuit involving V2 on the clock chassis as a monostable producing a precise pulse length. It's actually a Schmitt trigger, which produces a pulse of precise amplitude, with an output that snaps quickly between two voltage levels. Note that the small capacitor across the 56K resistor is not a timing capacitor - its' function is to compensate for the Miller capacitance arising from the grid-plate capacitance of the second triode, thus enabling a fast rise and fall time of the output pulse.
The high school I went to had an old computer that was donated to the school during the mid-1970's. The computer was made by 3M (the Scotch Tape (tm) people). It was designed in around '62 or '63, using discrete transistor logic. The CPU was considerably smaller than the G15 as a result, the whole CPU and the magnetic drum main memory fit in the rough equivalent of about 14U of rack space today. The machine had a 24-bit word, and was, like the G-15, completely bit-serial in architecture. It didn't store the working registers on the drum like the G15, instead opting for chains of flip flops organized as shift registers. It had a 24-bit accumulator, a 24-bit "B" register, a 5-bit opcode register, and some temporary registers for storing the operand address and next instruction address. Addresses were organized as block, track, and sector. Each block was 2K words, and there were 4 blocks, so the drum held a total of 8K words (by 24 bits, or the equivalent 24K bytes). Each block was divided into tracks which I can't remember the size of, and each word in a track was a sector. The previous life of the machine was as a process control computer for managing flow of natural gas in the distribution network of a gas utility. There was two complete CPUs, connected by a single 24-bit hardware-handshake register that allowed them to talk to each other. It was a master-slave arrangement, with a primary and secondary CPU. If the primary failed, the secondary had identical drum and register content to the primary (they were constantly synchronized through software) and could take over. The I/O system was extensive, with relay inputs and outputs, discrete logic inputs and outputs, counter/timers, and analog to digital and digital to analog converters. The I/O system was contained in a rack by itself that was 7ft. tall and stuffed with electronics. The instruction set was pretty basic, with just 24-bit sign-magnitude add/subtract, load & store, bitwise AND/OR/INVERT, test instructions, and a jump instruction. No index registers or indirect addressing. Any table accesses or computed branching involved self-modifying code. The instruction set documentation included calculations for operand and next instruction sector number based on the rotational speed of the drum. To optimize programming for maximum speed, you had to use these calculations to determine where to place an operand and next instruction. For example, for ADD, the operand sector needed to be the sector of the instruction plus three, and the next instruction sector was the sector of the instruction plus six. You didn't have to use these calculations, but if you didn't, the drum would likely have to make an extra revolution in order to find the operand sector, and yet another revolution to find the the next instruction to be executed. Needless to say, optimizing things like table lookups and computing branches were cumbersome. When I started at the high school, both CPUs ran just fine, although only the primary CPU had an interface to the 33ASR Teletype (110 baud) console, so any I/O the secondary machine wanted to do had to be routed through the primary CPU via the interconnect register. The machine also had a transistorized Parabam real time clock with BCD output that could be read all at once (HH:MM:SS), two banks of BCD thumbwheel switches that could be used for inputting into the accumulator (one was six digits with sign, the other was three digits without sign), and a very wide carriage IBM typebar typewriter that was used for logging output. I was never able to get that typewriter working, though. My crowning achievement on the machine was that I wrote a functional FOCAL (FORmula CALculator, a simple interactive computer language created by Digital Equipment Corp.(DEC) for the PDP-8) interpreter for the machine during my senior year. It would take about two seconds to take a simple line of a program entered from the Teletype, run basic syntax checking, and store it into program storage. Doing a direct command, such as "TYPE 2+2" took almost five seconds to execute and print out "= 4.0000" on the Teletype. The floating point math routines were particularly challenging to write to try to eke out as much speed as possible. The whole thing was painfully slow, but it worked. I wrote it to prove to myself it could be done, not as a practical tool. It was coded entirely in native machine code (8-digit Octal words). There was roughly room for about a 40 line program, and all of the rest of the space on the drum was filled by the interpreter. One day during my sophomore year, I was working at the machine, and suddenly this horrible shrieking noise filled the room. I immediately hit the master breaker and the noise lessened in frequency as the drums spun down. It turned out that a bearing had failed in the drum of the secondary CPU, and that caused a massive head crash. It was beyond repair, with deep grooves carved by the heads through the oxide layer into the aluminum substrate of the drum. The secondary CPU then became a parts donor to keep the primary CPU running. I was able to keep the primary CPU running through my senior year. I went back to the school the next year to visit, and found the machine was GONE! A new math director had taken over when the previous director moved into district administration, and the idiot determined that the machine was useless for anything even though it still ran, and had it junked. I was heartbroken. Very sorry to see the drum in the G15 was bad...but it looked nothing nearly as bad as the secondary CPU drum in the old 3M machine. I hope that the drum from System Source's other G-15 is good, and that the clock tracks are still OK. Without the clock tracks, the machine is just as useless as a one with a bad drum. If you re-coat the crashed drum, you'll have to figure out how to re-write the clock tracks properly. Most of the drum-based computers had the clock tracks written on the drum at the factory by a special fixture, and in many cases, the heads that wrote the clock tracks were only used to write them with the fixture, and were not user accessible. There obviously was no way to programmatically re-write the clock tracks, because it's a chicken-and-egg problem. Be aware that if you re-coat the drum, that some means will have to be found to make the aluminum drum surface completely smooth (no grooves), and that the coating will have to meet very exacting thickness and magnetic properties in order for the original heads to be able to properly pick up and alter the state of the magnetic field in the magnetic coating. Also make sure that there is no runout in the bearings, as that could be why the drum crashed in the first place. It'd suck to go to all the work to resurface the drum only to have it crash again on spin-up. Drum memory computers and accounting machines were quite common in the 1950s and early 1960's as it was less-expensive than magnetic core at the time. As automated methods became available for producing magnetic core arrays, the cost of core went down pretty dramatically, and by the mid-1960's, core had pretty much taken over even in the lower-end computers. Machines like the G-15, the LGP-30 (which was a genius design by Stanley Frankel using many tricks to reduce the number of tubes to a bare minimum), the IBM drum-memory machines, and this great old 3M machine (that I was so lucky to have been able to tinker with for four years...I learned so much from it) were wonderfully rube goldberg-ish ways of mixing moving parts with electronics to make low-cost computers possible. Interestingly, there were even two (only two) electronic calculators made in the mid-1960's that used spinning rust technology as their main storage...the Wyle Laboratories WS-01 (oldcalculatormuseum.com/w-wyle.html), and the Canon Canola 167 (oldcalculatormuseum.com/w-can167.html). Both of these calculators had short market lifetimes because of the issues with spinning media and head crashes, as well as being slow. Magnetostritive delay lines and magnetic core quickly eliminated spinning rust as a way to store working registers in electronic calculators.
I was excited to see a usagi electric upload and even more excited to see rotating drum memory. Can't wait to see it in action. I love seeing retro and vintage computing, but seeing really obscure and fascinating hardware like this just really is something else. Thank you usagi electric!
Wow.... it's magnificent! Always in awe of a good wire wrap.....Backside of the pre-amp is such beautiful construction. All the cards cleaned up look amazing. It's just such a damn cool looking machine!
You will also need to closely inspect the heads. They crashed hard and may be damaged as well.
Went here to post this.
They are most definitely damaged, the question is: are they unusable?
Absolutely!
They will have definitely picked up a large amount of oxide at the minimum. It's going to require quite a bit of effort, but I think if we approach it methodically, that we can bring it back, even if we have to make new heads
@@UsagiElectric Wait, what? Make new heads - is that even possible??? What do those heads look like?
@@UsagiElectric Is there existing software that we would like to run on the system?
If not I strongly suggest just not using the dead tracks, possibly rewiring the machine to make sure all the internal registers are on good tracks.
I see only a minority of unusable tracks, the remaining ones should be more than enough for museum demo purposes if custom software is executed.
As another poster commented, the heads are (apparently) fixed and could not actually "crash." Something like loose dirt may have caused the scarring on the drum. That said though, the heads should be inspected and cleaned. I wonder if routine head cleaning was a required maintenance procedure. Is the drum in a contained housing with filtered air. Is there an air filter?
@4:54, I have never seen the kind of display that shows counter of " Total Hours". I believe it was manufactured by Veeder-Root Counter Company. Which is similar to Luggage locker with counter where you put the luggage in locker and put quarter in slot before closing the door and take the key with you.
Another great vid. Your enthusiasm is infectious.
These machines are incredibly fascinating. Keep up the good work.
Awesome tech! And I totally agree with you; the fun is in the complete package. My passion is in slightly newer computers (c64, c128, Amiga) but Original hardware, floppies, CRTs etc.
This is such an exciting project! I'm fascinated by the engineering that went into the design and manufacture of these devices. The quick access storage is reminiscent of the VT240 video RAM, where the horizontal blanking period of the raster scan is used to move bits into the RAM and shift lines of text when scrolling. You steal that little bit of leftover resource for something else! I'm glad there's a plan to replace the broken drum because I'm really looking forward to see this project succeed. Good luck!
Kinda excited to see this line of work, almost as much as the Centurion printer!
A project I've kinda wanted to do for a while (but probably won't ever be able to) is to build a drum memory unit. Having a magnetic surface is quite the sticking point!
This video brings back some memories - of working on one of these units while I was at Rose-Hulman with Beta Iota Tau's founding chapter.
*NOT* all fun memories - though the machine was "only" 25 years old more-or-less at the time....
Ok, if that recoating thing works out, that'll be *incredible!!* There are *so many* old machines that could benefit! Not only spinning-drum RAM, but also (perhaps, eventually) restoring *hard drive platters!* I realize the density of the material on the HDD platters would have to be higher than that on the spinning-drum RAM, but I'd imagine that if you can solve the spinning-drum problem, it's only another few steps to solve the HDD platter problem! How exciting!!
I watched some great videos on youtube of companies testing and making disk packs from back in the day. My personal fav. was them spinning the clean metal platter around and pouring on the rust solution on the platter. Found an article saying at that time was one of the best ways to coat the platter. Reminded me of one of those art paint sets from as seen on tv. Spin the paper around and apply paint. I figured it wouldn't be to hard to make new metal platters and coat the platters. Most of the time would be testing the formula to see things about thickness after coating and how much (I forget) to store the data and read it back. A good example also is Curious Marc. He had the old HP terminals using a tape drive. After replacing the belts on the drive. He remove of swapped out a resistor? to increase the write head output. Then he was able to use newer and more common DC tapes.
Excellent explanation, must have taken a long time to comprehend.
Cool. I used to design drum memories for Univac and Control Date. The rage at the time was "flying" heads to reduce the gap between the head and the drum. We argued with early disk developers about which was best. Clearly, they won.
I used drum memory for an old real-time aircraft training simulator. The computer memory had only 1024 words of memory and the code was loaded in strips into the core memory. Real-time was controlled but the rotation speed of the drum. Reason for drum memory was the astronomical cost of core memory! Interesting times!
Bringing these old systems back to life is amazing. I sometimes wonder if any of the old National Cash Register CRAM (Card Random Access Memory) units survived the scrapyard. Cardboard cards had a magnetic stripe with mechanism to retrieve cards for read/write.
Yeah, I agree: part of the fun of making this old stuff work is when the OLD stuff _works!_
Recoating hit me at approx. 24:00, and of course Usagi himself way before me. Reason is that I saw an old film showing how IBM engineers literally “painted” their first hard disk prototype platters. The fast rotation of the platter was an integral part of creating an even coat (very messy too), but how that process can apply perpendicular to the rotation (a drum, not a disk) is of course beyond me. The chemistry was indeed very secretive according to said film, but I would think that Usagis friends have all relevant knowledge today, 60+ years later. Best of luck! 👍
Disks can be spin coated. that process give very precise deposition of material. Maybe he can test magnetic paint on his crashed hard disks first. I'm not sure how to coat a drum
I'd think rotating the drum at a say a 45 degree angle would work.
Let a combination of centrifugal force and gravity distribute the liquid from one end to the other.
Or maybe they blew air along the length of the drum, though that might create ripples.
Would be worth testing on surplus aluminum or cardboard tubes if the exact method isn't known.
Not the same film I saw, but same era…:
th-cam.com/video/8krGLyVRiRw/w-d-xo.html
I always thought drum memory was bigger, but maybe that's a nicely miniaturised version of the bigger stuff. If the recoating works on a drum, might it also work on those hard disk platters for the Centurion?
I really really want to see the recoting as soon as possible
There is a G15 at the Computer History Museum in Silicon Valley (Mountain View, CA), which belonged to Harry Huskey, who designed the CPU architecture, this machine has a gold plaque on it with his name, and was a present to him when he left Bendix. AFAIK, it was fully functional when it was put away in his barn in Santa Cruz, where it sat for a few decades before being shipped to the CHM, I think in the mid 90s? (I helped them pack it up for shipment).
edit to above. Wait, the Huskey G15 was sent to BOSTON, not Silicon Valley. I know there is one on static display at the CHM in Mountain View, but its a different one.
All the Boston CHM stuff went out West about 20-25 years ago.
Très belle machine ça fait plaisir de voir que c'est pas partie à la casse.
👍👍👍👍👍
This machine is amazing !!!... what a concept !!!!!! This drum deserves to be recoated.
These are fascinating machines. I worked in the computer Museum of the university of applied sciences in Kiel, Germany for a while. They have a couple of German Zuse computers. The Zuse Z22 is a tube computer that uses a drum as RAM too but I think it uses core memory for the registers. Unfortunately none of the computers there are in working condition and the museum was not interested in restoring them.
your enthusiasm is very cool
Ooooh! I am very interested in the chemistry and process R&D that's sure to happen.
Same. But i'm certain it's possible. Don't forget, the sort of technology and equipment that the engineers were working with in the 1950's is exactly the sort of things a hobbyist machinist has in his garage today. I'm honestly surprised there haven't been many hobbyist RDM units manufactured.
@@the123king Who's mad enough to work on seriously antique computers in a hobbyist workshop? (Our amigo, that's who.)
Agree completely with the concept, also I love when people do modern software that runs on 100% original hardware as you have shown before.
I say it's time to get a drum or recoat the one you have!
I really like this effort, and I love to have the methods back I case we lose the ability to make machines and have to recover from the end of the work. 8K ram on magnetic tape interleaving fast and slow memory genius….
When trying to describe computers to non technical people, "memory" being very short term is one of the hardest topics to get round... "memory" sounds like it should be long term. These old machines, where what we would call "memory" and "storage" getting kinda blurred together, make some sense of that "bad choice" of name.
This is why computers are easier to explain to old people.
First, give a cleaning to the drum, and coat it with a good layering of acrylic; as an Objet d'art it is its own beauty. Second, get a new cylinder of aluminum and start fresh. From the deep dive on TH-cam, the coatings (generalized) appear to be vapor deposited Nickle/Phosphorous layer and a final layer replace Phosphorous with Cobalt. Polish this as you go :)
I just gotta say that when I saw the power supply cord at the beginning of the video it really reminded me of the supply line that powered our entire mobile home.
It would be interesting to know how much power the RDM requires VS the rest of the system.
about 500watts from the motor size
“Good drum home “sounds like a good title for a song or an album
You had me at "recoat the drum"!!! That is so something I want to watch you try to do. Good luck!
With the help of a good paint/coating shop and machine shop... I doubt he's going to do it with a paint brush and sandpaper, lol.
Excellent series! I look forward to the next episode.
That Amperex 6463 is dreamy 😍. I bet it would sound amazing in an audio preamp!
From reading about the G15, the drum had a timing track that was written at manufacturing time and was never erased/re-written. If you were to recoat the drum, you would presumably need to find a way to re-write the timing track as well. All doable but certainly not trivial.
He has the advantage of modern computing equipment available to generate the required signals. Could probably program an Arduino to do it if information is available about what needs to go on the timing tracks.
That 'dual use' of the drum track for both long and short term use is ingenious. When I first saw "room for 108 29-bit words", I had assumed that was all the way around the drum. But now I understand those 108 words fit in about 7/8ths of the circumference and the other 1/8th is for the short term. But the 'short term' is being rewritten over a 'shifting' 1/8th of the drum constantly.
Two uses for the same magnetic track is really bizarre!!!
There is a fascinating book called "How to build a working digital computer" by Alcosser, Phillips and Wolk. It gives instructions on how build a computer on blocks of wood with paper clips, lamps, batteries, etc. Not joking! It has a chapter on how to build drum memory out of a can!
When you get it working and put it all back together, don't forget to put a couple desiccant bags in there! 😁
When I was in high school we had a computer called an ECP-18 whose main memory was a drum. Very basic by modern standards, but at the time it was the coolest thing in the world.
If you can re-coat the drum, what stops you modifying the process for hard disk platters?
not much, though you may need a more difficult to build vacuum chamber
The accuracy requirements were different from hard drives
Maybe the Hawk drive platters can be fixed as well.
Essentially nothing. Tolerances are going to be the biggest issue though.
@@the123king that’s kinda what I figured
With these micro-sprinkle like surfaces, I had good experience with a good IPA scrub so far. The only ones I haven't managed to get clean so far are the ones where someone used a marker to write stuff on it :(
Absolutely can not wait to see this thing going, I think it’s worth it to simulate the hard drive if there’s no other choice, especially if they want it running in the museum, just for reliability. It would be epic though if you can recoat that drum. It was made 70+ years ago, we must still have the knowledge, just wondering if the machines that did it still exist lol
At (roughly) 21:00 - Wow. I actually understood all of that. That is amazing! Looking forward to more of this computer that's three years older than me!
Almost forgot to add: PUPPY! 🐕🥰