Restoring Y Combinator's Xerox Alto, day 3: Inside the disk drive | Hacker News Reader

Restoring Y Combinator's Xerox Alto, day 3: Inside the disk drive

righto.com

129 pointskens21 comments

Ah the RK05 (that was the DEC version of the Diablo drive). I used to have a replacement set of R/W heads for one but gave them to a friend who was restoring a couple of drives for a PDP-11 project. Fun times.

I've wondered what it would take to build a drive and its media from scratch. DEC used to do it all the time of course but given today's tools and tech. Could you build a 5MB removable platter drive? What sort of seek rate could you get on it? Clearly one of those projects that people that build their own vacuum tubes would get in to.

For the platter, you'd need a substrate with the ferromagnetic material applied – I'm not sure what process is involved here, electroplating, coating? Some equipment might be borrowed from the tape manufacturing process if that is not considered cheating.

I'd go with a spin coating, easiest to create. However coming up with a material that could be easily spin coated that would be a bit harder. There is a lot of math involved [1] :-). There was an excellent "Secret Life of Machines"[2] video about tape which made magnetic tape by sprinkling ferrous oxide (rust) over "sticky tape" (which in the US is often called 'Scotch Tape' as that is the dominant brand).

[1] http://ecee.colorado.edu/~mcleod/teaching/EandM3400/Lab%20Bo...

[2] http://www.secretlifeofmachines.com/secret_life_of_the_video...

Scotch tape is something slightly different to what I'd consider "sticky tape", which is what I'd also know as Sellotape. We have Scotch tape here as well (W. European seaboard ;) but it's a neater more refined (cloudy) tape specifically used for paper, whereas Sellotape is a more general purpose clear tape.

https://en.wikipedia.org/wiki/Spin_coating, most likely.

Ah, the joys of 1970s electronics, or "it's full of parts!" Electronics was finally good enough that you could design and build complex systems and have them work. But it took a huge number of 7400 series TTL ICs to do anything.

That disk emulator fpga would be helpful for people looking to resurrect other old computers (looking at a couple of closets full of pdp-11/70 boards in my office...) so hopefully the design will be published/open sourced. RK-05s I'm sure are not identical but probably sufficiently similar for the design to be a good starting point.

There's a MFM disk emulator project here, but I don't know if it's compatible with your drives: http://www.pdp8.net/mfm/mfm.shtml

I'd like to see a photo of the linear motor that drives the heads. They come in massive and MASSIVE sizes...

Unfortunately the drive motor is underneath the head assembly, so you can't see it without more disassembly of the drive. The servo motor moves the heads via a rack and pinion gear.

There's a (bad) photo of the motor on page 52 of the manual: http://bitsavers.trailing-edge.com/pdf/diablo/81503-03_Serie...

http://technology.niagarac.on.ca/staff/mcsele/pdp11.htm

About halfway down the page.

(Note: either the author is mistaken or the page is actually from ~30 years ago, since hard drives have not used stepper motors for a long time.)

While watching this video, the significance of the advent of SSD really strikes home.

Steve Wozniak's disk controller for the Apple II used the same kind of approach: a few simple chips on the disk controller card with all the timing and marshalling done by code running on the CPU. Doing it that way is practical if you have confidence that the CPU speed won't change.

I think Wozniak would have fit in well on the Alto team; he had the same approach of minimizing hardware as much as possible.

One unusual thing about the Alto is the code to run the disk (and other devices) was in microcode, not machine code. (Are there other computers that used microcode for programming, not just to implement the instruction set?)

The MIT Lisp Machine has a fair bit of extra stuff in microcode other than the high level instruction set, it is at a slightly higher level than the Wildflower microcode though as more was done in hardware.

There are also some recent Freescale SoCs like the MPC5200 that use a programmable DMA engine to handle copying bytes between devices and memory.

Wasn't the Alto expensive because they didn't minimize the hardware enough?

It was expensive because minimized discrete hardware (not soon enough for any kind of real integration) was still expensive. Even the successor 2901-based D-machines were expensive in absolute terms. By the time Alto-level functionality was relatively cheap to implement, the software requirements had grown well beyond. Hence PARC moving eventually to the likes of ECL for research platform development.

To expand on what pinewurst said, the Alto was built minicomputer-style out of TTL chips, so even with minimized hardware the CPU took three large boards of chips. Within a few years, microprocessors took over and Moore's law crushed computers built from discrete chips.

Here are some bonus pictures of boards from the Xerox Dorado, which used ECL chips for higher speed. First is the ALU board; the large chips are ECL versions of the 74181 ALU. Second is a detail of chips from the 10K family, as well as unusual square memory chips.

https://lh3.googleusercontent.com/--KYoSBUuOhg/V4Rxs7tLB-I/A... https://lh3.googleusercontent.com/-MWYhn7enXdA/V4RxsQ8YHRI/A...

Woz wouldn't have been welcome sans degree as he was at the time.

it wasn't uncommon back in those days, I spent a bunch of years porting Unix to various pieces of hardware in the mid to late 80s. Disk drives that required you to build an entire sector image (including headers, trailers CRCs/etc) to be DMA'd at the device when triggered by a passing sector mark were common

I found a drive just like that in a dumpster in Eugene Oregon in 1997. No other parts, and I couldn't get it to do anything interesting, so it eventually went back in the trash.

Reply on news.ycombinator.com