8 ms·

Restoring YC's Xerox Alto: how our boot disk was trashed with random data

149 points by dwaxe 10y ago

vcarl 10y ago
I love these posts. I don't properly understand the majority of what's being done, but a step by step of somebody solving a difficult problem in a space I know nothing about is so fascinating.
- KC8ZKF 10y ago
  This series, assembled and fleshed out, would make a fine book.
  - dmd 10y ago
    Or a great web site!
- monkpit 10y ago
  I love these posts too, but man is it a pain to read on mobile. Quit hijacking my input! Edit: I know I can use the readability mode, but I don't like being essentially forced to do so.
  - kens 10y ago
    Can you explain what the problem is? I recently switched the mobile formatting so Google would stop complaining that my page wasn't optimized for mobile. But maybe that made things worse.
    - monkpit 10y ago
      I left a more detailed reply (https://news.ycombinator.com/item?id=12514009 https://news.ycombinator.com/item?id=12514009) to the other person in the thread, but basically while trying to zoom I get sent to other posts or images.
  - kiwijamo 10y ago
    Looked fine on iPhone. What phone do you have?
    - monkpit 10y ago
      iPhone 6s and safari. If I zoom in on a picture and touch the image at all, when I let go the page navigates to the image as if I clicked on it. Also in my efforts to zoom in on things, the page registers a left/right swipe and decides I want to go to the next/previous blog post.
      - kens 10y ago
        I don't quite understand the problem. You say when you touch an image and let go, it navigates to the image as if you clicked on it. Isn't touching and letting go the definition of clicking?
ChuckMcM 10y ago
Hah! Nice job on inverting the random number generator. And note that RNGs for testing data patterns have much less stringent requirements than cryptographically secure RNGs :-). I'm hopeful that a known good disk will actually boot something.
- jandrese 10y ago
  Getting a good RNG on 1970s hardware is almost impossible without some kind of custom device. There is no end to examples of horrible RNGs from that era, sometimes manifesting in amusing ways. There was an implementation of Minesweeper that naively used the rand() call from C's standard library and generated utterly predictable boards. I am rather impressed at how the author of the article actually tracked down the appropriate RNG function to verify exactly what happened. This is only possible because the PARC community was so small there would only be a couple of ways something like this could have possibly happened?
  - kens 10y ago
    What made it possible to track down the RNG function was that the Computer History Museum has a big archive of Alto code [1]. By searching that site with Google, I found the random number source code. Of course figuring out the code is a bit of a challenge since it's in BCPL or Nova assembly. Hopefully as part of our Alto restoration we'll be able to add more old Alto disks to the online archive. Maybe the disk labeled "Alan Demo" will have an exciting long-lost Alan Kay demo on it :-) [1] http://xeroxalto.computerhistory.org http://xeroxalto.computerhistory.org
dboreham 10y ago
Typo: "7141 TTL chip" should be "7414 TTL chip"
- kens 10y ago
  Fixed. The spell-check didn't catch that for some reason :-)
  - dboreham 10y ago
    Where's that machine learning when you need it ;)
dalke 10y ago
I know it's not really relevant to the repair work, but I jumped back to the first article, at http://www.righto.com/2016/06/y-combinators-xerox-alto-restoring.html http://www.righto.com/2016/06/y-combinators-xerox-alto-resto... . It says > The biggest impact of the Alto was in 1979 when Steve Jobs famously toured Xerox and saw the Alto and other machines. When Jobs saw the advanced graphics of the Alto and Star, he was inspired to base the user interfaces of the Lisa and Macintosh systems on Xerox's ideas, making the GUI available to the mass market. The reference is " Malcolm Gladwell describes Steve Jobs' visit to Xerox in detail in Creation Myth." But quoting http://web.stanford.edu/dept/SUL/sites/mac/parc.html http://web.stanford.edu/dept/SUL/sites/mac/parc.html : > The closest thing in the history of computing to a Prometheus myth is the late 1979 visit to Xerox PARC by a group of Apple engineers and executives led by Steve Jobs. According to early reports, it was on this visit that Jobs discovered the mouse, windows, icons, and other technologies that had been developed at PARC. ... It's a good story. Unfortunately, it's also wrong in almost every way a story can be wrong. ... > Both the Macintosh and Lisa projects were underway before the 1979 visit. Documents in The Book of Macintosh (a collection of essays, technical specs, and brainstorms written by Jef Raskin on and others) dating from the fall of 1979-- months before the PARC visit-- show that the Macintosh was going to feature user-friendly interfaces; a screen that could handle multiple fonts (that is, bitmapped screens); graphics capabilities; and a graphical input device. ... > Finally, as several authors have pointed out, there were actually two visits by groups from Apple to Xerox PARC in 1979. Steve Jobs was on the second of the two.
- GFK_of_xmaspast 10y ago
  You can't seriously expect factual accuracy from Malcolm Gladwell.
  - protomyth 10y ago
    Dealers in Lightning would seem to contradict Gladwell's account plus the first video with the others being background. https://www.youtube.com/watch?v=ZOF-j6Nxm04 https://www.youtube.com/watch?v=ZOF-j6Nxm04 https://www.youtube.com/watch?v=J33pVRdxWbw https://www.youtube.com/watch?v=J33pVRdxWbw https://www.youtube.com/watch?v=_1rXqD6M614 https://www.youtube.com/watch?v=_1rXqD6M614
russellsprouts 10y ago
The shift-xor "random" number generator described actually has an interesting property. The algorithm looks like this: 1. If the value is 0, output the magic number. 2. If the value has only its high bit set, output 0. 3. Otherwise, shift the value to the left 4. if there was an overflow (the high bit was 1 before the shift), xor the value by a magic constant. For certain values of the magic constant, this has the property of generating every integer in the range exactly once.
- mzs 10y ago
  And with 077213 and 012345 octal that is what happens. That's why the distribution was so uniform when Ken checked.
  - userbinator 10y ago
    https://en.wikipedia.org/wiki/Linear-feedback_shift_register#Uses_in_circuit_testing https://en.wikipedia.org/wiki/Linear-feedback_shift_register... I suspect it was deliberately chosen to do this for the purposes of testing the disk; the sequence is random-looking so it exercises all the bits, but is also deterministic so that written data can be verified by reading it back and comparing without having to store a copy of all the written data somewhere else.
drbawb 10y ago
Out of curiosity: what would it take to turn that disk back into a working boot disk? I can only imagine how difficult it must be to find an Alto _with two disk drives_ these days to dub a boot disk onto your scrubbed one. Does the Alto have any sort of recovery environment? e.g: boot from the (3Mbit?) Ethernet, or boot from a floppy/tape, etc. that would aid in such a task?
- kens 10y ago
  The easiest thing would be to boot the Alto off a good disk, then swap the disks and write a good boot environment to the bad disk. (You can switch disks while the system is running.) One of the guys working on the restoration (Carl) is building a FPGA-based disk emulator, which we could use to re-write the bad disk (as well as read and archive disks). The Alto supports network boot over Ethernet; I'm working on a BeagleBone-based Ethernet emulator/gateway so that's another option. (The Alto's 3 Mbit Ethernet isn't compatible with anything modern, so I can't just plug an Ethernet cable into it.)
  - rogerbinns 10y ago
    > You can switch disks while the system is running Is that something you can do at any time, or only at the "home" screen when no apps are running?
    - kens 10y ago
      I think you can switch disks pretty much any time. The Alto doesn't swap to disk so the disk is only being used when you're loading a program or loading/saving a file. As far as how it's used, think of the Alto's disk more like a floppy drive than a hard drive. (Although since it takes a minute for the disk to spin up to speed, you don't want to switch disks too often.)
coverband 10y ago
Reading this effort is as exciting as watching a good detective show.. I'll almost be disappointed if they don't come across any other blocking issues on the way to success (as long as they find a way to overcome it, of course ;-)
Animats 10y ago
They'll have to find or build a 3Mb/s Ethernet anyway. Ask around; maybe someone still has a 3Mb/s to 10Mb/s gateway. If they can get to 10Mb on coax, they can get an off the shelf box to go to 10baseT and to the modern world.
- kens 10y ago
  The Living Computer Museum is working on a 3Mb/s Ethernet gateway. I'm also looking into making one with a BeagleBone (which is why I've been looking at its PRU microcontrollers lately).

149 points by dwaxe 10y ago

vcarl 10y ago

I love these posts. I don't properly understand the majority of what's being done, but a step by step of somebody solving a difficult problem in a space I know nothing about is so fascinating.

KC8ZKF 10y ago
This series, assembled and fleshed out, would make a fine book.
- dmd 10y ago
  Or a great web site!
monkpit 10y ago
I love these posts too, but man is it a pain to read on mobile. Quit hijacking my input! Edit: I know I can use the readability mode, but I don't like being essentially forced to do so.
- kens 10y ago
  Can you explain what the problem is? I recently switched the mobile formatting so Google would stop complaining that my page wasn't optimized for mobile. But maybe that made things worse.
  - monkpit 10y ago
    I left a more detailed reply (https://news.ycombinator.com/item?id=12514009 https://news.ycombinator.com/item?id=12514009) to the other person in the thread, but basically while trying to zoom I get sent to other posts or images.
- kiwijamo 10y ago
  Looked fine on iPhone. What phone do you have?
  - monkpit 10y ago
    iPhone 6s and safari. If I zoom in on a picture and touch the image at all, when I let go the page navigates to the image as if I clicked on it. Also in my efforts to zoom in on things, the page registers a left/right swipe and decides I want to go to the next/previous blog post.
    - kens 10y ago
      I don't quite understand the problem. You say when you touch an image and let go, it navigates to the image as if you clicked on it. Isn't touching and letting go the definition of clicking?

ChuckMcM 10y ago

Hah! Nice job on inverting the random number generator. And note that RNGs for testing data patterns have much less stringent requirements than cryptographically secure RNGs :-). I'm hopeful that a known good disk will actually boot something.

jandrese 10y ago
Getting a good RNG on 1970s hardware is almost impossible without some kind of custom device. There is no end to examples of horrible RNGs from that era, sometimes manifesting in amusing ways. There was an implementation of Minesweeper that naively used the rand() call from C's standard library and generated utterly predictable boards. I am rather impressed at how the author of the article actually tracked down the appropriate RNG function to verify exactly what happened. This is only possible because the PARC community was so small there would only be a couple of ways something like this could have possibly happened?
- kens 10y ago
  What made it possible to track down the RNG function was that the Computer History Museum has a big archive of Alto code [1]. By searching that site with Google, I found the random number source code. Of course figuring out the code is a bit of a challenge since it's in BCPL or Nova assembly. Hopefully as part of our Alto restoration we'll be able to add more old Alto disks to the online archive. Maybe the disk labeled "Alan Demo" will have an exciting long-lost Alan Kay demo on it :-) [1] http://xeroxalto.computerhistory.org http://xeroxalto.computerhistory.org

dboreham 10y ago

Typo: "7141 TTL chip" should be "7414 TTL chip"

kens 10y ago
Fixed. The spell-check didn't catch that for some reason :-)
- dboreham 10y ago
  Where's that machine learning when you need it ;)

dalke 10y ago

I know it's not really relevant to the repair work, but I jumped back to the first article, at http://www.righto.com/2016/06/y-combinators-xerox-alto-restoring.html http://www.righto.com/2016/06/y-combinators-xerox-alto-resto... . It says > The biggest impact of the Alto was in 1979 when Steve Jobs famously toured Xerox and saw the Alto and other machines. When Jobs saw the advanced graphics of the Alto and Star, he was inspired to base the user interfaces of the Lisa and Macintosh systems on Xerox's ideas, making the GUI available to the mass market. The reference is " Malcolm Gladwell describes Steve Jobs' visit to Xerox in detail in Creation Myth." But quoting http://web.stanford.edu/dept/SUL/sites/mac/parc.html http://web.stanford.edu/dept/SUL/sites/mac/parc.html : > The closest thing in the history of computing to a Prometheus myth is the late 1979 visit to Xerox PARC by a group of Apple engineers and executives led by Steve Jobs. According to early reports, it was on this visit that Jobs discovered the mouse, windows, icons, and other technologies that had been developed at PARC. ... It's a good story. Unfortunately, it's also wrong in almost every way a story can be wrong. ... > Both the Macintosh and Lisa projects were underway before the 1979 visit. Documents in The Book of Macintosh (a collection of essays, technical specs, and brainstorms written by Jef Raskin on and others) dating from the fall of 1979-- months before the PARC visit-- show that the Macintosh was going to feature user-friendly interfaces; a screen that could handle multiple fonts (that is, bitmapped screens); graphics capabilities; and a graphical input device. ... > Finally, as several authors have pointed out, there were actually two visits by groups from Apple to Xerox PARC in 1979. Steve Jobs was on the second of the two.

GFK_of_xmaspast 10y ago
You can't seriously expect factual accuracy from Malcolm Gladwell.
- protomyth 10y ago
  Dealers in Lightning would seem to contradict Gladwell's account plus the first video with the others being background. https://www.youtube.com/watch?v=ZOF-j6Nxm04 https://www.youtube.com/watch?v=ZOF-j6Nxm04 https://www.youtube.com/watch?v=J33pVRdxWbw https://www.youtube.com/watch?v=J33pVRdxWbw https://www.youtube.com/watch?v=_1rXqD6M614 https://www.youtube.com/watch?v=_1rXqD6M614

russellsprouts 10y ago

The shift-xor "random" number generator described actually has an interesting property. The algorithm looks like this: 1. If the value is 0, output the magic number. 2. If the value has only its high bit set, output 0. 3. Otherwise, shift the value to the left 4. if there was an overflow (the high bit was 1 before the shift), xor the value by a magic constant. For certain values of the magic constant, this has the property of generating every integer in the range exactly once.

mzs 10y ago
And with 077213 and 012345 octal that is what happens. That's why the distribution was so uniform when Ken checked.
- userbinator 10y ago
  https://en.wikipedia.org/wiki/Linear-feedback_shift_register#Uses_in_circuit_testing https://en.wikipedia.org/wiki/Linear-feedback_shift_register... I suspect it was deliberately chosen to do this for the purposes of testing the disk; the sequence is random-looking so it exercises all the bits, but is also deterministic so that written data can be verified by reading it back and comparing without having to store a copy of all the written data somewhere else.

drbawb 10y ago

Out of curiosity: what would it take to turn that disk back into a working boot disk? I can only imagine how difficult it must be to find an Alto _with two disk drives_ these days to dub a boot disk onto your scrubbed one. Does the Alto have any sort of recovery environment? e.g: boot from the (3Mbit?) Ethernet, or boot from a floppy/tape, etc. that would aid in such a task?

kens 10y ago
The easiest thing would be to boot the Alto off a good disk, then swap the disks and write a good boot environment to the bad disk. (You can switch disks while the system is running.) One of the guys working on the restoration (Carl) is building a FPGA-based disk emulator, which we could use to re-write the bad disk (as well as read and archive disks). The Alto supports network boot over Ethernet; I'm working on a BeagleBone-based Ethernet emulator/gateway so that's another option. (The Alto's 3 Mbit Ethernet isn't compatible with anything modern, so I can't just plug an Ethernet cable into it.)
- rogerbinns 10y ago
  > You can switch disks while the system is running Is that something you can do at any time, or only at the "home" screen when no apps are running?
  - kens 10y ago
    I think you can switch disks pretty much any time. The Alto doesn't swap to disk so the disk is only being used when you're loading a program or loading/saving a file. As far as how it's used, think of the Alto's disk more like a floppy drive than a hard drive. (Although since it takes a minute for the disk to spin up to speed, you don't want to switch disks too often.)

coverband 10y ago

Reading this effort is as exciting as watching a good detective show.. I'll almost be disappointed if they don't come across any other blocking issues on the way to success (as long as they find a way to overcome it, of course ;-)

Animats 10y ago

They'll have to find or build a 3Mb/s Ethernet anyway. Ask around; maybe someone still has a 3Mb/s to 10Mb/s gateway. If they can get to 10Mb on coax, they can get an off the shelf box to go to 10baseT and to the modern world.

kens 10y ago
The Living Computer Museum is working on a 3Mb/s Ethernet gateway. I'm also looking into making one with a BeagleBone (which is why I've been looking at its PRU microcontrollers lately).