Hacker News new | past | comments | ask | show | jobs | submit login
The search for the lost Cray supercomputer OS (gigaom.com)
144 points by bane on Jan 16, 2014 | hide | past | favorite | 42 comments

Couldn't find a link to the project in the article, so here you go:


Also, gigaom has a much better article up here:


.. with a link to an interesting walkthough of the emulator in-progress:


> ...apparently SGI destroyed Cray’s old software archives before spinning them off again in the late 90′s.

Why would anyone do this?

Maybe the Pentagon asked them to do so?

Link to the verilog source: https://code.google.com/p/cray-1x/

I'm confused: isn't the "much better" gigaom article you linked to exactly the same article submitted here?

It was originally a much worse Atlantic article, but the submission was updated.

"Seymour Cray's first computer, the Cray 1, debuted in 1976"

No, that was Cray Research's first computer. Seymour Cray was designing computers for Control Data Corporation long before that: the CDC 1604 (in 1960); the CDC 3000 series; the CDC 6600 (the first commercial supercomputer) and its successors.[1]

[1] https://en.wikipedia.org/wiki/Seymour_Cray

I used to work under a former Cray engineer who wrote the linker for UNICOS (Cray Unix). He was instructed by the architects to write a single-pass linker for $some_theoretical_reason, but he did what he felt was right and wrote a double-pass linker instead. Customers later praised the stability and usefulness of the UNICOS linker relative to its competitors at the time, so he felt validated.

From him, I learned the phrase "It's better to ask for forgiveness than permission". Still my favorite manager ever, and one of the reasons I believe mentoring is so important in our industry.

(I actually worked with a lot of former Cray engineers in the late 1990s. Brilliant, all of them. I learned a lot.)

was there more than one OS or is this UNICOS what they're trying to get running?

There was more than one OS. Supercomputer manufacturers (not just Cray) were basically hardware companies, and operating systems were an afterthought. Software was written on other systems and loaded in from the outside. UNICOS was a regular System V Unix port, to give users the sort of rich experience of a modern operating system.

No the Cray OS would have preceded UNICOS.


Also some customers developed their own operating systems.

Note that systems like this also required a separate management computer to boot the mainframe and load the OS.

We still have our Cray-1 (with serial number #001 in the Bradbury). If someone is really serious about doing novel research, I can probably pass along your name to the folks who can give you access (after they approve a proposal).

Here's a picture of it that you probably haven't seen before (in this resolution anyway): http://i.imgur.com/OHqUfmk.jpg

Great! Is it still in operation mode, or only a museum piece? I found an article that may be related (also Bradbury) and they call it "true museum piece": http://news.cnet.com/8301-17938_105-57411871-1/crave-visits-...

I read on Wikipedia that e.g. NCAR turned off their Cray-1 in 1988. Though it would be really cool if there is still a Cray-1 in an useable state.

I stood next to Cray-1, Cray X-MP and Cray-2 supercomputers in museums in Europe (London, Munich). The unique cooling system that also acts as bench was a nice idea. And the second one was submerged in an (special) oil tank.

Yep, that article described the one I'm talking about. It's behind glass in a museum showroom now (about a mile from where I'm sitting at the moment), but it's in very good shape. I wouldn't be surprised if it still works.

According to ESR: "Seymour Cray, designer of the Cray line of supercomputers, was among the greatest. He is said once to have toggled an entire operating system of his own design into a computer of his own design through its front-panel switches. In octal. Without an error. And it worked."


Have they tried to ask the Chinese scientists? I heard rumors that the Yinhe (银河, the Milky Way in Chinese) I supercomputer was kinda `compatible' with Cray I.







> [...] kinda `compatible' [...]

Probably used `metric inches', like in the Soviet Union? (See https://en.wikipedia.org/wiki/Metric_inch)

Although I have never seen either of the two, I have heard rumors that people claimed that the Yinhe-I was actually cloned Cray-I. Considering that the subtle equilibrium among north east Asian countries, the Soviet and the U.S. during the Cold War and that China was used by the U.S. as leverage to neutualize the influence of the Soviet, after Richard Nixon's visit, and before the eventual collapse of the Soviet Union, maybe the Chinese clones were even closer to the genuine ones than the Soviet Bloc computer systems.

I don't think it's finished until we can run the Cray OS in javascript in our browser.

1000 times faster than the original.

It is not only the computing advances that are amazing, those "seats" around the Cray are actually the power supply!

A Cray-1 does 80MFLOPS. 80GFLOPS is around what a current i7 CPU can do. You can't emulate a 80GFLOPS machine in javascript right now. A Cray X-MP is 800MFLOPS and a Cray 2 is 1.9GFLOPS. These are not PDP-11s we're talking about here.

Bellard's x86 js emulator, which is as good as it can get, has a performance comparable with a 486. A 486 is slower than a Cray 1. It seems that even today you can't emulate a cray in the browser at native speed, nothing can come close to that 1000 times faster idea.

Please also remember that these system had massive memory bandwidth, and supported non-sequential workloads better than current computers which rely a lot on caches. Even if you can get the emulated peak performance of Cray 1 in javascript, for real tasks that this computers was used for it would be slower.

The memory architecture is a very good point. It's fundamentally different from the x86-derived systems we all use today.

Besides, if you really want to emulate a Cray, it's not the OS that's important so much as that almighty FORTRAN compiler! Part of the fun of working with old Cray guys was listening to the ranting about how FORTRAN is the Best Language Ever and that by removing a layer of memory indirection, it's faster than even C. Of course, you're then hardcoding the living crap out of everything, but that was handwaved away as a minor price to be paid for real performance....

Was anyone interested in forth on the cray too?

No idea, but my understanding of the architecture is that it wouldn't work well.

A Cray-1 does 80MFLOPS. 80GFLOPS is around what a current i7 CPU can do. You can't emulate a 80GFLOPS machine in javascript right now.

You switched units from mega to giga there.

Ha, I thought there were cooling modules. Anyway, back in those days your supercomputer double as a bench, take that air thin tablets.

Well the power supply did need its own cooling too since it was 115,000 watts.

Here's a cutaway http://upload.wikimedia.org/wikipedia/commons/6/6e/Cray_1_IM...

Think about the fact not only an iphone has more computing power but 115 kilowatts vs 5 watts.

I know I know, and I like to think that there are millions more iPhones than Crays yet almost nothing comes out of them. jiggles

  “For these machines (Cray-1 or X-MP) you couldn’t really go into a store 
  and buy an application, like you do for a PC these days. Now, you just
  ‘install’ Word and it runs. For these machines, everything came in 
  source-code format and you needed to compile it before you could run 
  it. You use the … compiler to turn it into machine code the machine 
  could understand,” Tantos said. “That was the main way you interacted with 
  these machines. Without the compiler, you can’t feed it that.”
And at last, we have located Gentoo's antecedent!

IBM mainframes were the same way... installing the OS required a compile process.

If you're interested in experiencing the joy of setting up OS/360 first hand, it's more possible than you might think. Here are instructions for building OS/360 on the Hercules emulator:


None of the steps in the process is all that involved, but some of the mechanics are very different. 60's vintage OS/360 on an IBM mainframe is very unlike the Unix model that most people are used to these days.

Gentoo's antecedent was freebsd ports. Hence the name portage.

Indeed. I'm a Gentoo user and contributor since late 2002 :) I suspect I should've surrounded my post in <JOKE> </JOKE> tags, I wasn't actually suggesting one is anyway descendant from the other.

It is fascinating to see how far back the methodology goes though, I wasn't aware of this.

I once tried to track down the original source of NeXTStep when I was working on adding NeXT black hardware support to QEMU, but had no real idea who to ask/where to look. I know it was sold at a hefty price back in the day, but I have no idea if anyone actually purchased a copy.

You probably just need to find the right person at Apple.

A very good book about this topic:

The Supermen: The Story of Seymour Cray and the Technical Wizards Behind the Supercomputer


It's a great book about Seymour Cray (biography) that details all his work. He was one of the very best, a real hero. Sadly he died in a car accident in the nineties.

Always fond of the story that Steve Jobs used a Cray to design Apple computers while Cray used an Apple to design the Cray [1]

[1] http://c2.com/cgi/wiki?AppleCrayComputer

Yes and sadly:

Seymour Cray, the supercomputer architect, died of head and neck injuries suffered in a traffic collision in 1996. His vehicle — a Jeep Cherokee — was designed using a Cray supercomputer.

[source: "The Supermen: The Story of Seymour Cray and the Technical Wizards Behind the Supercomputer" book]

The thing I remember about Unicos was that it was fanatical about job prioritization, which processes had permissions to use processor cpu time etc. You had to jump through hoops to get things running, even for commands that would run for less than a minute on a less robust system.

I'm waiting for the JavaScript emulator, which I assume will run faster than the original Cray 1.

submitted the same article 1 day ago https://news.ycombinator.com/item?id=7058448

[I wonder about the filters - this submission has some discussion to it]

Yeah, seems like they nerfed the submission filter today. A TCL story was 2nd, for example.

Fascinating project, amazing to see what can be done with still incomplete hardware and software.

Odd that only hobbyists seem to be involved with this.

Google for "museum of computing", I find a multitude of institutions. What else would they do but something like this?

Most computing museums are run by hobbyists.

Larger museums are only interested in famous milestone computers (The Apple II, Commodore 64 etc).

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact