- Feature Articles
- CodeSOD
- Error'd
-
Forums
-
Other Articles
- Random Article
- Alex's Soapbox
- Announcements
- Best of…
- Best of Email
- Best of the Sidebar
- Bring Your Own Code
- Coded Smorgasbord
- Mandatory Fun Day
- Off Topic
- Representative Line
- News Roundup
- Editor's Soapbox
- Software on the Rocks
- Souvenir Potpourri
- Sponsor Post
- Tales from the Interview
- The Daily WTF: Live
- Virtudyne
Admin
I actually worked at Thinking Machines Corporation (TMC) and I have a few comments. To the best of my recollection:
The communication topologies of the Connection Machines evolved. The CM-1 used a grid, the CM-2 used a hypercube, and the CM-5 used a fat tree.
We didn't, of course, provide full CM-5s for the set of Jurassic Park; we just provided a few CM-5 front covers. The lights could be programmed to blink in several different patterns. The most widely used pattern was the one produced when the lights were set to "random-and-pleasing mode". The only mode that reflected any aspect of the machine's internal operation was the mode used during boot-up. If an error condition occurred the lights could be used for diagnostics. For all modes except boot-up the lights were controlled by a single cheap dedicated microprocessor that had no connection whatsoever with the rest of the machine, so it was easy to provide Spielberg's crew with stand-alone blinking lights.
We provided compilers for parallel versions of Lisp, C, Fortran, and PARIS (Parallel Instruction Set). PARIS supported low-level parallel primitives.
The initial "vision" may have been to someday (in the very distant future) produce a "thinking machine", but our focus wasn't on AI. Our machines were used mainly for scientific computing. The NSA bought several of them too (presumably for code-breaking). I'm not sure where the author of the post came up with the notion that CMs were "pretty much useless for business and scientific purposes" and that they couldn't do floating point calculations. We used dedicated floating point coprocessors to crunched numbers.
Sheryl Handler went on to found Ab Initio Software Corporation, where she is still the CEO. (Quite an accomplishement for someone referred to as "impressively inept" in the post.) Several thunkos (former TMC employees) joined her in the founding of the company. Sheryl made sure everything we did was aesthetically pleasing and our machines and literature were works of art. Danny Hillis was in charge of the technical aspects of our products.
Yes, our office was magnificent and yes, a gourmet chef and her entourage came in every day to cook our lunch. The meals were awesome.
I had six full days of interviews and I thoroughly enjoyed them (and no, I'm not a masochist). If a single person had "black balled" me I wouldn't have been hired. The company we just looking for a certain type of person and they were very careful about who they hired. I just happened to fit right in and I enjoyed every minute of the five years I spent there.
Hey, the coke machine was cool! It was way ahead of its time. Only in the last few years have companies started to try to sell us toasters, refrigerators, and coffee makers that are hooked up to the internet.
Uh, I hardly think the downfall of TMC can be attributed to a Wall Street journal article and an embarrassed Bush administration. During the same period of time I also did research at MIT and United Technologies Research Center (UTRC) and they were also hit very hard by the end of the cold war. UTRC's staff shrunk from 1,500 to 500 in a year or two and everyone was told only to focus on projects that could make a profit in two years or less instead of working of strategic projects that looked five or ten years into the future. TMC also made some very bad business decisions.
Admin
Not only did they have no business strategy, they methodically put contraints in place to ensure that noone with a different outlook could get in. Sounds a lot like my current employer.
My company does occasionally hire developers with sense but potential management must be increasingly vetted ideologically to ensure that they whole-heartedly embrace the status quo before being being considered for promotion. This may not be a problem for management structures that have a shred of competence but before sharing a case where this might not be so bat consider some highlights of the kind of it management I am talking about:
In a now legendary and infamous decision in 2002, management decided to hold off on developing for the internet to "wait and see if it sticks around"
As late as 2005 it was decided that the best way to implement a massive web application would be with COBOL.
When I was hired in 2006 the company's entire public internet presense was a simple set of static html pages with no meaningful info.
A single act of vandalism soured this fortune 500 company on wikipedia so that the company's page is regularly removed. It is speculated to be a weekly chore of someone in upper management.
Newly hired Computer Science graduates fully vetted in object oriented design are immediately trained in COBOL where they are expected to work for their first couple of years.
A 2 week in-house training workshop in java is all that is provided to train programmers whose only experience is decades of COBOL before throwing them into large J2EE web projects. (Wanna see lots of java classes that were written as COBOL modules and then translated? I have.)
New silver bullets are vigorously being pursued with the hope of allowing unskilled employees to create web apps without any substantial training.
From the project level to the component level, new development almost always starts by copying an existing quagmire which also serves to instruct newly trained developers how to code. As a result, antipatterns accumlate and pile upon each other like geological layers.
I could go on and on with slighly more moderate WTFs I see on a daily bases but those are the real gems!
Admin
After seeing the comments, I can add some more info to the story. As usual in stories, things are changed and interpreted somewhat in the retelling.
I had a background in mathematics, business applications, and large-system software development. I understood that Thinking Machines was interested in interviewing me because I might bring some business experience and some practical insights. Clearly, I couldn't bring experience in programming massively parallel computers.
My interviews were with project managers who were working on applying the Connection Machine to various business areas. I didn't meet a manager who was applying the machine to mathematical areas. Each one described the business application that he was working on, or wanted to work on. Of course, I would ask about various parts of what they were doing. I thought this would show understanding and insight.
It seemed to me (an unexpressed thought) that massively-parallel was not a good fit for many business areas, because order processing and inventory control didn't seem to need it, given the expense. On the other hand, some types of optimization and search might fit. There was also the problem of how businesses that used COBOL (shudder)would program it.
My mind was open, and I wanted to understand how they saw this technology being applied. They were in the business, and I was new to it. Along with saying "Yes, that would be great", I asked in each case about what I saw as the possible problems in the "business plan" that came through the description. I expected to be told how the machine actually worked well for the problem area. Instead, I got the cold feeling that this didn't show enough enthusiasm.
As people have noted, I wasn't so much interested in applying cool technology as in applying a technology well and in a profitable way. The coke machine was a cool application, but to me it showed a lot of free time and a search for something interesting to do. That is not a good sign for a company.
It is unfortunate that, then and now, companies try to compensate for bad interviewing technique by assigning more interviewers. If everyone in the development staff, or the entire company (!), "likes" the candidate, then at least no one is blamed if the candidate doesn't work out so well. I think that Thinking Machines wanted enthusiasm above all, not someone who would gladly simplify a project in order to make it useful or profitable.
Admin
Admin
Admin
(click for image)
Admin
Admin
Admin
Admin
Admin
This sounds eerily like the business model of the failed car company Twentieth Century Motor Works in Atlas Shrugged.
Admin
How can you be sure they didn't?
Admin
A lot of rubbish has been talked on here about why Thinking Machines went bust.
In reality, it was because of the really badly designed GUI. For instance, the file system (which was similar to the Unix file system tree) was rendered in a sort of 3D view which the user would fly over in slow motion. This made navigating the directory hierarchy very slow and it could be a real problem in safety critical situations such as when trying to find the command to activate the door locks when the velociraptors are trying to get in.
Admin
As a person smarter than me pointed out somewhere else (and didn't post it here), at http://www.ddj.com/java/184406029 you can find a nice interview with Guy L. Steele (among other legendary achievements, one of the creators of Scheme) and how he proved a CM-1 was roughly equivalent in computing power to a Cray 1.
And the problem with Lisp is not in Lisp. It's in FORTRAN programmers. ;-)
As for the bizantine hiring process... Well. That's perhaps the only part that deserves a WTF.
Admin
+1 Funny.
Oh... Wrong website
Admin
One quibble - maybe it's just me, but when you say "Cambridge", I don't automatically complete it ", Massachusetts" - we have another one which has been around rather longer (and is also a major centre of hi-tech companies). It only caused me a moment's double-take (why was DARPA funding a UK company?), but please remember the 40% of your readers outside North America....
Admin
It was a triumph. I'm making a note here: huge success.
Admin
Admin
Wow. I can only assume 'Dave C' is a nom-de-plume and we have worked for the same boss :-). Or gawd help us, there's more than one boss that fits the exact description!
Admin
except it was serialised after all...
Admin
Admin
Ah, the mind fades... It's worse than that. The front end must have been the Sun 4/490 that was placed next to the DEC workstations that were running Ultrix. The Dec workstations were just what the engineers used when they were in the data center. They probably had similar workstations in their offices upstairs.
Like I said in my original post, I was a system operator. I didn't do much with the CM-5 system except run backups. The TMC engineers did everything else. I had more responsibility for the multiple Cray and SGI systems, since the onsite Cray engineers were limited to hardware repairs. We maintained the system and user environments. It was a very fun job. Best datacenter experience available in my opinion. I do information security system integration now.
Admin
One 18.25 ounce package chocolate cake mix. One can prepared coconut pecan frosting. 3/4 cup vegetable oil. Four large eggs. One cup semi-sweet chocolate chips. 3/4 cups butter or margarine. 1 2/3 cups granulated sugar. Two cups all purpose flour.
Don't forget garnishes such as: Fish shaped crackers. Fish shaped candies. Fish shaped solid waste. Fish shaped dirt. Fish shaped ethyl benzene. Pull and peel licorice. Fish shaped organic compounds and sediment shaped sediment. Candy coated peanut butter pieces. Shaped like fish.
One cup lemon juice. Alpha resins. Unsaturated polyester resin. Fiberglass surface resins. And volatile malted milk impoundments.
Nine large egg yolks. Twelve medium geosynthetic membranes. One cup granulated sugar.
An entry called 'how to kill someone with your bare hands.'
Two cups rhubarb, sliced. 2/3 cups granulated rhubarb. One tablespoon all-purpose rhubarb. One teaspoon grated orange rhubarb. Three tablespoons rhubarb, on fire. One cross bore hole electro-magnetic imaging rhubarb. Two tablespoons rhubarb juice.
Adjustable aluminum head positioner. Slaughter electric injector. Cordless electric needle injector. Injector needle driver. Injector needle gun. Cranial caps.
And it contains proven preservatives, deep penetration agents, and gas and odor control chemicals. That will deodorize and preserve putrid tissue.
Admin
"The NSA bought several of them too (presumably for code-breaking). "
That, and the snooper farm at Fort Meade. It takes a lot of power to listen for keywords in every single telephone conversation happening in America!
There were several other special applications for massively-parallel computers. No, I'm not going to tell you what.
Admin
Admin
OMG, a computer cannot do floating point math. . . wait a tick. . . my $5000+ desktop can't do it either. In fact no computer really can. They all estimate floating point numbers. So, what's the point?
Admin
You know, there's a line in Jurassic Park (the movie, not the novel) that I never understood. When they're taking the tour on how the dinosaurs are created, the little DNA animated character says "...Thinking Machines supercomputers break down the strands in minutes..." I capitalized that now because I now understand that it's the name of a company/manufacturer. Until this point, though, that line seemed incredibly odd to me.
Thanks Daily WTF, for helping me understand Jurassic Park even better. Heh.
captcha: vulputate
Admin
I was going to say, I've programmed on those machines many moons ago and could have sworn it was FORTRAN and C..
Admin
People who would rather treat code as data than write code by hand?
Admin
I wonder at all the disparaging remarks about the Connection Machine. It's pretty obvious that a massively parallel MIMD architecture is not for everyday business computing. Who would suppose otherwise? But it might, unsurprisingly, prove to be very useful for massively parallel computations, especially if the problem domain had topological features in common with the arrangement of processor interconnects. Computer vision, simulation and modelling, and search come immediately to mind.
It doesn't take much imagination to consider that there might possibly be interesting business applications to some of these problem domains as well, perhaps enough to justify hiring someone with a background in business applications development to work with the CM, especially if the candidate might also have some experience and interest with parallel systems at the small scale that would have been typically available to university undergrads of the day.
The mid-1980s was a time when there was a lot of interest in alternative processor architectures. The CM was just one example among many ideas that were being explored. Distributed computing was a very active area of research, board-level processors with fast interconnects were readily available, and funding for all this type of work was excellent. There was significant cross-pollination with AI research as well, giving rise to several varieties of Lisp Machine for example (Boeing Aircraft had hundreds of these at one time), excellent graphical user interfaces for data visualization, and novel languages such as Prolog and Occam, not to mention neural nets.
You could argue that these were solutions in search of a problem. They also anticipated some of the architectural pressures that are now beginning to emerge in earnest as we reach the speed limit for single processor cores. Those pressures have created a growing demand for multicore CPUs, supercomputer clusters, and grid computing, not to mention the Internet as a kind of distributed computing engine in its own right.
But all these solutions are imperfect in various respects. They will eventually be discarded or revised beyond recognition. Why then should the CM come in for such criticism? It was a legitimate exploration of some very timely questions about how to make computation scale to interesting problem domains.
Admin
coach sneakers
Admin
Great company. I worked for them for nearly 4 years and left minutes before the troubles began. In the Eighties, I worked for Sun, Meiko and TMC. TMC was loads of fun as well as being very profitable (for me:-)
Someone above said; "I guess that depends on how badly you want to work on one of the coolest pieces of computing hardware in the world at that time."
Hell, that's right! It was the coolest machine on the planet with the coolest applications and the coolest corporate culture by miles. The people I met and worked with there were awesome.
Innovative too. Coke machine toys were trivia. I remember using the www in late 1989 with some software the guy in the next room wrote (and gave to the community) called Lynx.
I was trying to find ways to present the CM5 (we called it CMX in development) and what applications became feasible with a teraflops machine.
My take on the fall is that the operational boss (not Sheryl Handler) truly believed that CM200 (not CM2) sales would stop the minute the CMX was announced.
A few of us tried to present a survival plan based on the proven success of the CM200 (which was the computer in Jurassic Park, not the CM5 as someone said above). He didn't believe it.
At that point I became persona non grata</> so I left before the axe fell. Just in time as it turned out.
Best wishes to all those I worked with if they ever read this.
I have a mint condition copy of the cookbook in front of me now! I wonder if it is worth anything on E-bay?
Admin
"The real WTF is that they make their developers pay for SODA out of pocket."
Actually, you just telnetted to the soda machine and pulled 35 cents from your account. No pocket involved!
You could also remotely check the stocking levels of the different drinks, so you didn't have to walk to the machine to find out that it didn't have what you wanted.
Same for lunch. Which was AWESOME! I could go for a week without any cash, just a T card and a TMC login.
Admin
I've got my TMC cookbook. Amazing food.
Admin
My favorite application of the CM-5 was at American Express, where they loaded up all AmEx transactions for the last 'x' years into RAM, then ran all sorts of interesting AI algorithms on them. The CM-5 outperformed a building full of IBM mainframes (NB: IBM was not pleased by this, thus the IBM SP parallel computing clusters), allowing them to do "big data" analysis in seconds instead of hours. The ROI on the investment was amazing - supposedly they paid for a very expensive supercomputer in a matter of months by using it to optimize the value of ad inserts vs. mailing costs in the monthly bills.
Admin
Even after so many years, I find it odd that people still comment on their time at TMC. I too worked at TMC, for 6 years in support. It was one of the greatest jobs I have ever had. Although many people have commented about the well know foibles of TMC, like the gourmet kitchen and coke machine, the memories I have were born out hard work, late hours, and being in such a creative, consuming, and intense environment. I survived the layoff, because I was in support, but my remaining 6 months, were sad and depressing. Very few Thinkos were around after the massive layoff on black friday. I came back in and walked around the silent building. Personal effects, half filled mugs, coats on the back of doors, pictures, and mementos were all over the place, just no people. It was very eerie and sad. When they started to organize what was left of the company, hundreds of staplers were put on one white folding table(You do remember the folding white tables, right??). Another table held tape dispensors, and so forth. It was sad that you knew that everything was in use 1 week before. Anyway, I bought a few mementos, 4 white tables and one of the "air craft carrier" desks, which I still use to this day. I also have ~6 TMC mugs, t-shirts, my TMC black bag. I ended up with lots of TMC vcr tapes of training tapes, demo videos created on the CMs and recordings of internal events, that I have not had the heart to throw out. I think about my life at TMC often.
Admin
this reminded me of a side-story in the book, "@large", about a "super" computer company, whose system was supposed to be able to run programs faster because it had multiple CPUs. BUT, the programs had to be specially written in "module" style so that the multiple CPUs could run the different subroutines simultaneously, otherwise it made no difference! since most people thought that was too much trouble, they made few sales and eventually went bankrupt.