In many cases, these companies aren't US companies anymore; they're multinationals, who happen to be headquartered in the US. Hippies used to talk about wanting a "world without borders", and in the end, they got it. Pity it's the version thought up finance guys who can afford to live in gated communites, while the peace, security, and community stability version (doesn't contribute to next quarter's numbers, so who needs it?) remains the province of folk songs.
(and yes, I believe in the whole free-market, optimize the economic system for the greatest benefit, etc, but if outsourcing is so great, why don't we outsource upper-management rather than engineering, manufacturing, and sales/support?)
Presumably, though, the system versions of programs and system-wide configurations remain untouched elsewhere on the machine. In this scenario, you can hurt yourself, but not anyone else. To clean up a virus/worm in this environment, I delete your account, and leave the machine and all of the officially installed programs alone.
So, as long as the user doesn't rewrite their path (defined by the administrator and stored in the system registry), they may not even have a problem beyond the next reboot. Programs such as WP would still be installed system-wide by someone with admin privs, but would then run for each user as if they have exclusive access.
Ideally, this hack would have never been necessary, as the people writing the NT code-base would have had the political throw to make the hacks over in 95-land be forward compatible. In reality, this solution is easier to implement than either time-travel or scrapping the entire edifice.
Unfortunately, since the OS we're talking about is NT-based, the aphorism should read:
Those who do not understand VMS are condemned to reimplement it, poorly
This is what amazes me about these discussions: they hired Cutler, the architect of a very successful OS, that had all of the necessary security features. They updated and reimplemented his architecture for modern PC hardware. They then mangled it beyond all recognition by insisting that programs written for Win 3.1 and later Win95 run under NT/2K/XP as if they were still on single-user, no priv separation, versions, and we're still living with that behaviour today.
I tried to run my users with no privs on the last job, and always got bitten by programs such as WordPerfect, which insisted they had to run with PowerUser privs. Meanwhile, complex, computationaly demanding, graphics-heavy programs such as Spartan (visual environment for quantum chemistry), quietly installed in their own folder, didn't write to the registry, and could be moved without breaking because they didn't install anything to the system directories.
The second one is no less complex than WP, yet it behaved for non-priv'd users while popular programs with large development teams funded by reasonable-sized corporations, didn't.
Personally, I think there needs to be a local copy or version of the registry and system folders for such programs, so that they can write to it and be happy, without the user actually having manager privs. That way people with software written for 95/98/ME that they aren't ready to give up can still run it, while the administrator can screw down their machines and keep them relatively safe. This is probably better than the real solution, which would be MS deciding with Vista: Normal users will run as non-priv'd users, and have no write access to system folder or registry. Older programs expecting that ability will simply not run.
The Truly Best Answer would be someone at Redmond deciding, "hey, the next version of our OS will be Microsoft VMS!" Just put the Vista graphical environment on top of a real VMS core, remember that the default SYSTEM account should not ship with password MANAGER, and finally do it right.
I can see the adjustment of a certain Windows box: "Uhhh... I think that I'll just walk down campus, and go hide in the machine room behind the mainframes, and maybe you'll just forget that I exist...."
Actually, I'm having a hard time parsing these percentages, because every time I try to visualize the painting, all that comes up is the Gary Larson cover with her as a Cow.
Personally, I would have settled for watching them be run through the screening the Mercury astronauts went through, if the movie, "The Right Stuff" is to be believed. The running of the stairs test would just about guarantee great Nielsen numbers.
Picture one of those shows, where instead of being voted off/out/fired, etc, your teammates would vote that you had to undergo that challenge again.
Or the boot camp one, where instead of getting to quit, if you quit they sign you up for somebody's military for real.
I'll second IBM support. We bought the original xSeries 325, early enough that some of our machines were actually release candidates, rather than official hardware (IBM was back-ordered at the release, and the choice was take the betas, or wait two months; I thought it would take that long to get the cluster retooled for them anyway, so I took the betas). IBM treated them identically to the rest of the order, and when I called in an issue, including something vague along the lines of "eth0 refuses to negotiate with gigabit switch, while eth1 reliably syncs with same switch", they sent a technician out with spare parts, firmware updates, diagnostics, etc, and he stayed until that machine was working and stable. No "why don't you apply this patch, reboot a couple of times, and call us later", or , "this problem is obviously your switch; talk to that vendor first before calling us again". IBM was there when they said they would be there, and they stayed until things were right.
Actually, the hard part of having the service contract with them was that things I could do myself, such as unpacking, mounting, and intial wiring, they insisted on doing as well, just in case. They said that I had already paid for it, so I might as well let them use up their hours, rather than me dedicate time to that task. I give them credit; they ran wires through that rack in ways that I never would have tried. It was like watching an origami master practicing.
I had similar experiences with Sun (a V880 diagnosed an unstable RAM module, told me which one was faulty, and Sun had a new one to me the next morning) and HP (nothing broken, but I needed some insight into Linux on a ZX-6000; I was given more phone time and documentation than expected). Maybe I'm lazy, but there was a certain confidence in being able to tell the boss, "problem has been called in, a technician with parts is on the way".
We went the white-box route on our first compute cluster, which were then converted to desktops later. Decent machines, but the power-supplies weren't up to the 24x7 operation and tended to eventually have the fans sieze up, causing the ps to overheat. Eventually other components showed that they could have been better, and we cannibalized some machines to keep others running. They were replaced by HP and IBM boxes under 3-year, next-day, service contracts.
The advantage of calling IBM, HP, or even Dell, is not simply the service contract (though your time is worth something), or the fact that their QC is superior to wherever you're getting your parts from, but that they have real engineers, who worry about such issues as optimizing air-flow, choosing proper fan-sizes, etc. Take apart an IBM xSeries 345 some time, then try to decide if you could actually buy parts to build a machine like that, for less than just calling IBM.
White-box systems may have once made sense, ( I remember a 386/40 AMD-based system that I wrote my thesis on that was still running when I came to visit years later), but with modern components, heat-loads, etc, it pays to invest in properly engineered hardware, backed up by a company willing to service it on short notice. WB hardware may still make sense for desktops, if your environment keeps the data in non-local storage, so that a new desktop can be dropped, booted, and put into production immediately. Never with servers.
We adopted an informal, simple, but effective policy: Do not buy any machine that doesn't come with a three-year warranty, or hard-drive that doesn't come with a five year warranty.
Sun seems to have entered the, "if you strike me down, I'll become more powerful than you can imagine" mode of business. Let's hope (for those of us who ran and like Solaris) that they're right.
It would be interesting if there is enough of this technology running released for either the up and comers (China and India) or smaller firms here, to start making Sparcs, thereby encouraging it to spread as an alternate platform. Since Sun still sells support, consulting, etc, and the tools to drive it are free, it's a great way of creating mind-share.
They may be edging back towards the technical computing market as well, given that their Studio 11 includes Fortran95 for Opteron/Solaris. Nice to see that they're going to go out with a fight, rather than quietly fade away like SGI has.
My Oxford American Dictionary claims that "Orangutan" "Orangutang" and "Orangoutang" are all acceptable spellings.
The OED (compact edition), (the full OED, but you need a special magnifier to read it), says "Orangutan" or "Orangoutang", with the first being more correct.
One of the issues prior to the 2001 incident was reported to be that the FBI, CIA, etc, were more worried about turf-protection than criminal interdiction. If you actually called someone, you'd have to share the credit, possibly help his budget get increased, and therefore lose face within your own organization. So, while modernizing the FBI computer system (and while they're at it the great gantlet of forms that will be digitized) is a great idea, someone has to do something about the FBI culture as well. More easily searched forms that nobody but the FBI can access, on the FBI's schedule, aren't really going to help much.
Translation: we're spending money on a technical solution to a sociological problem.
Considering what I've spent on service contracts over the last several years, that's not a bad deal. The box only has to flake once, and it's probably paid for itself. This presumes, of course, that your downtime is worth something. I bought desktops from IBM and HP for the last job because we got a three-year service contract with them, and paid slightly more (total) for hardware only for PIV/Xeon-based machines.
The best part of a three-year contract is that the company is betting that it won't see that box again during that period, so you have some hope that it's built to a reasonable quality standard. There's nothing worse (computer-wise) than getting a supposedly great price on a piece of equipment, just to watch it act flaky (eating into your productivity) for months before ultimately dying decisively (eating even further).
I wish them the best of luck. They have good tools, and maybe they can make enough off support and hardware to keep going. I personally think they should charge some minimal amount for the bundle, as Apple does, just for psychological reasons, but if they've thought this through, then let's see how it works.
We ran some AIX servers under 3.2.5 until a year or so back. We'd replaced harddrives a few times (the last time with wide drives with narrow adapters), but finally upgraded mainly out of paranoia. They wouldn't run AIX 4, IBM wasn't patching 3.2.5, and they were too slow to use as compute servers, even for teaching. (simulation codes also get larger with age, and we wanted to keep current with what the research groups were using). Rather than wait for them to get Rooted, we turned them off and bought replacements.
All that being said, there was nothing wrong with those machines if we'd kept the same expectations as when they were new. They would have been just fine email servers, or light programming stations. However, the price of upgrading their networking to 100mb/s, replacing the monitors/video cards with ones that sync'd above 60Hz and weren't faded, and other tweaks was prohibitive versus replacing them with small Linux boxes. Sadly, those RS/6Ks would probably have outlasted the current PC's we replaced them with, even though we did buy for a reliable vendor. It's sad that there is so little left built to a spec where you upgrade it because it doesn't meet you requirements, as opposed to because it died suddenly because of a wonky power-supply or underspec'd capacitor.
I hate to say it openly, but that was the point. I'm not saying the man didn't have some "issues", but if you look at his governing, he's holding up pretty well.
Come out from under your AC, put your hand in the air, admit you bought the R.M. Nixon commemerative stamps, and loudly say, "I'm a Republican Centrist, and I'm Proud!"
Back when you could still run DOS, my boss got a bumper sticker with "I (heart) DOS" on it with a new machine. He promptly stuck it to the side of our VAX/VMS system.
10% improvement, I agree with you. 25%, we start to pay attention, and when my runtime drops by half or more, then it's worth the trouble. Regrettably, while the computer gets faster every year, the size of problem I'm expected to tackle gets larger as well, so no net win.
It would be nice if the comp-sci types would, every now and then, mention that some of those algorithms have huge prefactors, and those matter. Optimized, versus stock Netlib BLAS makes a huge difference on my codes. On the other hand, I certainly wouldn't recommend that level of detail on the whole program.
If you watch Dr. Strangelove , I believe you'll get a pretty good picture of what happens during a BSOD. You don't need the whole movie, just the scenes in the War Room as Dr. Strangelove describes the Doomsday Device.
As for what a Borg BSOD looks like, I haven't the foggiest. If we're lucky, it'll be like a Greatful Dead Concert (lots of people staring and going "oh wow"), but without the scent of Patchouli oil in the air.
Do you *do* numerical computing for a living? I've had a properly optimized math library cut my run-time in half, on 1-2 week runs. The issue becomes, when your best algorithm runs in O(N^7), then time spent optimizing your math libraries to cut the prefactors down as far as possible actually matter. You're still limited to around 10 atoms (I'm doing coupled-cluster), but now you get results back in a usable amount of time. More importantly, you get out of the way of the other users of the supercomputer.
You rely as much as possible on better compilers, faster procs, and better algorithms, but at some point you have to make sure that basic operations done billions of times are running as quickly as possible.
Amazing. I only found the NEC/Westinghouse connection from its founding, and the various foreign stock-exchange listings. My apologies. It generally amazes me how big some of the Japanese companies really are as well.
"For those with a short memory, NEC is a "baby bell" (an AT&T spin-off)."
BZZT! NEC http://www.nec.com/ is Nippon Electric Corporation, an immense japanese conglomerate founded in 1899. You're probably thinking of NCR, which was swallowed, never digested, and subsequently regurgitated by AT&T.
Your third real option is probably Fujitsu or IBM. The issue isn't just the interconnects, as you can buy those from Myricom http://www.myri.com/ or Quadrics.http://www.quadrics.com/ PNNL did this for their monstrous Itanium-2 system. It's also memory bandwidth, disk throughput, and that some jobs really require vector processors.
What Microsoft brings to the table, beyond incompatibility, overhead, and confusion, is really beyond me at the moment. They have a possibility of making an impact on high-throughput systems, but I can't see what they offer high-performance users.
I think that "not so super" product you're referring to was the Sparc-based system, which became the Starfire E10K. SGI/Cray couldn't make money on it, but Sun used it to eat their lunch.
Like the old IBM, Microsoft is now big enough that various pieces are running their own projects, and it will be interesting to see how this plays out. Windows that seamlessly clusters, where you could just add machines transparently in a manner similar to a Condor flock, would be an interesting competitor. They may be a lumbering, http://www.eps.mcgill.ca/jargon/jargon.html#evil%2 0and%20rude/ Evil and Rude corporation, but there are some really bright people in there working on more than Office.
Slashdot Mirror
In many cases, these companies aren't US companies anymore; they're multinationals, who happen to be headquartered in the US. Hippies used to talk about wanting a "world without borders", and in the end, they got it. Pity it's the version thought up finance guys who can afford to live in gated communites, while the peace, security, and community stability version (doesn't contribute to next quarter's numbers, so who needs it?) remains the province of folk songs.
(and yes, I believe in the whole free-market, optimize the economic system for the greatest benefit, etc, but if outsourcing is so great, why don't we outsource upper-management rather than engineering, manufacturing, and sales/support?)
Presumably, though, the system versions of programs and system-wide configurations remain untouched elsewhere on the machine. In this scenario, you can hurt yourself, but not anyone else. To clean up a virus/worm in this environment, I delete your account, and leave the machine and all of the officially installed programs alone.
So, as long as the user doesn't rewrite their path (defined by the administrator and stored in the system registry), they may not even have a problem beyond the next reboot. Programs such as WP would still be installed system-wide by someone with admin privs, but would then run for each user as if they have exclusive access.
Ideally, this hack would have never been necessary, as the people writing the NT code-base would have had the political throw to make the hacks over in 95-land be forward compatible. In reality, this solution is easier to implement than either time-travel or scrapping the entire edifice.
Unfortunately, since the OS we're talking about is NT-based, the aphorism should read:
Those who do not understand VMS are condemned to reimplement it, poorly
This is what amazes me about these discussions: they hired Cutler, the architect of a very successful OS, that had all of the necessary security features. They updated and reimplemented his architecture for modern PC hardware. They then mangled it beyond all recognition by insisting that programs written for Win 3.1 and later Win95 run under NT/2K/XP as if they were still on single-user, no priv separation, versions, and we're still living with that behaviour today.
I tried to run my users with no privs on the last job, and always got bitten by programs such as WordPerfect, which insisted they had to run with PowerUser privs. Meanwhile, complex, computationaly demanding, graphics-heavy programs such as Spartan (visual environment for quantum chemistry), quietly installed in their own folder, didn't write to the registry, and could be moved without breaking because they didn't install anything to the system directories.
The second one is no less complex than WP, yet it behaved for non-priv'd users while popular programs with large development teams funded by reasonable-sized corporations, didn't.
Personally, I think there needs to be a local copy or version of the registry and system folders for such programs, so that they can write to it and be happy, without the user actually having manager privs. That way people with software written for 95/98/ME that they aren't ready to give up can still run it, while the administrator can screw down their machines and keep them relatively safe. This is probably better than the real solution, which would be MS deciding with Vista: Normal users will run as non-priv'd users, and have no write access to system folder or registry. Older programs expecting that ability will simply not run.
The Truly Best Answer would be someone at Redmond deciding, "hey, the next version of our OS will be Microsoft VMS!" Just put the Vista graphical environment on top of a real VMS core, remember that the default SYSTEM account should not ship with password MANAGER, and finally do it right.
I can see the adjustment of a certain Windows box: "Uhhh... I think that I'll just walk down campus, and go hide in the machine room behind the mainframes, and maybe you'll just forget that I exist...."
Actually, I'm having a hard time parsing these percentages, because every time I try to visualize the painting, all that comes up is the Gary Larson cover with her as a Cow.
Modernity slays another classic, I guess.
Personally, I would have settled for watching them be run through the screening the Mercury astronauts went through, if the movie, "The Right Stuff" is to be believed. The running of the stairs test would just about guarantee great Nielsen numbers.
Picture one of those shows, where instead of being voted off/out/fired, etc, your teammates would vote that you had to undergo that challenge again.
Or the boot camp one, where instead of getting to quit, if you quit they sign you up for somebody's military for real.
But I digress.
I'll second IBM support. We bought the original xSeries 325, early enough that some of our machines were actually release candidates, rather than official hardware (IBM was back-ordered at the release, and the choice was take the betas, or wait two months; I thought it would take that long to get the cluster retooled for them anyway, so I took the betas). IBM treated them identically to the rest of the order, and when I called in an issue, including something vague along the lines of "eth0 refuses to negotiate with gigabit switch, while eth1 reliably syncs with same switch", they sent a technician out with spare parts, firmware updates, diagnostics, etc, and he stayed until that machine was working and stable. No "why don't you apply this patch, reboot a couple of times, and call us later", or , "this problem is obviously your switch; talk to that vendor first before calling us again". IBM was there when they said they would be there, and they stayed until things were right.
Actually, the hard part of having the service contract with them was that things I could do myself, such as unpacking, mounting, and intial wiring, they insisted on doing as well, just in case. They said that I had already paid for it, so I might as well let them use up their hours, rather than me dedicate time to that task. I give them credit; they ran wires through that rack in ways that I never would have tried. It was like watching an origami master practicing.
I had similar experiences with Sun (a V880 diagnosed an unstable RAM module, told me which one was faulty, and Sun had a new one to me the next morning) and HP (nothing broken, but I needed some insight into Linux on a ZX-6000; I was given more phone time and documentation than expected). Maybe I'm lazy, but there was a certain confidence in being able to tell the boss, "problem has been called in, a technician with parts is on the way".
We went the white-box route on our first compute cluster, which were then converted to desktops later. Decent machines, but the power-supplies weren't up to the 24x7 operation and tended to eventually have the fans sieze up, causing the ps to overheat. Eventually other components showed that they could have been better, and we cannibalized some machines to keep others running. They were replaced by HP and IBM boxes under 3-year, next-day, service contracts.
The advantage of calling IBM, HP, or even Dell, is not simply the service contract (though your time is worth something), or the fact that their QC is superior to wherever you're getting your parts from, but that they have real engineers, who worry about such issues as optimizing air-flow, choosing proper fan-sizes, etc. Take apart an IBM xSeries 345 some time, then try to decide if you could actually buy parts to build a machine like that, for less than just calling IBM.
White-box systems may have once made sense, ( I remember a 386/40 AMD-based system that I wrote my thesis on that was still running when I came to visit years later), but with modern components, heat-loads, etc, it pays to invest in properly engineered hardware, backed up by a company willing to service it on short notice. WB hardware may still make sense for desktops, if your environment keeps the data in non-local storage, so that a new desktop can be dropped, booted, and put into production immediately. Never with servers.
We adopted an informal, simple, but effective policy: Do not buy any machine that doesn't come with a three-year warranty, or hard-drive that doesn't come with a five year warranty.
Would that be the monkeys Kansas School Board members evolved from, or the Flying Monkeys that I'm trying to get H1B visas for?
..because sometimes, the right solution to the problem involves Flying Monkeys.
Sun seems to have entered the, "if you strike me down, I'll become more powerful than you can imagine" mode of business. Let's hope (for those of us who ran and like Solaris) that they're right.
It would be interesting if there is enough of this technology running released for either the up and comers (China and India) or smaller firms here, to start making Sparcs, thereby encouraging it to spread as an alternate platform. Since Sun still sells support, consulting, etc, and the tools to drive it are free, it's a great way of creating mind-share.
They may be edging back towards the technical computing market as well, given that their Studio 11 includes Fortran95 for Opteron/Solaris. Nice to see that they're going to go out with a fight, rather than quietly fade away like SGI has.
My Oxford American Dictionary claims that "Orangutan" "Orangutang" and "Orangoutang" are all acceptable spellings.
The OED (compact edition), (the full OED, but you need a special magnifier to read it), says "Orangutan" or "Orangoutang", with the first being more correct.
Dr. Zaius, we salute you.
One of the issues prior to the 2001 incident was reported to be that the FBI, CIA, etc, were more worried about turf-protection than criminal interdiction. If you actually called someone, you'd have to share the credit, possibly help his budget get increased, and therefore lose face within your own organization. So, while modernizing the FBI computer system (and while they're at it the great gantlet of forms that will be digitized) is a great idea, someone has to do something about the FBI culture as well. More easily searched forms that nobody but the FBI can access, on the FBI's schedule, aren't really going to help much.
Translation: we're spending money on a technical solution to a sociological problem.
Considering what I've spent on service contracts over the last several years, that's not a bad deal. The box only has to flake once, and it's probably paid for itself. This presumes, of course, that your downtime is worth something. I bought desktops from IBM and HP for the last job because we got a three-year service contract with them, and paid slightly more (total) for hardware only for PIV/Xeon-based machines.
The best part of a three-year contract is that the company is betting that it won't see that box again during that period, so you have some hope that it's built to a reasonable quality standard. There's nothing worse (computer-wise) than getting a supposedly great price on a piece of equipment, just to watch it act flaky (eating into your productivity) for months before ultimately dying decisively (eating even further).
I wish them the best of luck. They have good tools, and maybe they can make enough off support and hardware to keep going. I personally think they should charge some minimal amount for the bundle, as Apple does, just for psychological reasons, but if they've thought this through, then let's see how it works.
So this would make their usual OS as "Microsoft Anti-Public Virus Beta"?
We ran some AIX servers under 3.2.5 until a year or so back. We'd replaced harddrives a few times (the last time with wide drives with narrow adapters), but finally upgraded mainly out of paranoia. They wouldn't run AIX 4, IBM wasn't patching 3.2.5, and they were too slow to use as compute servers, even for teaching. (simulation codes also get larger with age, and we wanted to keep current with what the research groups were using). Rather than wait for them to get Rooted, we turned them off and bought replacements.
All that being said, there was nothing wrong with those machines if we'd kept the same expectations as when they were new. They would have been just fine email servers, or light programming stations. However, the price of upgrading their networking to 100mb/s, replacing the monitors/video cards with ones that sync'd above 60Hz and weren't faded, and other tweaks was prohibitive versus replacing them with small Linux boxes. Sadly, those RS/6Ks would probably have outlasted the current PC's we replaced them with, even though we did buy for a reliable vendor. It's sad that there is so little left built to a spec where you upgrade it because it doesn't meet you requirements, as opposed to because it died suddenly because of a wonky power-supply or underspec'd capacitor.
I hate to say it openly, but that was the point. I'm not saying the man didn't have some "issues", but if you look at his governing, he's holding up pretty well.
Come out from under your AC, put your hand in the air, admit you bought the R.M. Nixon commemerative stamps, and loudly say, "I'm a Republican Centrist, and I'm Proud!"
Makes you start missing Richard Nixon, doesn't it?
(apparently an introvert as well)
Back when you could still run DOS, my boss got a bumper sticker with "I (heart) DOS" on it with a new machine. He promptly stuck it to the side of our VAX/VMS system.
10% improvement, I agree with you. 25%, we start to pay attention, and when my runtime drops by half or more, then it's worth the trouble. Regrettably, while the computer gets faster every year, the size of problem I'm expected to tackle gets larger as well, so no net win.
It would be nice if the comp-sci types would, every now and then, mention that some of those algorithms have huge prefactors, and those matter. Optimized, versus stock Netlib BLAS makes a huge difference on my codes. On the other hand, I certainly wouldn't recommend that level of detail on the whole program.
If you watch Dr. Strangelove , I believe you'll get a pretty good picture of what happens during a BSOD. You don't need the whole movie, just the scenes in the War Room as Dr. Strangelove describes the Doomsday Device.
As for what a Borg BSOD looks like, I haven't the foggiest. If we're lucky, it'll be like a Greatful Dead Concert (lots of people staring and going "oh wow"), but without the scent of Patchouli oil in the air.
Do you *do* numerical computing for a living? I've had a properly optimized math library cut my run-time in half, on 1-2 week runs. The issue becomes, when your best algorithm runs in O(N^7), then time spent optimizing your math libraries to cut the prefactors down as far as possible actually matter. You're still limited to around 10 atoms (I'm doing coupled-cluster), but now you get results back in a usable amount of time. More importantly, you get out of the way of the other users of the supercomputer.
You rely as much as possible on better compilers, faster procs, and better algorithms, but at some point you have to make sure that basic operations done billions of times are running as quickly as possible.
Bummer dude. Better have a beer.
Amazing. I only found the NEC/Westinghouse connection from its founding, and the various foreign stock-exchange listings. My apologies. It generally amazes me how big some of the Japanese companies really are as well.
"For those with a short memory, NEC is a "baby bell" (an AT&T spin-off)."
BZZT! NEC http://www.nec.com/ is Nippon Electric Corporation, an immense japanese conglomerate founded in 1899. You're probably thinking of NCR, which was swallowed, never digested, and subsequently regurgitated by AT&T.
Your third real option is probably Fujitsu or IBM. The issue isn't just the interconnects, as you can buy those from Myricom http://www.myri.com/ or Quadrics.http://www.quadrics.com/ PNNL did this for their monstrous Itanium-2 system. It's also memory bandwidth, disk throughput, and that some jobs really require vector processors.
What Microsoft brings to the table, beyond incompatibility, overhead, and confusion, is really beyond me at the moment. They have a possibility of making an impact on high-throughput systems, but I can't see what they offer high-performance users.
I think that "not so super" product you're referring to was the Sparc-based system, which became the Starfire E10K. SGI/Cray couldn't make money on it, but Sun used it to eat their lunch.
2 0and%20rude/ Evil and Rude corporation, but there are some really bright people in there working on more than Office.
Like the old IBM, Microsoft is now big enough that various pieces are running their own projects, and it will be interesting to see how this plays out. Windows that seamlessly clusters, where you could just add machines transparently in a manner similar to a Condor flock, would be an interesting competitor. They may be a lumbering, http://www.eps.mcgill.ca/jargon/jargon.html#evil%