On the other hand, if you can't trust your suppliers to keep your confidential information confidential, should you really be shipping their products at all? What if they had told Compaq or Dell that Apple was going to announce this cube thing (instead of leaking it publicly) and Compaq announced one last Monday? Who would look like the imitator? Unfortunately, with marketing, you really *do* need to keep some things secret. Plus, it seems like the retaliation itself is just a rumor. For all we know, ATi couldn't get Apple enough parts in time and *that* was why they pulled the plug on it (I'm not saying that that's what happened).
Re:oh yes, because OS X is Manna from heaven!
on
MacOSX and X11
·
· Score: 1
I would consider myself to be pretty well educated on how to put computers together. I'm sure you'll feel free to disagree... This is basically a similar response to one I posted before about a "roll your own" vs "buy the solution" deal.
For a home user, you definetly won't save money (if you know what you're doing) by buying a Mac over a built-it-yourself box (though the iMac is close, it doesn't really expand well). However, if you're in the IT department of any corporation you will *not* be building your own machines, period. How much does someone like that make per hour? Add to that the space they occupy/benefits/electricity/etc and I'd be they are making close to $75/hour. It doesn't take that many hours to make it worth buying a machine from someone who will support it.
Now, as for Mac vs. PC, I think it's more of a religious issue of which interface you prefer. I prefer MacOS by far. For whatever reason, it just makes more sense to me. Also, when something breaks, which it inevitably does on any system, I know how to fix it easily on a Mac. Even a total system reinstall is easy since it won't overwrite extensions that applications have installed. Speaking of extensions to the OS, drivers seem to "just work" on Macs more than on Windows.
In case you're wondering, I use Unix for pretty much everything except typing random things (like papers when I was in college) and then I used MacOS. Fortunately, I can use Unix for everything I do at work (being a Unix kernel programmer helps:)
If the reporter wants to continue to be a reporter, he is not an idiot for protecting his sources. This is every bit as much about protecting his own career as it is about protecting the crackers. If you knew that a reporter would reveal his sources, would you talk to him? Probably not. And most other people wouldn't either. I think the editor of the Washington Post either went to jail or nearly did during Watergate to protect sources that Woodward and Berstein had.
As the other people have mentioned, you pay for the packaging and support. Think of it this way: a networking admin making $65,000/year (which is probably low for Silicon Valley) makes roughly $300/day not counting benefits + office space which is probably almost as much. So if buying this thing saves the guy a week of not having to taylor the setup, the price is well worth it.
I also assume that they have *tested* the stuff on the CD so you won't do an install and find out you have to upgrade 10 different aps.
Heh - whoops - forgot it was 128 bits - I was thinking 64 bits for some reason. Yeah, 2^128 is a big number:) I think the inefficiency in IPv6 is that addresses are assigned geographically. Can anyone confirm/deny that?
No, it would certainly not. The reason is that IP addresses directly affect the way the internet works - you'd have to make every device (like routers, switches, etc) with an IP stack recogize the new format. The format of all IP packets would have to be changed. This is, as you would imagine, non-trivial and it doesn't solve the namespace problem anyway (it solves a different problem which I'll talk about next) since www.whatever.com would still just point to 1.2.3.4.us or 1.2.3.4.5 or whatever. The way that the IP address is actually stored is not as a string of numbers - it gets converted into a 32 bit integer which in base 16 would work out to the number "1020304" for the above example. Hence, it's not easy to add to it.
This brings up the second problem on the internet - running out of IP numbers. There are 4 billion possible combinations of numbers in 32 bits, but for various reasons, there are a bunch that are invalid and because of the allocation system, which assumed that the internet was small, there is a lot of waste. For instance, MIT has 24 million IP addresses (a "class A" network - Apple also has one). MIT is obviously not using all of them, but nobody else can have them. So, any IP address that starts with "18.x.x.x" belongs to MIT. Carnegie Mellon University (and a bunch of others) has a class B address space with 65,536 addresses (anything in 128.2.x.x) belongs to them. Smaller organizations have "class C" (256 addresses) spaces. There was (originally) nothing in the middle. Then they came up with this thing called CIDR (classless interdomain routing) that allowed for arbitrary numbers of addresses per network (as long as they are of the size of a power of 2). That has saved the internet so far. The end solution is IP version 6 (present version is version 4 - what happened to 5???) which has a few trillion IP's so that should solve the problem nicely:) However, it's taking forever to get implemented because everyone has to change their OS to support it.
Anyway, that is why you can't just add something to the IP address.
It is definetly non-flamable. The problem with it is that if it gets hot enough (ie, a module shorts and burns a hole in itself, which can happen) it tends to molecularly break down and turn into (basically) mustard gas, which is not exactly health stuff to be breathing in your machine room... I doubt the guys mentioned in this article have enough juice to do that, but a Cray does...
The T90 was much later (1994 or '95) than the Cray 2 (1985? not too sure on that date) and it certainly uses that approach. See my post further up talking about what Crays are cooled in what method. Only the T90 and Cray 2 are cooled by total immersion, though a bunch of others circulate Flourinert through the system.
You make a couple good points but you are, however, wrong about the Crays. Crays *do* have the "don't turn the Flourinert into mustard gas" problems. I'm not sure where he got the oxygen mask bit though - there isn't one on the T90 downstairs. The concern is not that normal operation will get too hot. The concern is that a short in the board will cause the board to burn.
From what I heard from the guy who sets up the Crays here (and he should know since this building *is* Cray) the Cray 2 and the Cray T90's (the ones that are actually immersed in a bath of Flourinert) have this problem. The Cray 2 would vent the gas out a pipe in the roof and the T90 actually has a filtering system to catch it which could be where the guy got the O2 mask story from. So you're right that the CPU will not be functioning when this happens, but that doesn't mean that it's not a problem for the system as a whole which will have more than one CPU and these CPU's are hooked up to pretty nice sized power supplies...
If you're asking if Cray determined the freezing point, probably not. Cray's stuff is not supercooled because the temperature excursion the boards go on when you take the out/put them back in can be damaging. Crays are generally cooled to around 55 or 65 degrees F. That way, you can use a watercooled heat exchanger to cool the flourinert. (The heat exchanger for the Cray 2 and the T90 both look *really* cool, as a totally off topic asside:)
Cray 1 and Cray X-MP were freon cooled with a cold plate between the layers of the cicuit board. The X-MP was in the same chasis as the Cray 1. This caused hot and cold pockets to form on the board so for the next machine....
The Cray 2 was cooled by immersing it into a big vat 'o flourinert which was circulated through a heat exchanger with water to keep it at around 55 (or 65?) degrees.
Y-MP has a flourinert pocket between cicuit boards so it is cooled that way. That (apparently) solved the cooling problems they had with the Cray 1/Cray X-MP.
C90, T3D and T3E were cooled much like the Y-MP. No Cray uses water cooling since water is a conductor. They have heat exchangers to cool circulating flourinert with water.
There was a deskside version of the Y-MP and that along with J90, low end T3E's, and SV1 are all air cooled.
Highly doubtful - it would most probably run AIX or a special microkernel type OS. The problem with Linux on a supercomputer right now is that the scalability of a single kernel image is "only" around 8 processors (Windows NT is like 2 or 3 processors, so it Linux still wins there:). I don't know IBM's architecture very well, but I would assume that they have a node size of more than 8 processors. In ASCI Blue Mountain (the SGI machine) the node size is 128 processors. There are 48 (I think) nodes in the cluster.
That said, the group I'm in at SGI is working on some software (about to be release in Irix) to partition single systems up into multiple smaller systems. That software will eventually run on Linux (it actually is a combination of hardware and software) and allows multiple kernel images to run on the same physical address space and communicate with direct memory copies between them. In this way we plan to build a 512p Linux system (maybe even more) where the size of a single kernel image is only 16, 32, or 64 processors.
So basically, the only way to scale Linux well past 8 CPU's without modifying the kernel heavily (I can elaborate on exactly what that means if someone wants) is to build some sort of a cluster.
I seriously doubt the DOT will have much to say about it unless he puts it in an overweight truck... Anyway, if you read/. regularly, you have seen any number of frivolous lawsuits. The FAA *does* need to at least say "ok, you can launch" (or NASA if he goes in their airspace). If he blows himself to bits, would you like to be known as the guy who signed off on it? I certainly wouldn't. Second to that, NASA rockets, including manned rockets, have a self destruct that keeps the rocket from nailing a populated area. I don't know if this guy has such a device or not, but you can bet the FAA will be interested in having one on there - the last thing they need is it flying off course and into the path of an oncomming 747.
As for seeking forgiveness later, if you launch in the wrong spot, you won't be alive to seek forgiveness and neither will the people on the plane you hit.
That said, assuming that the design is not just totally stupid, which it sounds like it isn't, I don't see why the FAA would try to stop it as long as he can convince them he'll only kill himself.
As for having all the technical challenges worked out, I dunno. For one thing, he's going to be travelling at the speed of the Earth's rotation so, while gravity may keep him upright if the engines fire with exactly the same amount of thrust, I doubt he's going to come down in the same spot he took off in (or anywhere near it for that matter). There's also the technical challenge of making all the parts work together which is never easy.
So I certainly wouldn't wanna be in that thing. But I say let him fly - he'll either make space or the Darwin Awards:)
Ok, so this guy may be endangering lives by doing this, but a better question is if that is true, why did the CIA give the "secret" report to the New York Times???? Isn't that the whole point of keeping it classified in the first place? "Government intelligence" at its best...
Oh - two last points and I'll shut up. On the media deal, I *can't* go and buy from another OEM since MS intends to force *all* OEM's to do this. How does this help consumers? If it had always been done this way, I suppose I would accept it, but people are now expecting to get a Windows license and they are not getting it.
The other point is about your Solaris vs. Win2k - if you are rebooting Win2k and not Solaris, you're not really comparing up-times. I consider a forced reboot to be almost as bad as a crash. Worse if I'm forced to reboot on a regular basis.
I didn't say "switch to Unix". I said look at what else is out there. There's a lot out there other than Unix and Microsoft (MacOS and BeOS come to mind). In fact, *any* time you get entrenched in something, start looking for what else is out there 'cause somewhere out there is the thing that's going to replace what you are using now...
I don't, honestly, care if MS opens their API's. I was responding to the "who would want Netscape" point, though I wouldn't mind if they opened the specs on.doc so I could read 'me on Unix.
Which Outlook problem: How many #$@$$@#- VB script viruses have there been in the last few weeks???? I certainly call it a security problem to have my mail reader auto-run a script which could toast anything on my system. If anyone comes out with a mail reader that does the same thing on Unix (ie, can run a shell script) I'll consider *that* a security hole too.
Internet Explorer now, by many reports (I don't use it) is better than Netscape now. It was certainly *not* better at the point that it won. It also was only originally available on one platform. The Mac port sucked *majorly* in its first few versions (dunno where it's at now), and the Solaris port still sucks from what I here. I'll definetly concede Quicken, though - I had forgotten about it. Oracle makes a better database, but they are steadily losing market share (from what I hear from people I know in Oracle) to Access - the book isn't closed on this one yet. I guess I have to concede PhotoShop since i don't even know what PhotoDraw is.
As for the final point, I am sickened every time I see someone agree with a party line completely. I certainly don't agree with every descision SGI makes (and I even work for them) nor every descision Sun makes, or Apple. Irix does, of course, need a reboot to upgrade the kernel. However, since it never crashes, I don't exactly feel a pressing need to reinstall very often. I can say the same of my Solaris box which I've owned for 3 years now and hasn't crashed once in the entire time I've owned it (except when I did something really stupid with the hardware). Your experience with Win2k seems, by all accounts, to be the exception rather than the rule.
I will spare you (and everyone else) my thoughts on the relative scalability for large servers between Unix and WinXX, since it's really off topic. In fact, 90% of this thread including what I posted is off topic...
Who wants netscape? How about those of us who don't wanna use Windows?
Here's a few more ways they hurt the industry: they continue to do nothing to fix the problems with Outlook security so virus after virus sweeps across the country. To use analogies from the auto-industry, if your car had a "feature" where you could accelerate *really* fast but if you went a bit too fast, it blew up, and car after car blew up and people died and the company did nothing to fix it, this would be called "Gross Negligence." Microsoft calls this "inovation".
Here's another one - not distributing Windows media. So that means that if I buy a PC and choose to upgrade it, I am no longer able to use a legally licensed copy of the software that I own?!?!?!
Basically, Microsoft is so large, that *any* application that they choose to integrate and sell with the OS will win. Can you give me an example of a Microsoft ap that has lost to a competing product in the last few years? *That* is how they hurt the industry - if I have the most fantastic word processor ever written, I don't stand a chance of beating Word.
Lastly, you would do well to explore some other OS's. Your bias towards M$ products is making me sick... I use Unix only, every day (ok, so I'm a Unix kernel programmer). My machine *never* gets rebooted. By never, I really mean never. The only time it goes down is when the power goes out. I'm not saying that Unix (Irix in my case) is the best OS that can possibly be written, but it's a whole lot better than anything from Microsoft in terms of stability. When I really need productivity, I use a Mac.
> Why pay the 1k? For research? For development? Hell... MS didn't get this big if these two things were really that expensive.
I hate egrigious software licensing as much as the next guy, but R&D *is very* expensive. If you have a team of 20 people working on a product (and remember, it's not just the people writing the code - there's people doing builds, testing, managing releases, getting the stuff manufactured, etc.), and they each cost the company (including benefits/office space/etc) $100,000/year, you're looking at almost $170,000/month in development costs just for that product. That means that they need to have 170 users just to justify continued development. For various pieces of business software that are *very* specialized, that may be difficult. Microsoft got as big as they are because they were able to capture millions of users. How they were able to do *that* is the subject of this minor court case......
Actually, the original Cray was a general purpose supercomputer. The front end machine basically told it what jobs to run. The I/O was basically a seperate computer, though it was very specialized. Crays have continued to use this philosophy. The I/O used by the J90/SV1/T90/T3E is called gigaring and uses (among other things) SPARC chips running VxWorks to handle all of the I/O. Their next machine (the SV2) may actually have I/O running on the mainframe, though I think that's very up-in-the-air at this point.
Anyway, my point in all that was the the Cray's are designed for general purpose computation, even if they aren't designed to be as general as, say, database servers.
For MacOS, two weeks up is pretty good. I can get about that if I strip out the cruft on my machine. Windows 95/98 seem to crash on me daily when I use them. I think a Mac in heavy use with many aps is probably closer to this than to your two week number. NT seems much more stable - maybe a crash a month (though there are people who get a crash a day out of it - YMMV). Then we get to Unix. My SPARC (running Solaris) hasn't crashed in the last two or 3 years. The only time it goes down is when the power goes out. I've had uptimes of upwards of 180 days and this is small change for a lot of people running Unix. I'm sure we'll now get a nice list of "my uptime is larger than yours!":)
In all fairness to MacOS and Windoze, it's kinda hard to tell whether a crash is really a crash of the OS or the GUI. X has died on my SPARC multiple times, but I just kill it and keep going...
Yeah, multi-path I/O is a *very* cool thing. FribreChannel cables are also so much nicer to work with than SCSI:) These things work well in clusters with a cluster wide filesystem too to provide a single huge filesystem that can be shared across the front end nodes and compute nodes.
The interconnect on Origin 2000 (formerly "CrayLink", now "NUMAlink" with the sale of Cray...) is 800 MBytes/sec bidirectional from the node to the nearest router. The follow on will have twice the bandwidth.
It's kinda hard to describe. At 32p, the machine is a 3D cube with extra links going in an X across the cube. At 64p the machine is a 4D hypercube. At 128p, it's sort of two 4D hypercubes with 4 of their corners touching. Above 128p, I can't visulize it anymore enough to describe it:) Fortunately, my software doesn't need to know exactly how the machine is connected:)
You don't have to imagine - check out ASCI Blue - it's not a Beowulf, but it is a cluster of 48 boxes where each box is an SGI Origin 2000 with 128 processors. It is pretty high up the Top 500 list of the world's fastest supercomputers:)
Our (SGI's) ccNUMA machines go to 512p in a single system image. The next generation has a hardware limit of 1024 CPU's, though we are not, at this point, committed to building one larger than 512p for a variety of technical reasons I'm not going to go into here. However, large sums of money tossed in our direction could push the CPU count up:)
We are definetly working on a IA64/Linux version of this hardware that will scale to the same number of processors as the MIPS version, though it is unclear how far it will go as a single system image. However, we can split the thing up and run multiple kernel images that talk over shared memory in the same box.
Slashdot Mirror
On the other hand, if you can't trust your suppliers to keep your confidential information confidential, should you really be shipping their products at all? What if they had told Compaq or Dell that Apple was going to announce this cube thing (instead of leaking it publicly) and Compaq announced one last Monday? Who would look like the imitator? Unfortunately, with marketing, you really *do* need to keep some things secret. Plus, it seems like the retaliation itself is just a rumor. For all we know, ATi couldn't get Apple enough parts in time and *that* was why they pulled the plug on it (I'm not saying that that's what happened).
For a home user, you definetly won't save money (if you know what you're doing) by buying a Mac over a built-it-yourself box (though the iMac is close, it doesn't really expand well). However, if you're in the IT department of any corporation you will *not* be building your own machines, period. How much does someone like that make per hour? Add to that the space they occupy/benefits/electricity/etc and I'd be they are making close to $75/hour. It doesn't take that many hours to make it worth buying a machine from someone who will support it.
Now, as for Mac vs. PC, I think it's more of a religious issue of which interface you prefer. I prefer MacOS by far. For whatever reason, it just makes more sense to me. Also, when something breaks, which it inevitably does on any system, I know how to fix it easily on a Mac. Even a total system reinstall is easy since it won't overwrite extensions that applications have installed. Speaking of extensions to the OS, drivers seem to "just work" on Macs more than on Windows.
In case you're wondering, I use Unix for pretty much everything except typing random things (like papers when I was in college) and then I used MacOS. Fortunately, I can use Unix for everything I do at work (being a Unix kernel programmer helps :)
If the reporter wants to continue to be a reporter, he is not an idiot for protecting his sources. This is every bit as much about protecting his own career as it is about protecting the crackers. If you knew that a reporter would reveal his sources, would you talk to him? Probably not. And most other people wouldn't either. I think the editor of the Washington Post either went to jail or nearly did during Watergate to protect sources that Woodward and Berstein had.
I also assume that they have *tested* the stuff on the CD so you won't do an install and find out you have to upgrade 10 different aps.
Heh - whoops - forgot it was 128 bits - I was thinking 64 bits for some reason. Yeah, 2^128 is a big number :) I think the inefficiency in IPv6 is that addresses are assigned geographically. Can anyone confirm/deny that?
This brings up the second problem on the internet - running out of IP numbers. There are 4 billion possible combinations of numbers in 32 bits, but for various reasons, there are a bunch that are invalid and because of the allocation system, which assumed that the internet was small, there is a lot of waste. For instance, MIT has 24 million IP addresses (a "class A" network - Apple also has one). MIT is obviously not using all of them, but nobody else can have them. So, any IP address that starts with "18.x.x.x" belongs to MIT. Carnegie Mellon University (and a bunch of others) has a class B address space with 65,536 addresses (anything in 128.2.x.x) belongs to them. Smaller organizations have "class C" (256 addresses) spaces. There was (originally) nothing in the middle. Then they came up with this thing called CIDR (classless interdomain routing) that allowed for arbitrary numbers of addresses per network (as long as they are of the size of a power of 2). That has saved the internet so far. The end solution is IP version 6 (present version is version 4 - what happened to 5???) which has a few trillion IP's so that should solve the problem nicely :) However, it's taking forever to get implemented because everyone has to change their OS to support it.
Anyway, that is why you can't just add something to the IP address.
It is definetly non-flamable. The problem with it is that if it gets hot enough (ie, a module shorts and burns a hole in itself, which can happen) it tends to molecularly break down and turn into (basically) mustard gas, which is not exactly health stuff to be breathing in your machine room... I doubt the guys mentioned in this article have enough juice to do that, but a Cray does...
The T90 was much later (1994 or '95) than the Cray 2 (1985? not too sure on that date) and it certainly uses that approach. See my post further up talking about what Crays are cooled in what method. Only the T90 and Cray 2 are cooled by total immersion, though a bunch of others circulate Flourinert through the system.
From what I heard from the guy who sets up the Crays here (and he should know since this building *is* Cray) the Cray 2 and the Cray T90's (the ones that are actually immersed in a bath of Flourinert) have this problem. The Cray 2 would vent the gas out a pipe in the roof and the T90 actually has a filtering system to catch it which could be where the guy got the O2 mask story from. So you're right that the CPU will not be functioning when this happens, but that doesn't mean that it's not a problem for the system as a whole which will have more than one CPU and these CPU's are hooked up to pretty nice sized power supplies...
If you're asking if Cray determined the freezing point, probably not. Cray's stuff is not supercooled because the temperature excursion the boards go on when you take the out/put them back in can be damaging. Crays are generally cooled to around 55 or 65 degrees F. That way, you can use a watercooled heat exchanger to cool the flourinert. (The heat exchanger for the Cray 2 and the T90 both look *really* cool, as a totally off topic asside :)
The Cray 2 was cooled by immersing it into a big vat 'o flourinert which was circulated through a heat exchanger with water to keep it at around 55 (or 65?) degrees.
Y-MP has a flourinert pocket between cicuit boards so it is cooled that way. That (apparently) solved the cooling problems they had with the Cray 1/Cray X-MP.
C90, T3D and T3E were cooled much like the Y-MP. No Cray uses water cooling since water is a conductor. They have heat exchangers to cool circulating flourinert with water.
There was a deskside version of the Y-MP and that along with J90, low end T3E's, and SV1 are all air cooled.
That said, the group I'm in at SGI is working on some software (about to be release in Irix) to partition single systems up into multiple smaller systems. That software will eventually run on Linux (it actually is a combination of hardware and software) and allows multiple kernel images to run on the same physical address space and communicate with direct memory copies between them. In this way we plan to build a 512p Linux system (maybe even more) where the size of a single kernel image is only 16, 32, or 64 processors.
So basically, the only way to scale Linux well past 8 CPU's without modifying the kernel heavily (I can elaborate on exactly what that means if someone wants) is to build some sort of a cluster.
As for seeking forgiveness later, if you launch in the wrong spot, you won't be alive to seek forgiveness and neither will the people on the plane you hit.
That said, assuming that the design is not just totally stupid, which it sounds like it isn't, I don't see why the FAA would try to stop it as long as he can convince them he'll only kill himself.
As for having all the technical challenges worked out, I dunno. For one thing, he's going to be travelling at the speed of the Earth's rotation so, while gravity may keep him upright if the engines fire with exactly the same amount of thrust, I doubt he's going to come down in the same spot he took off in (or anywhere near it for that matter). There's also the technical challenge of making all the parts work together which is never easy.
So I certainly wouldn't wanna be in that thing. But I say let him fly - he'll either make space or the Darwin Awards :)
Ok, so this guy may be endangering lives by doing this, but a better question is if that is true, why did the CIA give the "secret" report to the New York Times???? Isn't that the whole point of keeping it classified in the first place? "Government intelligence" at its best...
The other point is about your Solaris vs. Win2k - if you are rebooting Win2k and not Solaris, you're not really comparing up-times. I consider a forced reboot to be almost as bad as a crash. Worse if I'm forced to reboot on a regular basis.
I didn't say "switch to Unix". I said look at what else is out there. There's a lot out there other than Unix and Microsoft (MacOS and BeOS come to mind). In fact, *any* time you get entrenched in something, start looking for what else is out there 'cause somewhere out there is the thing that's going to replace what you are using now...
Which Outlook problem: How many #$@$$@#- VB script viruses have there been in the last few weeks???? I certainly call it a security problem to have my mail reader auto-run a script which could toast anything on my system. If anyone comes out with a mail reader that does the same thing on Unix (ie, can run a shell script) I'll consider *that* a security hole too.
Internet Explorer now, by many reports (I don't use it) is better than Netscape now. It was certainly *not* better at the point that it won. It also was only originally available on one platform. The Mac port sucked *majorly* in its first few versions (dunno where it's at now), and the Solaris port still sucks from what I here. I'll definetly concede Quicken, though - I had forgotten about it. Oracle makes a better database, but they are steadily losing market share (from what I hear from people I know in Oracle) to Access - the book isn't closed on this one yet. I guess I have to concede PhotoShop since i don't even know what PhotoDraw is.
As for the final point, I am sickened every time I see someone agree with a party line completely. I certainly don't agree with every descision SGI makes (and I even work for them) nor every descision Sun makes, or Apple. Irix does, of course, need a reboot to upgrade the kernel. However, since it never crashes, I don't exactly feel a pressing need to reinstall very often. I can say the same of my Solaris box which I've owned for 3 years now and hasn't crashed once in the entire time I've owned it (except when I did something really stupid with the hardware). Your experience with Win2k seems, by all accounts, to be the exception rather than the rule.
I will spare you (and everyone else) my thoughts on the relative scalability for large servers between Unix and WinXX, since it's really off topic. In fact, 90% of this thread including what I posted is off topic...
Here's a few more ways they hurt the industry: they continue to do nothing to fix the problems with Outlook security so virus after virus sweeps across the country. To use analogies from the auto-industry, if your car had a "feature" where you could accelerate *really* fast but if you went a bit too fast, it blew up, and car after car blew up and people died and the company did nothing to fix it, this would be called "Gross Negligence." Microsoft calls this "inovation".
Here's another one - not distributing Windows media. So that means that if I buy a PC and choose to upgrade it, I am no longer able to use a legally licensed copy of the software that I own?!?!?!
Basically, Microsoft is so large, that *any* application that they choose to integrate and sell with the OS will win. Can you give me an example of a Microsoft ap that has lost to a competing product in the last few years? *That* is how they hurt the industry - if I have the most fantastic word processor ever written, I don't stand a chance of beating Word.
Lastly, you would do well to explore some other OS's. Your bias towards M$ products is making me sick... I use Unix only, every day (ok, so I'm a Unix kernel programmer). My machine *never* gets rebooted. By never, I really mean never. The only time it goes down is when the power goes out. I'm not saying that Unix (Irix in my case) is the best OS that can possibly be written, but it's a whole lot better than anything from Microsoft in terms of stability. When I really need productivity, I use a Mac.
I hate egrigious software licensing as much as the next guy, but R&D *is very* expensive. If you have a team of 20 people working on a product (and remember, it's not just the people writing the code - there's people doing builds, testing, managing releases, getting the stuff manufactured, etc.), and they each cost the company (including benefits/office space/etc) $100,000/year, you're looking at almost $170,000/month in development costs just for that product. That means that they need to have 170 users just to justify continued development. For various pieces of business software that are *very* specialized, that may be difficult. Microsoft got as big as they are because they were able to capture millions of users. How they were able to do *that* is the subject of this minor court case......
Anyway, my point in all that was the the Cray's are designed for general purpose computation, even if they aren't designed to be as general as, say, database servers.
In all fairness to MacOS and Windoze, it's kinda hard to tell whether a crash is really a crash of the OS or the GUI. X has died on my SPARC multiple times, but I just kill it and keep going...
The interconnect on Origin 2000 (formerly "CrayLink", now "NUMAlink" with the sale of Cray...) is 800 MBytes/sec bidirectional from the node to the nearest router. The follow on will have twice the bandwidth.
It's kinda hard to describe. At 32p, the machine is a 3D cube with extra links going in an X across the cube. At 64p the machine is a 4D hypercube. At 128p, it's sort of two 4D hypercubes with 4 of their corners touching. Above 128p, I can't visulize it anymore enough to describe it :) Fortunately, my software doesn't need to know exactly how the machine is connected :)
You don't have to imagine - check out ASCI Blue - it's not a Beowulf, but it is a cluster of 48 boxes where each box is an SGI Origin 2000 with 128 processors. It is pretty high up the Top 500 list of the world's fastest supercomputers :)
We are definetly working on a IA64/Linux version of this hardware that will scale to the same number of processors as the MIPS version, though it is unclear how far it will go as a single system image. However, we can split the thing up and run multiple kernel images that talk over shared memory in the same box.