I would have to at least partially disagree with this. Remember that if you decide to do something dumb and OD on some drug you prescribed for yourself, your insurance company is going to have to pay for your resulting (possibly quite large) medical bills. That results in high insurance costs for me. So when you do that, you don't just affect yourself, you affect everyone else - the 911 response staff who have to come pick you up instead of someone who may be having a heart attack, the doctors in the emergency room, your insurance company, etc. Unless, of course, you want people to pay for their own mistakes which brings up a whole other set of moral/ethical issues that will only result in different regulation.
Can a doctor make a mistake that ends up with the same result as your mistake? Sure. They are human. But I'd wager that a doctor is going to make a lot fewer mistakes prescribing drugs for me than I would if I did it myself.
Well, that is certainly true for anything that is completely distributed by computer, such as, say, the silly attempts at regulating encryption. However, at some point, said off-shore pharmacy is going to have to ship you something which will go through customs. Hence, it is enforceable.
To me, this doesn't look like it's an attempt to tax and regulate the internet, it looks like an attempt to make sure that that www.we-sell-drugs.com plays by the same rules (or at least similar rules) to the neighborhood drug store. What is wrong with that?
Well, that all depends on what you call "scales well". Compared to NT, Linux scaling is fantastic (up to 8 processors or so). Compared to big iron OS's like Irix (up to 512 procesors) or Unicos/mk (up to 2048 processors), it doesn't (I should probably include OS/390 in here as a "big iron OS", but I know nothing about it). Unicos/mk is a very hard comparison to make since it is a microkernel based OS, but you can pretty easily compare Irix and Linux - Irix has a very well threaded kernel that allows you to declare certain CPU's (generally the ones close to the I/O) to handle interupts while the others keep processing. There are very few large grained locks in the kernel. Kernel threads are premptible and schedulable in Irix (which allows you to run real time aps, in addition to making the kernel more responsive to high priority tasks).
Linux, of course, is getting much better very quickly (and one of the projects that I am on the periphery of at SGI is working on this). The zone memory allocator and underpinnings of NUMA support are excellent first steps. Even so, I don't think you'll see a 512 processor single system image Linux machine that has reasonable scaling any time soon. Basically, Linux scales well on the "low end" but not the high end.
(Ok - so this really isn't related to the issue of Linux porting, but....)
With all due respect to Linus, I wouldn't say that microkernel research has been a total waste. While it hasn't turned into what the researchers probably intended (replacing regular monolithic kernels in all OS's), it has produced useful OS's. The T3D operating system (forget what it is called) and Unicos/mk on the T3E are both microkernel based. I've heard that ASCI/Red (the Intel monster machine - check the Top500 list) runs a microkernel based OS. So, some of the fastest machines in the world run microkernel based OS's. Also, MacOS X is based on the Mach microkernel.
The reason a microkernel based OS is easier to port is that there's less there to port. Linux and the *BSD OS's have, however, become a marvel at how easy they are to port.
...but why are people so concerned about this? Every workstation I've used has a built in ethernet card with a hard (or impossible) to change MAC address. Any software that wants can read it and embed it any way they like. Intel's CPU ID is just another thing like this. If I want to be sure that something I write isn't traced back to me, I'd hardly do it using a Microsoft piece of software anyway. If I fire up vi and write something, I seriously doubt any information that I didn't put there is getting appended.
Also, you can only trace the processor if you know who it is sold to. All guns sold in the US have serial numbers (well, most of them at least). Can the police trace every single one of them to the person who owns it? No. So just don't send in your registration card and it's going to be pretty hard to trace the number back to you.
Besides, you should be yelling at Microsoft, not Intel - they are the ones who are using the number in a manner that is silly. If you are running on the machine instead of WIndows and generating documents in Emacs or vi, no serial numbers are getting attached. Hence, it is a software issue - blame the software people.
(Ok, I'm posting late, so nobody will probably read this anyway, but....)
The T3D has been end of life'd - get a T3E. Also, how many processors were in that T3D? I seriously doubt you're going to compete with a T3E, however. It's hard to argue with 2048 DEC Alphas at 600 MHz...
Also, let's talk about memory bandwidth. I believe that the present T3E model gets 45 gigabits/second *BETWEEN* processors. I'd like to see you get that to local memory in your Beowulf cluster. Does that make a difference? Yes. In raw CPU speed, your Beowulf cluster may win. But that doesn't matter if your CPU's are idle half the time waiting for the RAM. Some of our (Cray's - well, technically I work for SGI but I wish I worked on the Cray side of the split - at least I still have access to their machine room:) customers will buy Beowulf. The biggest ones won't. The reason is that poor memory bandwidth could double your run time. While that makes hardly any difference for a 5 minute benchmark, for a 2 week MPI job, that could be a bit of a letdown if it now takes a month.
The other thing is NUMA - a Beowulf cluster is not a NUMA environment - you can't DMA from one node into another node without kernel intervention. On the T3E (and the SMP based vector Crays) you can do this. On the Origin (SGI designed) ccNUMA boxes, you can actually allocate several hundred gigabytes of RAM into a pthreads job and access it all normally. You won't be doing this on Beowulf any time soon.
Now, most of that doesn't make a difference for lower end customers. However, Cray has never targeted the low end of anything:) Check out the Top 500. There are a lot of Crays. There aren't too many Beowulf clusters.
Slashdot Mirror
Can a doctor make a mistake that ends up with the same result as your mistake? Sure. They are human. But I'd wager that a doctor is going to make a lot fewer mistakes prescribing drugs for me than I would if I did it myself.
To me, this doesn't look like it's an attempt to tax and regulate the internet, it looks like an attempt to make sure that that www.we-sell-drugs.com plays by the same rules (or at least similar rules) to the neighborhood drug store. What is wrong with that?
Linux, of course, is getting much better very quickly (and one of the projects that I am on the periphery of at SGI is working on this). The zone memory allocator and underpinnings of NUMA support are excellent first steps. Even so, I don't think you'll see a 512 processor single system image Linux machine that has reasonable scaling any time soon. Basically, Linux scales well on the "low end" but not the high end.
I speak for myself, not SGI.
With all due respect to Linus, I wouldn't say that microkernel research has been a total waste. While it hasn't turned into what the researchers probably intended (replacing regular monolithic kernels in all OS's), it has produced useful OS's. The T3D operating system (forget what it is called) and Unicos/mk on the T3E are both microkernel based. I've heard that ASCI/Red (the Intel monster machine - check the Top500 list) runs a microkernel based OS. So, some of the fastest machines in the world run microkernel based OS's. Also, MacOS X is based on the Mach microkernel.
The reason a microkernel based OS is easier to port is that there's less there to port. Linux and the *BSD OS's have, however, become a marvel at how easy they are to port.
Also, you can only trace the processor if you know who it is sold to. All guns sold in the US have serial numbers (well, most of them at least). Can the police trace every single one of them to the person who owns it? No. So just don't send in your registration card and it's going to be pretty hard to trace the number back to you.
Besides, you should be yelling at Microsoft, not Intel - they are the ones who are using the number in a manner that is silly. If you are running on the machine instead of WIndows and generating documents in Emacs or vi, no serial numbers are getting attached. Hence, it is a software issue - blame the software people.
The T3D has been end of life'd - get a T3E. Also, how many processors were in that T3D? I seriously doubt you're going to compete with a T3E, however. It's hard to argue with 2048 DEC Alphas at 600 MHz...
Also, let's talk about memory bandwidth. I believe that the present T3E model gets 45 gigabits/second *BETWEEN* processors. I'd like to see you get that to local memory in your Beowulf cluster. Does that make a difference? Yes. In raw CPU speed, your Beowulf cluster may win. But that doesn't matter if your CPU's are idle half the time waiting for the RAM. Some of our (Cray's - well, technically I work for SGI but I wish I worked on the Cray side of the split - at least I still have access to their machine room :) customers will buy Beowulf. The biggest ones won't. The reason is that poor memory bandwidth could double your run time. While that makes hardly any difference for a 5 minute benchmark, for a 2 week MPI job, that could be a bit of a letdown if it now takes a month.
The other thing is NUMA - a Beowulf cluster is not a NUMA environment - you can't DMA from one node into another node without kernel intervention. On the T3E (and the SMP based vector Crays) you can do this. On the Origin (SGI designed) ccNUMA boxes, you can actually allocate several hundred gigabytes of RAM into a pthreads job and access it all normally. You won't be doing this on Beowulf any time soon.
Now, most of that doesn't make a difference for lower end customers. However, Cray has never targeted the low end of anything :) Check out the Top 500. There are a lot of Crays. There aren't too many Beowulf clusters.