Volume-based pricing makes sense: the industry can't give you faster and faster access and at the same time allow unlimited volume--they just don't have the hardware and network infrastructure to support it, and, yes, some people will try to stream at the maximum speed whenever they can.
The real question is what the volume pricing should look like. A 5GB limit is too low--if they charge that, they will likely lose lots of customers. Something that would make more sense to me would be:
You get 5GB of peak Internet usage (9am-9pm).
You get unlimited off-peak Internet usage (9pm-9am).
Only traffic above 128kbps counts towards the volume usage (i.e., you can listen to Internet radio 24h/day)
my computer theory professor would have deducted points if I said "NP problems cannot be solved in polynomial time"
As well he should: that statement is wrong. All we can say is that it isn't known whether NP-complete problems can be solved in (deterministic, sequential) polynomial time. Which is what I said.
Now, please explain *why* those statement are wrong for parallelism and non-determinism
Because you are not guaranteed that a problem consisting of "2^1000 computations" can be carried out by "2^1000 computers" in parallel. For example, if "encrypt" is a strong encryption algorithm that takes a day to run, how would you parallelize this:
s = "hello"
for i=1 to 2**1000 do s = encrypt(msg=s,key=i) end
print(s)
And for parallelizing an NP-complete problem by removing non-determinism, you can only parallelize as you encounter choice points--afterwards, you are still left with a problem in P (i.e., worse than constant time in general).
No, it isn't. When people talk about "solvable in polynomial time" without further qualification, they mean the deterministic case.
Look at it this way: if you have a problem that needs 2^1000 different calculation and each of those calculations lasts 1 day. Given you have 2^1000 computers, you can solve the problem in 1 day, which is *definately* polynomial (it's constant!)
That's neither an accurate statement about non-determinism nor about parallelism.
Technically, it's 'NP-hard,' meaning that there is no efficient way to calculate the necessary moves to "win," even if you know in advance the complete order of pieces, and are given all the time you need to make each move.
Even if we stick with the traditional meaning of "efficient" as "solvable in polynomial time", that is wrong: we simply don't know whether NP-hard problems can be solved in polynomial time or not.
Of course, the whole definition of "efficiency" used in the theory of NP completeness is bogus. Just because something runs in polynomial time doesn't mean it can be solved "efficiently" or even that it "scales well", and just because something is NP-hard doesn't mean that it's not solvable efficiently in most or all cases you would be interested in.
NP completeness is a cute theory, but the misleading use of the term "efficient" it has brought into vogue in some computer science circles has really done a lot of harm and caused a lot of confusion.
These philosophical arguments are tedious. Of course, BitKeeper can publish their software under whatever license they like, but that doesn't mean that we have to like it or use it or acquiesce.
What RMS is saying that if you use software with that kind of license, it's going to be bad for you and for others in the long run. It's similar to saying that SUVs are bad for the world or that smoking is bad for you--you may disagree, and you may insist on your right to do something that's legal, but there is no reason why others should stop complaining about it.
Of course, RMS may be wrong, but nobody seems to have made a convincing counter-argument. And whether the GPL is "more free" or "less free" is completely irrelevant to anything.
DNA is an example of a molecule where the position of billions of individual atoms matters. And, guess what, it's quite stable and it works very well for information storage.
Did it ever crossed the minds of people demanding windoze-like installers that the very fact questions ARE asked are because those who know what they doing might want to make CHOICES?
There is a difference between giving users choices and pestering them with unnecessary questions. I can make most choices easily using admin interfaces that are much better than the installer once the basic system is up and running. And the few choices I might occasionally have to make during the install, I can make with Control-Alt-Fn and typing at a console.
DO NOT DUMB DEBIAN DOWN!!!
That is exactly what asking a lot of questions during the install is: dumbing down Debian. People who know what they are doing don't need to be asked those questions (they already know what to do), and people who don't know what they are doing shouldn't be asked those questions.
"We're smart, we're careful, we can write code in C that doesn't have buffer overflows." Yeah, right. If MIT hackers can't do it, if Microsoft can't do it, who can?
Whether graphical or text based, Linux installers still ask too many unnecessary questions, and usually at the wrong time. RedHat or Mandrake's installers may be graphical, but they are just as annoying.
There are only very few questions that the installer really, really needs to ask the user, and for those, a text interface should be sufficient.
The most wonderful thing that would happen would be that people can finally dump that messy piece of software and move to a better toolset.
Unfortunately, Microsoft won't let it happen. The data may be "in XML", but that doesn't mean you can read it or generate it well. Instead, Microsoft will give you just enough to serve their business interests and nobody else's.
How? Office will probably stick undocumented base64 encoded binary stuff into the output, containing formatting information. You can use the document content, for example, with a database, but you can't load it into another word processor and preserve all the formatting. And in the other direction, sure, you can generate simple documents that Office will import, but you can't generate arbitrary Word documents--they will, again, have weird, undocumented tags and binary stuff.
In short: don't hold your breath. Microsoft isn't stupid.
SCSI configuration is not exactly user friendly: you need to work out the SCSI IDs and worry about termination. There are also a bunch of different SCSI versions you need to worry about. If you get things wrong, nothing may tell you about it, the drives may just fail in subtle ways.
IDE is much simpler: with cable select, you can just plug in anything anywhere and it works. Serial ATA will preserve that simplicity and improve on it.
It's not all the same. You have a limited budget: are you going to put it into additional wires or additional electronics? Since the cost of high speed electronics has dropped through the floor while wires aren't getting cheaper, serial is becoming increasingly attractive. And that's even not taking into account all its other advantages.
I'm not a Doctor Who expert, but I just saw the Key to Time series on DVD. In it, K9 does get the Doctor out of a lot of tight spots. Also, the Tardis is locked and can only be opened by the Doctor. And, the Doctor really seems to like K9 and does pet it quite a bit. And, yes, that tail does wag.
So it doesn't look like it's that horribly anti-US biased, it just looks like it's tracking a number of things that we don't usually look at in terms of press-freedom.
Well, that kind of bias is the usual case. When Europeans say that the US commits serious human rights violations, Americans don't even understand: they just don't consider executing minors or prison labor or a host of other issues "human rights violations". Similarly, when the US complains about lack of press freedom or various abuses in other countries, the US also fails to understand that those many not be considered problems in those countries.
Which would you rather have? Landings on all the major planets, together with exploratory rovers, chemical analyses, and photography, and space telescopes looking for planets around nearby stars? Or a handful of aging space cowboys spending a lot of their time cleaning toilets and keeping in shape at the Lagrange point? I know which one I would rather have.
Sure, it would be fun to go into space in person. But that's entertainment and tourism, and the best way to finance that is through private funding. It's the science, the big questions, that require government funding, and there we should concentrate on what gives the biggest payoff--and that is unmanned space flight with robotic probes.
The New Scientist is to Nature what the National Enquirer is to the New York Times. But, hey, lots of people read the National Enquirer for fun as well. Only that when people start taking it seriously that people get hurt.
Converting from one speed to the other was robbing Perlman of his crisp high end! [...]
"The hi-def screen spoils you," says Cuban. "I can't watch regular TV anymore. It just isn't worth the effort."
I can't watch regular TV anymore because the content (mostly) sucks.
These people confuse the medium with the message. A scratchy mono recording of Casals playing the Bach Suites beats pretty much all of that over-processed digital entertainment that this gear is designed for.
One of my machines is running Windows XP, and it is calling home to various Microsoft machines, frequently. Part of it is probably auto-updating, but there are apparently other things it does as well. And many major Windows programs check their own home server whether there is an update, and many of them don't take "No" for an answer.
Corporate security officers really should be concerned about this. From a security and privacy point of view, Windows XP is already out of control, and it looks like it's getting worse. Even if all those connections were harmless, it's hard to even identify a real trojan horse with all that junk going on.
Software updates and contacts to other services are much more sensible under Linux: nothing happens unless you explicitly enable it, you have the option of updating via media or mirrors, and all software updates can happen through a single server.
I don't quite get what the news is. I mean, these kinds of efforts have been around for a couple of decades, in various forms. Nor are Canadian academics particularly deprived--people in the US and Europe feel that they have to set up the same kinds of projects to get the cycles they need.
So, why is this news? Is there some new technology they are using?
It's just an idea, my NeXT had Zilla (it's version of this) years ago
- seems a shame that this hasn't caught on more widely.
And before that, other people did the same thing. And there is
at least a dozen projects worldwide that are doing this already on a wide
scale.
So come on Apple - let's see it, put it in the Darwin project and put a nice
UI on it in Mac OS X.
And what, pray tell, should that "nice UI" actually do that current software
isn't already doing?
It's a pretty good bet that Linux can easily handle that.
Actually, it turns out, it's not just a bet, it's a fact. AMD demonstrated
Opteron 4-way multiprocessing using Linux. It seems the only other
OS they have demonstrated is Windows with SMP, but only with 2-way multiprocessing.
Linus was working on increasing the SMP scalability of the Linux kernel, but I don't think that more than 4 to 8 cpus per node is available yet.
The largest amount of SMP AMD has demonstrated for Opteron is 4-way. It's a pretty good bet that Linux can easily handle that.
If you want to link 16000 cpu's together in a useful way, 2- or 4-way nodes are going to be marginally useful at best.
Quite to the contrary. A cluster of 16000 CPUs is very, very useful, without any form of SMP. In fact, some degree clustering is unavoidable for that many nodes. The question is whether SMP is useful at all in that kind of environment, and it may well not be: it actually complicates software (because you have now have to write for two levels of parallelism) and it may not help you much (because achiving the kind of necessary data locality may be hard). If you have the money, a cluster of 16000 single CPU machines is usually better than a cluster of 8000 dual CPU machines.
Anyway, my thought was that to scale to that number of cpus, they would probably need heavier nodes, and that would require them to adapt an existing SMP or NUMA architecture to fit their needs
Well, even assuming that your spurious reason was correct--what OS are you talking about? Come on, tell us!
Slashdot Mirror
since all NP-hard problems can be reformulated as another NP-hard problem
You obviously don't have the slightest clue of what you are talking about. Get yourself a good book on computational complexity.
The real question is what the volume pricing should look like. A 5GB limit is too low--if they charge that, they will likely lose lots of customers. Something that would make more sense to me would be:
As well he should: that statement is wrong. All we can say is that it isn't known whether NP-complete problems can be solved in (deterministic, sequential) polynomial time. Which is what I said.
Now, please explain *why* those statement are wrong for parallelism and non-determinism
Because you are not guaranteed that a problem consisting of "2^1000 computations" can be carried out by "2^1000 computers" in parallel. For example, if "encrypt" is a strong encryption algorithm that takes a day to run, how would you parallelize this:
s = "hello"
for i=1 to 2**1000 do s = encrypt(msg=s,key=i) end
print(s)
And for parallelizing an NP-complete problem by removing non-determinism, you can only parallelize as you encounter choice points--afterwards, you are still left with a problem in P (i.e., worse than constant time in general).
No, it isn't. When people talk about "solvable in polynomial time" without further qualification, they mean the deterministic case.
Look at it this way: if you have a problem that needs 2^1000 different calculation and each of those calculations lasts 1 day. Given you have 2^1000 computers, you can solve the problem in 1 day, which is *definately* polynomial (it's constant!)
That's neither an accurate statement about non-determinism nor about parallelism.
Even if we stick with the traditional meaning of "efficient" as "solvable in polynomial time", that is wrong: we simply don't know whether NP-hard problems can be solved in polynomial time or not.
Of course, the whole definition of "efficiency" used in the theory of NP completeness is bogus. Just because something runs in polynomial time doesn't mean it can be solved "efficiently" or even that it "scales well", and just because something is NP-hard doesn't mean that it's not solvable efficiently in most or all cases you would be interested in.
NP completeness is a cute theory, but the misleading use of the term "efficient" it has brought into vogue in some computer science circles has really done a lot of harm and caused a lot of confusion.
What RMS is saying that if you use software with that kind of license, it's going to be bad for you and for others in the long run. It's similar to saying that SUVs are bad for the world or that smoking is bad for you--you may disagree, and you may insist on your right to do something that's legal, but there is no reason why others should stop complaining about it.
Of course, RMS may be wrong, but nobody seems to have made a convincing counter-argument. And whether the GPL is "more free" or "less free" is completely irrelevant to anything.
DNA is an example of a molecule where the position of billions of individual atoms matters. And, guess what, it's quite stable and it works very well for information storage.
There is a difference between giving users choices and pestering them with unnecessary questions. I can make most choices easily using admin interfaces that are much better than the installer once the basic system is up and running. And the few choices I might occasionally have to make during the install, I can make with Control-Alt-Fn and typing at a console.
DO NOT DUMB DEBIAN DOWN!!!
That is exactly what asking a lot of questions during the install is: dumbing down Debian. People who know what they are doing don't need to be asked those questions (they already know what to do), and people who don't know what they are doing shouldn't be asked those questions.
"We're smart, we're careful, we can write code in C that doesn't have buffer overflows." Yeah, right. If MIT hackers can't do it, if Microsoft can't do it, who can?
There are only very few questions that the installer really, really needs to ask the user, and for those, a text interface should be sufficient.
Unfortunately, Microsoft won't let it happen. The data may be "in XML", but that doesn't mean you can read it or generate it well. Instead, Microsoft will give you just enough to serve their business interests and nobody else's.
How? Office will probably stick undocumented base64 encoded binary stuff into the output, containing formatting information. You can use the document content, for example, with a database, but you can't load it into another word processor and preserve all the formatting. And in the other direction, sure, you can generate simple documents that Office will import, but you can't generate arbitrary Word documents--they will, again, have weird, undocumented tags and binary stuff.
In short: don't hold your breath. Microsoft isn't stupid.
IDE is much simpler: with cable select, you can just plug in anything anywhere and it works. Serial ATA will preserve that simplicity and improve on it.
It's not all the same. You have a limited budget: are you going to put it into additional wires or additional electronics? Since the cost of high speed electronics has dropped through the floor while wires aren't getting cheaper, serial is becoming increasingly attractive. And that's even not taking into account all its other advantages.
If you want to put 10000 people into a completely inhospitable environment, the Sahara or Antarctica is a lot cheaper than L5.
I'm not a Doctor Who expert, but I just saw the Key to Time series on DVD. In it, K9 does get the Doctor out of a lot of tight spots. Also, the Tardis is locked and can only be opened by the Doctor. And, the Doctor really seems to like K9 and does pet it quite a bit. And, yes, that tail does wag.
Well, that kind of bias is the usual case. When Europeans say that the US commits serious human rights violations, Americans don't even understand: they just don't consider executing minors or prison labor or a host of other issues "human rights violations". Similarly, when the US complains about lack of press freedom or various abuses in other countries, the US also fails to understand that those many not be considered problems in those countries.
Sure, it would be fun to go into space in person. But that's entertainment and tourism, and the best way to finance that is through private funding. It's the science, the big questions, that require government funding, and there we should concentrate on what gives the biggest payoff--and that is unmanned space flight with robotic probes.
The New Scientist is to Nature what the National Enquirer is to the New York Times. But, hey, lots of people read the National Enquirer for fun as well. Only that when people start taking it seriously that people get hurt.
Is this thing bus powered/bus rechargeable, or does it require the use of a wall wart?
I can't watch regular TV anymore because the content (mostly) sucks.
These people confuse the medium with the message. A scratchy mono recording of Casals playing the Bach Suites beats pretty much all of that over-processed digital entertainment that this gear is designed for.
Corporate security officers really should be concerned about this. From a security and privacy point of view, Windows XP is already out of control, and it looks like it's getting worse. Even if all those connections were harmless, it's hard to even identify a real trojan horse with all that junk going on.
Software updates and contacts to other services are much more sensible under Linux: nothing happens unless you explicitly enable it, you have the option of updating via media or mirrors, and all software updates can happen through a single server.
So, why is this news? Is there some new technology they are using?
It's just an idea, my NeXT had Zilla (it's version of this) years ago - seems a shame that this hasn't caught on more widely.
And before that, other people did the same thing. And there is at least a dozen projects worldwide that are doing this already on a wide scale.
So come on Apple - let's see it, put it in the Darwin project and put a nice UI on it in Mac OS X.
And what, pray tell, should that "nice UI" actually do that current software isn't already doing?
It's a pretty good bet that Linux can easily handle that.
Actually, it turns out, it's not just a bet, it's a fact. AMD demonstrated Opteron 4-way multiprocessing using Linux. It seems the only other OS they have demonstrated is Windows with SMP, but only with 2-way multiprocessing.
The largest amount of SMP AMD has demonstrated for Opteron is 4-way. It's a pretty good bet that Linux can easily handle that.
If you want to link 16000 cpu's together in a useful way, 2- or 4-way nodes are going to be marginally useful at best.
Quite to the contrary. A cluster of 16000 CPUs is very, very useful, without any form of SMP. In fact, some degree clustering is unavoidable for that many nodes. The question is whether SMP is useful at all in that kind of environment, and it may well not be: it actually complicates software (because you have now have to write for two levels of parallelism) and it may not help you much (because achiving the kind of necessary data locality may be hard). If you have the money, a cluster of 16000 single CPU machines is usually better than a cluster of 8000 dual CPU machines.
Anyway, my thought was that to scale to that number of cpus, they would probably need heavier nodes, and that would require them to adapt an existing SMP or NUMA architecture to fit their needs
Well, even assuming that your spurious reason was correct--what OS are you talking about? Come on, tell us!