As the hypersonic craft demonstrated so well, there is a vast gulf between what theory says could be done and what can actually be done in practice. The only way to know is to build.
Secondly, and more importantly for NASA, something like a BWB airliner could perform high altitude research currently impossible because no conventional aircraft has the lift capacity or the space availability. NASA use aircraft extensively in research, but are usually stuck with aircraft wholly unsuited to what they want to do.
Finally, and most importantly of all, NASA Langley IS a design house for aviation. LaRC is also where NASA began and what a key part of NASA is.
Obviously, in the case of New Horizons, you couldn't reproduce the result except by building a new probe. Likewise, reproducing the results of ITER or LHC would be difficult. There, you'd be utterly reliant on analyzing the steps up to the experiment.
I absolutely concur on publication of software (and anything else used in analysis, such as custom hardware designs). The complete setup that permitted the experiment to take place, plus the setup that permitted analysis of the results, should be described in a way that can be reproduced or - at the very least - be analyzed systematically to determine if it could do what was claimed.
I agree that papers should not be treated as currency and that "publish or perish" should itself perish, but I refer you to the experiences of Open Source and of Open Science -- "release early, release often" tends to produce better results overall. The challenge is how to make use of this.
The Open Source experience shows that multi-stage code reviews work better than single-stage, that testing is critical (even though we all hate doing it), that good documentation will always outperform bad documentation, that a central repository scores over fragmented repositories, and that finding stuff is one of the most important tasks a programmer has to perform.
Seems simple enough. We know this model works, we know that the bits of this model that have been back-ported to academia have worked exceedingly well, so it would seem obvious enough to try back-porting the entire model and using that as a starting point for discussing how to fix science. Turns out that many of the key components already exist in some form in academia, so we know those components are doable with no extra effort.
It may be that my model is too complex and that your suggestion is all that is needed, but I'd rather have a meaningful discussion on it than the trial-by-combat that I'm seeing from some of the other posters.
Research costs are always high. However, in principle, a basic waverider with a hydrogen-powered scramjet aught to be a lot cheaper than an ICBM (hydrogen being a much cheaper fuel, and 100% of the vehicle is reusable).
Note - that's in principle. Practice rarely pays much attention to theory. A space elevator would be nice, too, as would a pony.
The biggest problem is that the hypersonic craft was NASA's last shot at this for a while. The last time the test craft was covered, it was stated that the project is terminated. Kaput. (In short, in about the same shape as NASA's blended wing body passenger aircraft, whose sheer size would have made the Airbus 400 look like a unicycle.) NASA hasn't the funds to see projects to completion any more. That wouldn't be a total loss, if the data collected was made available. But, no, this sort of stuff is kept under very heavy wraps. Nothing learned will be made available to anyone, the research will get redone by others at massive expense (and possible loss of life), and all to stop those nations who could never build such a vehicle anyway from scoring a PR coup. ie: it's politics. It's not even military, it's politics.
Ok, correction, then. The biggest problem is that we have politicians.
In theory, yes. The type of design (a "waverider") places the hypersonic shockwave directly beneath the vehicle. Basically, you're surfing the shockwave. This reduces the stresses involved, improves stability and should allow considerably more control than could be achieved with the space shuttle (you have sufficient lift from a waverider to glide). Waveriders do have disadvantages - most designs only work at specific speeds, the wings have a habit of frying and they rely on cooling by radiation (only effective at high altitude).
NB: The STAAR group beat NASA and the US DoD to the first working waverider airfoil, as noted on their site. Perhaps NASA's problem with their current design is that they're not threatening the engineers with bagpipe music.
Just as important, there are multiple distributions. Just as it makes it hard to write commercial software that will run under any version of any distro, it makes it hard to write a virus that will work under any version of any distro. The odds are that Linux viruses will be predominantly scripts because those are relatively portable and applications which run scriptlets don't have nearly the same level of security as the OS itself.
Even then, the massive fragmentation of the application base will severely limit viruses not designed to attack near-universal software. A postfix/sendmail flaw would be more serious than a comparable glitch in blender. Not because of privileges (although postfix obviously needs more rights than blender) but because virtually every Linux box will have a mail server but only graphics modelers will need graphics modeling software.
With e-mail software, anything that attacks Outlook (regular or express) on Windows has a good chance of finding a machine that is vulnerable. For Linux, there are hundreds of clients and no one has sufficient market share to create a substantial vulnerability. For databases, the overwhelming majority use MS SQL Server or Oracle on a Windows box, but the Linux database scene is vast. Even in the MySQL arena, there are many forks where an attack on one may have no impact on another.
That's not to say Linux is impervious, merely that it isn't a good environment for virus writers.
I am not sure you have a good sense of how much scientific research costs
The UK spent about 1.2 billion pounds in the last financial year on science and medical research. That's less than I expected, to be honest. I think it's less, in relative terms, from when I worked for SERC.
how much the changes you're proposing will increase those costs
Not much. Anything that is ISO 9000 already handles all this, so there's no extra costs. It's already done - far in excess of what I'm outlining. My proposal - most "lab books" are no longer on paper, they're on computer. Rsync costs zilch. Duplicating the research is already done, since replication is central to science, the difference here is purely that the process can be verified as genuine replication. So the costs aren't as high as you feel.
For one, the amount of time that a scientist spends peer reviewing will balloon to be most of their time (and resources)
The idea is to distribute the workload as much as possible, so no scientist is going to be spending excess time peer reviewing and any time spent duplicating is time they'd spend replicating the result anyway because that's what they'd be doing anyway.
The difference on evaluation is that instead of muddling through trying to see if things make sense, they'd merely have to do a comparison of methods and results.
Much, perhaps most, scientific fraud in published studies has little to do with corporate R&D.
Cite, please. All references I can find are specifically to do with corporate sponsorship with journals and corporate sponsorship of reseach.
It's this "business/capitalism-is-the-root-of-all-evil" Marxist reductionism that is getting really tiresome to read. I get it. You don't like business.
This must be why I mentioned corporations putting money into the central pot. If there's any ideology here, it's yours, since you have evidently taken a few things utterly out of the context in which they were placed and imposed your own idea of what I "must have" meant according to some fantastically inaccurate wall-chart of phrases-to-politics.
And what did you even want to propose, anyway?
I said what I wanted to propose. In detail.
No private scientific R&D?
Plenty of private R&D in this framework. Private but decoupled.
Are you mad?
Those who have marked me as "foe" on Slashdot would say so. Those, like you, who simply don't read what I write and prefer to imagine what you want me to have written - well, that used to make me mad. These days, it makes me wish I could emigrate to Mars on the basis that microbes and amoeba offer better conversation.
It's also not a universally-applied fudge factor - globular clusters show no evidence of dark matter around their edges, regardless of how you perform the analysis.
There's some good science indicating dark matter, but there's just as much that conflicts with it. Even if dark matter does turn out to exist, the fact that there's any conflict at all means that current dark matter theories are not merely a little too simple but have some facts plain wrong within the bounds for which they are defined.
Agreed. However, it's all about "survival of the fittest". The current system favours the least work (since doing less means you can write more, and writing more means a higher citation score, which in turn means more funding), the work least likely to fail (negative results don't get published) and the work least likely to contradict prior work (repeat studies also don't get published).
In order for quality science to survive, it HAS to be the fittest for purpose, which means we've got to change the purpose so that the above three flaws are selected against and not for.
This is why I would argue for a shake-up of how science is funded and how papers are refereed, published and post-publish reviewed.
Science should NOT be corporate-funded, it should be grant-funded -- directly from a scientific organization like NIST, or indirectly via university (or other educational) departments. Corporations should be entitled to push money into a grant pool and should also be entitled to suggest problems to study, but there should be absolutely NO link between the providers of the money and the providers of the science. Scientists MUST be free to say a claim is wrong, obtain negative results or otherwise get results corporations aren't going to like. Sorry, the universe doesn't give a flying what your CEO says.
A paper should NOT be considered as having been refereed until the work has been reproduced. But what constitutes reproduction of a result? At least some forgeries have involved people taking prior published papers and doing a cut-and-paste on the tables of results. The values now necessarily agree. Is that reproduction of results? No. Conclusion - a copy of the lab notes during the experiments should be placed in escrow with the journal. Once the peer reviewers have also submitted their lab notes, the complete collection is released to a second-stage peer review to determine if the collection suggests anyone "cooked the books". Only when a paper passes second-stage review is it published.
Next, there need to be central scientific libraries that collect ALL journals (regardless of obscurity), ALL reviewed lab notes, etc, making that information available to absolutely anyone, with PROPER linkage between research (Semantic Web has nothing on this!). Journals will claim they need to make a profit -- fine, embargo new publications for N months after pay-per-view publication. Since I'm arguing for quality indexing, and given that takes time, such a library can't publish instantly anyway.
What to do with negative results, though? Journals hate publishing those. So, have the central funding agencies ALSO fund an "open journal" that ONLY publishes negative results. Journals can't complain that it's competing, since there's no overlap.
Ok, but even with all of that, nobody has time to read every paper and certainly nobody has time to go back and correlate current science with past papers even if all this information was available. Doesn't matter. If there's a central store of everything, and that everything is properly linked up, the reasoners that have already been written for Semantic Web logic will work on those links to determine if the data is internally consistent. That information can be passed back to the funding agencies to determine what experiments are needed (if any) to identify what results are good, what ones are fraud and what ones are merely incompetent.
This sort of framework is relatively open (anyone can join as a publisher, anyone can join as a researcher, anyone can throw money into the pool), but more importantly the information is open and the information lifecycle is a closed loop. Even if the majority of past data is bad in any given field, this system would make bad data unsustainable because it can't pass through a two-stage review anything like as easily as it can a one-stage because the criteria differ, and even if it did get through, it then has to handle an automated consistency check.
Yes, this is serious infrastructure we're talking. However, science journals cost many times more to publish in than open journals (roughly, $8,000 an article less, assuming the typical conversion rates). You don't need to hand that many papers being published before the cost of all the infrastructure needed matches the amount saved. The money then saved from eliminating the bad science then becomes pure profit, which can be ploughed into new work.
Yes it does. If you test software (rather than wait for something to break), use the patches for live kernel updates for Linux and have the kernel booted via Coreboot (nee LinuxBIOS), you can guarantee the above for even mainstream Linux quite easily.
"Testing" means actually developing test harnesses for different components, developing a full regression test suite (and keeping it up-to-date), hammering the kernel, using the various probes and trace toolkits as they were intended, documenting stuff properly, fixing stuff and then using said regression test suite to do continual testing of the repository (plenty of packages out there already provide this facility).
Why Coreboot? Because it can boot an OS in 3 seconds, meeting the deadline you give.
Why live updates? Then you only need to actually reboot when a bug is found that corrupts state irretrievably (so you can't simply update the software and plug state values in, recalculate, etc). How often do you hear of those being found? My guess is "not very". My other guess is that if you had a good testing regimen, given the extremely limited amount of hardware a phone will have (so you're dealing with not just one architecture but essentially one platform on one architecture) that a good testing regimen would find the majority of cases of that class of bug quickly.
The other thing to consider is that this is a probability, not an expression of an absolute. So if you've two phones and one has 2 seconds outage for that year but the other has 4 seconds outage, you've not violated the seven nines. Equally, if you've a million phones and at any given time one of them is out of service, you're still meeting the guarantee.
If you prefer, there is a 0.00001% chance of a given arc through the code having a bug severe enough to eliminate service. Linux, as of 2011, had 0.45 bugs per thousand lines of code. Assuming the average size of a code block to be 1,000 lines of code, you'd need to have a bug rate of 1/45th of that. That's the worst-case, where the bugs are uniformly spread. In practice, some modules will be less reliable than others. But even in the worst possible case, it's well within the capacity of a company like Microsoft or Google to hire a team of tester-developers whose sole remit is bug-stomping. Nothing else. No side issues, no distractions, just locating and exterminating bugs.
How many tester-developers would you need? Given that Linux is achieving such high quality when many kernel coders are known to despise testing of any kind whatsoever, if you needed more than a hundred for a narrow subset of Linux like the Android kernel, I would be very very surprised. And a hundred extra staff for the Googles of the world simply isn't going to be noticed in terms of cost. I doubt you'd need even that many.
That means bringing the bugs to 0.01 per thousand lines is very achievable, which means you meet the requirements for how likely a crash would be, which means that yes, you could achieve an average of 3 seconds downtime a year.
So it is indeed honestly realistic, not that expensive, and could be done any time a corporation wanted.
Could it be done for Windows 8 on a phone? The bug hunters would likely be a lot busier (Coverity rates Open Source as 150x more reliable than Closed Source) and you might only achieve five nines in the first couple of years, but after that I see no reason why seven nines couldn't be done.
The main obstacle is not difficulty, nor is it cost. It's that economics via "planned obsolescence" rather than "continuous maintenance" makes a lot more money. Makes a lot more pollution, makes a lot more hardship for users, but since we've always been at war with Eurasia, what does that matter?
Since lawyers are forbidden from asking certain types of question (eg: leading questions, questions without proper answers, etc), at the very least Ellison's lawyers failed to object or failed to prove the question had no meaning. Ergo, in the context the question was asked, it had enough of a meaning for Ellison to answer. Either that, or he should fire his legal department.
There were, however, clean-room Java implementations prior to 2005. Google may well have used one of those at that time. And, except where there is definitive proof that they're using Oracle Java now, they might well be using clean-room implementations today. We know very little about what Google are actually doing (and Google clearly wants to keep it that way).
Makes it hard for Oracle to make a compelling case that it is non-free if the man ultimately in charge of deciding doesn't know. They may well still make a compelling case, but even if they do, this admission will impact what they can claim in damages. (Google can legitimately claim that if Oracle doesn't know what it owns, Google cannot be wholly responsible for not knowing either.)
No, he CAN'T have his five minutes back - you know the rules, no refunds on Slashdot except through the complaints department.
"The Bad Astronomer" has been around long enough, and has enough credibility, that anyone who claims he's a karma whore is de-facto neither a nerd nor a geek and should gerroff our collective lawns.
Obvious Geometry is indeed Obvious. This guy, Euclid, wrote some of the Obvious Geometry down and used it as a teaching manual. Nobody had done that before. Everyone in his time knew the rules he was describing (Archimedes regarded them as insultingly simple), but few had understood the fundamentals (what was axiomatic, what was derivative) and absolutely nobody had thought of actually explaining things before. The result of him doing so caused the number of mathematicians and their skills to explode. The learning curve had become dramatically shallower.
This is really no different. Sure, it's basic but the learning curve of WHY it works, HOW it works and WHEN/WHERE it can be used is NOT common knowledge. This makes teaching the relationship of maths and astronomy a cakewalk. I have no objections to more people seeing why geometry and maths are relevant (a common complaint is that they aren't ever used anywhere, but that's because nobody explains why, yes, they are). I regard it as the sole opportunity for turning the world into people who can think for themselves.
I've never regarded "good enough" as Good Enough and I'm not about to start now. I remember phones that didn't crash -at all-. True, they were noisy and fixed into the wall, but replacing one defect with another has never appealed to me. I hate dumb cell phones because they drop connections (proper error-correction would eliminate that entirely, would fit onto the chips perfectly well, and would add practically no cost to the devices), have limited memory (you can get a flash with as many gigabytes as a dumb phone's memory has K) and have way too many signal shadows (at least some of which would be eliminated with better error-correction as it's not simply line-of-site that matters in such things).
I hate "smart phones" that crash - there are plenty of Carrier-Grade OS' out there, at least some can fit on phones, and there's no excuse for instability on an OS that's five nines rated.
In fact, given that programmers are cheap and plentiful in comparison to the actual amount of genuinely new work being done, there's really bugger all excuse for ANY system to not achieve five nines uptime. These days, that should be a minimum. With increasing reliance on fewer, smarter devices, we should be moving closer to the seven nines region as a target.
Oh, and the whole bloody world can gerroff my lawn.
I usually recommend waiting to SP2 as SP1 is usually poorly tested. XP, for example, only truly became stable after SP2 due to problems with SP1, and Windows 95 SP1 was notorious for adding massive security holes (beyond the usual ones).
Those who deliberately engineer masterpieces, those who "inadvertently" engineer masterpieces and those who write the (cough) software that causes the other two groups to act.
In this case, these accidental geniuses are responsible for work that mainstream GIMP developers had long claimed was impossible. From the looks of it, six impossible things were achieved, so said developers should round things off with a meal at Milliways.
Obvious claim is indeed obvious. Intelligence, however, is not merely not even a single trait, there isn't even a single definition for it (which is why the Turing Test - compare against something you think is intelligent and see if there's a difference - is still valuable).
In fact, no serious researcher has contended ANYTHING to be controlled by a single gene since sequencing genes became possible - and many were seriously doubting it long before then.
(Even something as basic as "Chronic Fatigue" - M.E. - has seven distinct genes involved in it.)
Quite the contrary. Most serious researchers now not only contend that just about everything is controlled by multiple genes, they're also saying that not all those genes are even in human cells. There are over 5,500 different species of organism in humans (known, that is - there may be more), constituting roughly 10x as many actual cells as there are human cells. The interactions between human cells and these symbiotic cells are vast, complex and extremely difficult to map (since the symbiotic cells can vary up to 50% in their genome between human hosts, whereas the human genome varies only about 1.5%), but because there are so many interactions and there is such a rich biochemistry as a result, there is no meaningful distinction between a genetic disease in the microflaura and a genetic disease in the host.
Slashdot Mirror
As the hypersonic craft demonstrated so well, there is a vast gulf between what theory says could be done and what can actually be done in practice. The only way to know is to build.
Secondly, and more importantly for NASA, something like a BWB airliner could perform high altitude research currently impossible because no conventional aircraft has the lift capacity or the space availability. NASA use aircraft extensively in research, but are usually stuck with aircraft wholly unsuited to what they want to do.
Finally, and most importantly of all, NASA Langley IS a design house for aviation. LaRC is also where NASA began and what a key part of NASA is.
Obviously, in the case of New Horizons, you couldn't reproduce the result except by building a new probe. Likewise, reproducing the results of ITER or LHC would be difficult. There, you'd be utterly reliant on analyzing the steps up to the experiment.
I absolutely concur on publication of software (and anything else used in analysis, such as custom hardware designs). The complete setup that permitted the experiment to take place, plus the setup that permitted analysis of the results, should be described in a way that can be reproduced or - at the very least - be analyzed systematically to determine if it could do what was claimed.
I agree that papers should not be treated as currency and that "publish or perish" should itself perish, but I refer you to the experiences of Open Source and of Open Science -- "release early, release often" tends to produce better results overall. The challenge is how to make use of this.
The Open Source experience shows that multi-stage code reviews work better than single-stage, that testing is critical (even though we all hate doing it), that good documentation will always outperform bad documentation, that a central repository scores over fragmented repositories, and that finding stuff is one of the most important tasks a programmer has to perform.
Seems simple enough. We know this model works, we know that the bits of this model that have been back-ported to academia have worked exceedingly well, so it would seem obvious enough to try back-porting the entire model and using that as a starting point for discussing how to fix science. Turns out that many of the key components already exist in some form in academia, so we know those components are doable with no extra effort.
It may be that my model is too complex and that your suggestion is all that is needed, but I'd rather have a meaningful discussion on it than the trial-by-combat that I'm seeing from some of the other posters.
Research costs are always high. However, in principle, a basic waverider with a hydrogen-powered scramjet aught to be a lot cheaper than an ICBM (hydrogen being a much cheaper fuel, and 100% of the vehicle is reusable).
Note - that's in principle. Practice rarely pays much attention to theory. A space elevator would be nice, too, as would a pony.
The biggest problem is that the hypersonic craft was NASA's last shot at this for a while. The last time the test craft was covered, it was stated that the project is terminated. Kaput. (In short, in about the same shape as NASA's blended wing body passenger aircraft, whose sheer size would have made the Airbus 400 look like a unicycle.) NASA hasn't the funds to see projects to completion any more. That wouldn't be a total loss, if the data collected was made available. But, no, this sort of stuff is kept under very heavy wraps. Nothing learned will be made available to anyone, the research will get redone by others at massive expense (and possible loss of life), and all to stop those nations who could never build such a vehicle anyway from scoring a PR coup. ie: it's politics. It's not even military, it's politics.
Ok, correction, then. The biggest problem is that we have politicians.
In theory, yes. The type of design (a "waverider") places the hypersonic shockwave directly beneath the vehicle. Basically, you're surfing the shockwave. This reduces the stresses involved, improves stability and should allow considerably more control than could be achieved with the space shuttle (you have sufficient lift from a waverider to glide). Waveriders do have disadvantages - most designs only work at specific speeds, the wings have a habit of frying and they rely on cooling by radiation (only effective at high altitude).
Old wisdom on waveriders:
http://research.lifeboat.com/surf.htm
http://www.aerospaceweb.org/design/waverider/waverider.shtml
Published theory:
http://www.waset.org/journals/waset/v79/v79-79.pdf
http://www.dept.aoe.vt.edu/~mason/Mason_f/ConfigAeroHypersonics.pdf
Multi-speed waveriders:
http://www.springerlink.com/content/x75nh2154nuh5464/
Amateur waverider research:
http://www.gbnet.net/orgs/staar/waveriders.html
NB: The STAAR group beat NASA and the US DoD to the first working waverider airfoil, as noted on their site. Perhaps NASA's problem with their current design is that they're not threatening the engineers with bagpipe music.
Just as important, there are multiple distributions. Just as it makes it hard to write commercial software that will run under any version of any distro, it makes it hard to write a virus that will work under any version of any distro. The odds are that Linux viruses will be predominantly scripts because those are relatively portable and applications which run scriptlets don't have nearly the same level of security as the OS itself.
Even then, the massive fragmentation of the application base will severely limit viruses not designed to attack near-universal software. A postfix/sendmail flaw would be more serious than a comparable glitch in blender. Not because of privileges (although postfix obviously needs more rights than blender) but because virtually every Linux box will have a mail server but only graphics modelers will need graphics modeling software.
With e-mail software, anything that attacks Outlook (regular or express) on Windows has a good chance of finding a machine that is vulnerable. For Linux, there are hundreds of clients and no one has sufficient market share to create a substantial vulnerability. For databases, the overwhelming majority use MS SQL Server or Oracle on a Windows box, but the Linux database scene is vast. Even in the MySQL arena, there are many forks where an attack on one may have no impact on another.
That's not to say Linux is impervious, merely that it isn't a good environment for virus writers.
The UK spent about 1.2 billion pounds in the last financial year on science and medical research. That's less than I expected, to be honest. I think it's less, in relative terms, from when I worked for SERC.
Not much. Anything that is ISO 9000 already handles all this, so there's no extra costs. It's already done - far in excess of what I'm outlining. My proposal - most "lab books" are no longer on paper, they're on computer. Rsync costs zilch. Duplicating the research is already done, since replication is central to science, the difference here is purely that the process can be verified as genuine replication. So the costs aren't as high as you feel.
The idea is to distribute the workload as much as possible, so no scientist is going to be spending excess time peer reviewing and any time spent duplicating is time they'd spend replicating the result anyway because that's what they'd be doing anyway.
The difference on evaluation is that instead of muddling through trying to see if things make sense, they'd merely have to do a comparison of methods and results.
Cite, please. All references I can find are specifically to do with corporate sponsorship with journals and corporate sponsorship of reseach.
This must be why I mentioned corporations putting money into the central pot. If there's any ideology here, it's yours, since you have evidently taken a few things utterly out of the context in which they were placed and imposed your own idea of what I "must have" meant according to some fantastically inaccurate wall-chart of phrases-to-politics.
I said what I wanted to propose. In detail.
Plenty of private R&D in this framework. Private but decoupled.
Those who have marked me as "foe" on Slashdot would say so. Those, like you, who simply don't read what I write and prefer to imagine what you want me to have written - well, that used to make me mad. These days, it makes me wish I could emigrate to Mars on the basis that microbes and amoeba offer better conversation.
It's also not a universally-applied fudge factor - globular clusters show no evidence of dark matter around their edges, regardless of how you perform the analysis.
There's some good science indicating dark matter, but there's just as much that conflicts with it. Even if dark matter does turn out to exist, the fact that there's any conflict at all means that current dark matter theories are not merely a little too simple but have some facts plain wrong within the bounds for which they are defined.
Agreed. However, it's all about "survival of the fittest". The current system favours the least work (since doing less means you can write more, and writing more means a higher citation score, which in turn means more funding), the work least likely to fail (negative results don't get published) and the work least likely to contradict prior work (repeat studies also don't get published).
In order for quality science to survive, it HAS to be the fittest for purpose, which means we've got to change the purpose so that the above three flaws are selected against and not for.
This is why I would argue for a shake-up of how science is funded and how papers are refereed, published and post-publish reviewed.
Science should NOT be corporate-funded, it should be grant-funded -- directly from a scientific organization like NIST, or indirectly via university (or other educational) departments. Corporations should be entitled to push money into a grant pool and should also be entitled to suggest problems to study, but there should be absolutely NO link between the providers of the money and the providers of the science. Scientists MUST be free to say a claim is wrong, obtain negative results or otherwise get results corporations aren't going to like. Sorry, the universe doesn't give a flying what your CEO says.
A paper should NOT be considered as having been refereed until the work has been reproduced. But what constitutes reproduction of a result? At least some forgeries have involved people taking prior published papers and doing a cut-and-paste on the tables of results. The values now necessarily agree. Is that reproduction of results? No. Conclusion - a copy of the lab notes during the experiments should be placed in escrow with the journal. Once the peer reviewers have also submitted their lab notes, the complete collection is released to a second-stage peer review to determine if the collection suggests anyone "cooked the books". Only when a paper passes second-stage review is it published.
Next, there need to be central scientific libraries that collect ALL journals (regardless of obscurity), ALL reviewed lab notes, etc, making that information available to absolutely anyone, with PROPER linkage between research (Semantic Web has nothing on this!). Journals will claim they need to make a profit -- fine, embargo new publications for N months after pay-per-view publication. Since I'm arguing for quality indexing, and given that takes time, such a library can't publish instantly anyway.
What to do with negative results, though? Journals hate publishing those. So, have the central funding agencies ALSO fund an "open journal" that ONLY publishes negative results. Journals can't complain that it's competing, since there's no overlap.
Ok, but even with all of that, nobody has time to read every paper and certainly nobody has time to go back and correlate current science with past papers even if all this information was available. Doesn't matter. If there's a central store of everything, and that everything is properly linked up, the reasoners that have already been written for Semantic Web logic will work on those links to determine if the data is internally consistent. That information can be passed back to the funding agencies to determine what experiments are needed (if any) to identify what results are good, what ones are fraud and what ones are merely incompetent.
This sort of framework is relatively open (anyone can join as a publisher, anyone can join as a researcher, anyone can throw money into the pool), but more importantly the information is open and the information lifecycle is a closed loop. Even if the majority of past data is bad in any given field, this system would make bad data unsustainable because it can't pass through a two-stage review anything like as easily as it can a one-stage because the criteria differ, and even if it did get through, it then has to handle an automated consistency check.
Yes, this is serious infrastructure we're talking. However, science journals cost many times more to publish in than open journals (roughly, $8,000 an article less, assuming the typical conversion rates). You don't need to hand that many papers being published before the cost of all the infrastructure needed matches the amount saved. The money then saved from eliminating the bad science then becomes pure profit, which can be ploughed into new work.
Yes it does. If you test software (rather than wait for something to break), use the patches for live kernel updates for Linux and have the kernel booted via Coreboot (nee LinuxBIOS), you can guarantee the above for even mainstream Linux quite easily.
"Testing" means actually developing test harnesses for different components, developing a full regression test suite (and keeping it up-to-date), hammering the kernel, using the various probes and trace toolkits as they were intended, documenting stuff properly, fixing stuff and then using said regression test suite to do continual testing of the repository (plenty of packages out there already provide this facility).
Why Coreboot? Because it can boot an OS in 3 seconds, meeting the deadline you give.
Why live updates? Then you only need to actually reboot when a bug is found that corrupts state irretrievably (so you can't simply update the software and plug state values in, recalculate, etc). How often do you hear of those being found? My guess is "not very". My other guess is that if you had a good testing regimen, given the extremely limited amount of hardware a phone will have (so you're dealing with not just one architecture but essentially one platform on one architecture) that a good testing regimen would find the majority of cases of that class of bug quickly.
The other thing to consider is that this is a probability, not an expression of an absolute. So if you've two phones and one has 2 seconds outage for that year but the other has 4 seconds outage, you've not violated the seven nines. Equally, if you've a million phones and at any given time one of them is out of service, you're still meeting the guarantee.
If you prefer, there is a 0.00001% chance of a given arc through the code having a bug severe enough to eliminate service. Linux, as of 2011, had 0.45 bugs per thousand lines of code. Assuming the average size of a code block to be 1,000 lines of code, you'd need to have a bug rate of 1/45th of that. That's the worst-case, where the bugs are uniformly spread. In practice, some modules will be less reliable than others. But even in the worst possible case, it's well within the capacity of a company like Microsoft or Google to hire a team of tester-developers whose sole remit is bug-stomping. Nothing else. No side issues, no distractions, just locating and exterminating bugs.
How many tester-developers would you need? Given that Linux is achieving such high quality when many kernel coders are known to despise testing of any kind whatsoever, if you needed more than a hundred for a narrow subset of Linux like the Android kernel, I would be very very surprised. And a hundred extra staff for the Googles of the world simply isn't going to be noticed in terms of cost. I doubt you'd need even that many.
That means bringing the bugs to 0.01 per thousand lines is very achievable, which means you meet the requirements for how likely a crash would be, which means that yes, you could achieve an average of 3 seconds downtime a year.
So it is indeed honestly realistic, not that expensive, and could be done any time a corporation wanted.
Could it be done for Windows 8 on a phone? The bug hunters would likely be a lot busier (Coverity rates Open Source as 150x more reliable than Closed Source) and you might only achieve five nines in the first couple of years, but after that I see no reason why seven nines couldn't be done.
The main obstacle is not difficulty, nor is it cost. It's that economics via "planned obsolescence" rather than "continuous maintenance" makes a lot more money. Makes a lot more pollution, makes a lot more hardship for users, but since we've always been at war with Eurasia, what does that matter?
Had the Vikings invented geometry.....
Since lawyers are forbidden from asking certain types of question (eg: leading questions, questions without proper answers, etc), at the very least Ellison's lawyers failed to object or failed to prove the question had no meaning. Ergo, in the context the question was asked, it had enough of a meaning for Ellison to answer. Either that, or he should fire his legal department.
There were, however, clean-room Java implementations prior to 2005. Google may well have used one of those at that time. And, except where there is definitive proof that they're using Oracle Java now, they might well be using clean-room implementations today. We know very little about what Google are actually doing (and Google clearly wants to keep it that way).
Makes it hard for Oracle to make a compelling case that it is non-free if the man ultimately in charge of deciding doesn't know. They may well still make a compelling case, but even if they do, this admission will impact what they can claim in damages. (Google can legitimately claim that if Oracle doesn't know what it owns, Google cannot be wholly responsible for not knowing either.)
No, he CAN'T have his five minutes back - you know the rules, no refunds on Slashdot except through the complaints department.
"The Bad Astronomer" has been around long enough, and has enough credibility, that anyone who claims he's a karma whore is de-facto neither a nerd nor a geek and should gerroff our collective lawns.
Obvious Geometry is indeed Obvious. This guy, Euclid, wrote some of the Obvious Geometry down and used it as a teaching manual. Nobody had done that before. Everyone in his time knew the rules he was describing (Archimedes regarded them as insultingly simple), but few had understood the fundamentals (what was axiomatic, what was derivative) and absolutely nobody had thought of actually explaining things before. The result of him doing so caused the number of mathematicians and their skills to explode. The learning curve had become dramatically shallower.
This is really no different. Sure, it's basic but the learning curve of WHY it works, HOW it works and WHEN/WHERE it can be used is NOT common knowledge. This makes teaching the relationship of maths and astronomy a cakewalk. I have no objections to more people seeing why geometry and maths are relevant (a common complaint is that they aren't ever used anywhere, but that's because nobody explains why, yes, they are). I regard it as the sole opportunity for turning the world into people who can think for themselves.
I've never regarded "good enough" as Good Enough and I'm not about to start now. I remember phones that didn't crash -at all-. True, they were noisy and fixed into the wall, but replacing one defect with another has never appealed to me. I hate dumb cell phones because they drop connections (proper error-correction would eliminate that entirely, would fit onto the chips perfectly well, and would add practically no cost to the devices), have limited memory (you can get a flash with as many gigabytes as a dumb phone's memory has K) and have way too many signal shadows (at least some of which would be eliminated with better error-correction as it's not simply line-of-site that matters in such things).
I hate "smart phones" that crash - there are plenty of Carrier-Grade OS' out there, at least some can fit on phones, and there's no excuse for instability on an OS that's five nines rated.
In fact, given that programmers are cheap and plentiful in comparison to the actual amount of genuinely new work being done, there's really bugger all excuse for ANY system to not achieve five nines uptime. These days, that should be a minimum. With increasing reliance on fewer, smarter devices, we should be moving closer to the seven nines region as a target.
Oh, and the whole bloody world can gerroff my lawn.
Yes, but he has to have a Mandelbrot Set on a t-shirt or on the back of his jacket.
I usually recommend waiting to SP2 as SP1 is usually poorly tested. XP, for example, only truly became stable after SP2 due to problems with SP1, and Windows 95 SP1 was notorious for adding massive security holes (beyond the usual ones).
It would solve the problem of what to do with all the European odd-balls (we're running a surplus).
Those who deliberately engineer masterpieces, those who "inadvertently" engineer masterpieces and those who write the (cough) software that causes the other two groups to act.
In this case, these accidental geniuses are responsible for work that mainstream GIMP developers had long claimed was impossible. From the looks of it, six impossible things were achieved, so said developers should round things off with a meal at Milliways.
According to yet other researchers, "researchers" have lost their meaning.
Obvious claim is indeed obvious. Intelligence, however, is not merely not even a single trait, there isn't even a single definition for it (which is why the Turing Test - compare against something you think is intelligent and see if there's a difference - is still valuable).
In fact, no serious researcher has contended ANYTHING to be controlled by a single gene since sequencing genes became possible - and many were seriously doubting it long before then.
(Even something as basic as "Chronic Fatigue" - M.E. - has seven distinct genes involved in it.)
Quite the contrary. Most serious researchers now not only contend that just about everything is controlled by multiple genes, they're also saying that not all those genes are even in human cells. There are over 5,500 different species of organism in humans (known, that is - there may be more), constituting roughly 10x as many actual cells as there are human cells. The interactions between human cells and these symbiotic cells are vast, complex and extremely difficult to map (since the symbiotic cells can vary up to 50% in their genome between human hosts, whereas the human genome varies only about 1.5%), but because there are so many interactions and there is such a rich biochemistry as a result, there is no meaningful distinction between a genetic disease in the microflaura and a genetic disease in the host.
Raising up the lower end would be a good start, but eliminating genetic defects responsible for there being sociopaths would also help.