Sweden is not a member of NATO, according to Wikipedia, and they never were one. They decided to sit the Cold War out and remain neutral.
Yeah, but the EU has lots of NATO members. It's somewhat plausible the EU parliament might vote to waive inviolability for that reason. If the PP gets into Riskdagen that changes since it's not illegal to divulge NATO secrets in Sweden (or the other way around), although generally the governments of friendly states try to avoid that. They usually also pressure local media not to divulge foreign secrets. In the case of some random server they could have the police seize it under some vague pretext (like to determine whether any Swedish secrets might be kept on them) - but with the inviolability of Riksdagen this would be impossible. It would require a 5/6 majority to waive inviolability, which just isn't going to happen. MPs handle classified information all the time, and without evidence of actual treason or espionage the vote would never pass; every MP would be asking themselves if they were next...
Odd to think of even being investigated on such boasts much less convicted.
You won't be arrested without evidence of a crime (no judge would issue a warrant), however it's enough for search warrants and holding you for questioning. Most likely this driver shot and posted his own video - dumb criminals and all that.
Now we just have to wait for the generation of programmers educated/weaned on MS tools to die off, and we're good:)
Which will probably happen sooner than many think if Windows loses market share. PCs and Windows is something you program not because you like working with them, but because someone pays you to do so and you're in need of money. People who aren't very adept and productive programmers tend to escape to engineering management or bizdev at first opportunity, and these tend to be people who have only experience with Windows and don't really think of programming as a craft. They will simply be unable to change and for this reason Windows programming will be as quickly forgotten among mainstream programmers as cobol.
They have every right to not be particularly happy. The programmers, these economics PhDs, know their worth and it isn't 125k a year.
People who build complex systems that are core to a business tend to also know their worth, and will eventually leave and start their own business building a competing system - only better. Then they sell it back for what it's actually worth. The funny thing is a lot of the dumbf*ck PHBs still won't get it, they're so utterly stuck in their trivial mental model.
But lets say I blended that price in with something else that they also wanted, and they'd pay for that over time as well? How about a phone contract?
Even if you jailbreak or unlock the phone you'll still have a contract. You can connect an unlocked iPhone to T-Mobile, but you'll still have a legal commitment to keep paying AT&T, who will keep billing you. If you want out of your original contract there's an early termination fee to cover the device cost.
The discussion of interest is that once the contract reaches the end of its term (two years or whatever) you 1) no longer have to keep paying to subsidize a device already amortized, and 2) you have the option to terminate your contract and change carriers if you feel the one you have is so poor it's worth the termination fee. In fact, the device should automatically be unlocked at the end of the contract term (since all obligations are met at that point), or the termination fee has been paid and the contract canceled.
For people who have paid off their devices in full we need a free market for services that reflect network cost, with zero device cost.
We all remember smart kids from high school that wasted their lives on pot.
Repeat after me: HS age teenagers are children, not adults. The social dynamics of high school don't apply to society at large. Children are DUMB - you were too, trust me. Lots of people wasted away their high school years, yet they're perfectly fine responsible adults. The A student might be the one who turns out to have an addictive or criminal make-money-fast personality. You can't fit adult behavior into HS age stereotypes, it simply doesn't apply. Life is not that trivial, because people change when they grow up.
In fact, it will likely be less detrimental than alcohol, and that's already legal.
A lot of people get belligerent and violent when drunk. I'd rather have them stoned. For the rest of us normal people, I don't know why the government would or even should care if I have a drink or a couple of puffs of pot. They should just mind their own business and go find something useful to occupy themselves with. To be frank I think the illicit market for dealing in contraband is far more detrimental than the contraband itself - at least for things like booze and pot. People get killed in turf wars to control the illicit trade and to show off their third-world peasant machismo, not from smoking pot.
QuickDraw is available from the same download page. It's also a lot more interesting than MacPaint. But at 17k lines of 68k assembly it has no real practical value; it's probably easier to reimplement it in C with an abstract hardware layer than try to port the original code to a different graphics system.
Example: When you were signed into your gmail account last, you searched for golf clubs, pricey restaurants and luxury spa getaways. You're probably more likely to care less about the price of your ticket than the guy who searched for hostels, cheap rental cars and other "budget" websites.
So you'd create an account solely for the purpose of making budget travel arrangements - then seed it with a history to yield the best possible deals. Works for me.
It's all part of the 'Knots per hour' and 'Watts per day' malaise that all journalists are infected with.
None of them* can use units correctly, leaving us to try to interpret what the scientist, who wrote the notes that were mismassaged into a press release which was misinterpreted by the journalist, was trying to say.
*unjustified absolute. YHBT
They should just stick to "high rate of speed". The reader gets to guess all the same, anyway.
There will be a few growing pains, but once APIs get straightened out and programmers get used to it (which shouldn't take too long for the ones writing HPC code), this is going to be a huge win for scientific computing.
All you say makes sense, but I for one don't understand the market for this. Today, if you need a compute server that's good for stream (e.g. SIMD) workloads you get a dozen 1U/2U rackmounts and fill them up with as many GPU boards as they'll take. You put a work scheduler on them that accepts tasks from a dispatch server, and hook them up to a NAS box (or just rsync data sets and results from a storage subsystem). Then you put a transaction server in front it all that exposes a job manager.
Throwing a couple of GPU boards in a transaction server will add some computational punch, but not enough to make it a real compute server. It'll be too expensive to rack and stack. It'll cost more than a plain old transaction server. The market clearly is whoever needs a little bit of computational power on their backend - but does it really exist?
Your rounding is incorrect. 1 ft = 0.305 m (exact), 1 m^2 = 1/(0.305)^2 ft^2 = 10.7497884... = 10.7 if you round it to the nearest 0.1.
0.305 has three significant digits, and as a rule of thumb you can't really get a result more accurate than this. 10.7 is as precise as 0.305 will afford you. With 0.3048 you'd get 10.76.
So somebody's free speech is being violated because a government agency doesn't want people reading Slashdot when they're supposed to be working? Are you utterly against workplaces trying to get their employees to not browse the web? Or just the government?
As long as they get their job done, who cares what they read, what radio station they listen to, podcasts, whatever. As long as it doesn't interfere with their job. People also have and need breaks. As an employer myself I only block sites that are illegal or can be construed as workplace harassment - like porn, and only tell my employees they can't click through EULAs or pretend to represent the company. Otherwise I recognize they are human beings who exist in a social context, and that context is to a large extent mediated by the net. If you treat people like mindless drones, surprise - that's what you get. Treat your employees like shit and you'll be hiring from the bottom of the talent pool. Of course, the TSA practically wrote the book on that...
Why should the employer care whether I read Slashdot on a break or when waiting for something? Remove the coffee machine and newspapers if you want people to stare at the roof when they're not actively doing anything.
You're talking about trust and integrity here. Clearly, the TSA doesn't think highly of its own staff in this department.
They used the "write zero" disk erase method, which in fact un-erase every NAND block of the disk, which in turn forces the disk to erase each block again as it writes. Thats why they see such consistency of results : they are measuring the worst possible case where the disk is forced to the slow path for each block.
To erase NAND, you need to erase it by block, and the resulting block is full of 1's. Writing to NAND is a question of writing zeros in places, you can't write 1's on NAND unless you erase it.
So in a way, their test is showing that OSX is much, much better than windows when the disk is dirty, but that apple hasn't implemented the "trim" that allows the disk to re-erase nand 'free' blocks.
While erased flash is all 1's, it's complemented before the host sees it, so the host reads erased flash as all zeros. So filling a drive with zeros is pretty much the same as erasing it. And if the drive controller is remotely intelligent it will detect that a sector is written with all 1's and not bother even allocating it; it can store a magic value in the sector map to indicate this logical sector is blank and has no physical allocation. The first non-zero write to it would then cause it to be allocated. The fact that at least one common OS tool writes zeros (and DU can't be the only one) to erase drive space is pretty good motivation to handle this correctly at the allocation level. I think it's pretty safe to assume that any firmware engineer who works on SSDs is perfectly aware of all these aspects - and more.
So actually giving classified secrets to a foreign power is a very serious crime. It is the kind of thing that can earn life in prison, or even death if it is done during war time. Working as an agent for a foreign government and trying to get classified data is also illegal, though less so.
As far as I can tell there was no classified information involved. They did things like collect media information on how Russia and its politics are portrayed and perceived in the U.S. There is no harm to the U.S. - it's purely a 'procedural' charge, for failing to register. Every country engages in these activities in just about every foreign country, it's not considered harmful or hostile, just plain simple information gathering. In addition, transferring files over WiFi is not all that advanced these days, and encryption is pretty much standard in the intelligence community and doesn't really mean a whole lot. To the 'spies' it was probably just contract work - they got paid like any other journalist or writer.
Of course, there could have been something else going on, but on the face of it as explained it seems pretty frivolous to bother arresting them. Instead of, say, reminding them to file paperwork and pay their fees.
There can also be a self-destruct password - which gives access to the decoy volume but begins a hard erase of the real data in background. You'd write that on a piece of paper and keep it in your wallet for a hostile to find. I'd expect the FBI to be a bit too savvy to bumble into traps like that.
They only have to look at the code to see what algorithms are used. Breaking two ciphers takes twice as long as breaking one, which is more or less equivalent to adding one bit to the key - i.e. not much difference. Also, it doesn't have to be two different ciphers, it could be one encrypted twice using one cipher with two different keys.
Men don't have to be passionate about computers and programming to do well in our field.
You don't have to be "passionate" in the sense that you're gadget obsessed, can assemble PCs, operate Linux to keep it from falling over, and know all the latest Intel processor code names, but you do have to be self-motivated. If you haven't learned programming basics on your own before you sit down for your first college CS class you're in the wrong field. Similarly, if you haven't wired up simple circuits and understand what basic discrete components do before your first EE class it's almost certainly a waste of effort. If the purpose of the education isn't a means to understanding but merely the start of a career path, then this will lead to a very common pattern: 1) poor to mediocre engineer, 2) gets very little done, 3) plays politics and takes credit for others' work, 4) bails to management at first opportunity because it's "more interesting" than engineering. It takes time and concentrated effort to become a good programmer - a recent MS graduate is barely employable, and doesn't start to become strong until after about ten years' experience. In another ten they're solid. If they're not motivated to do this but just see it as a career they'll never last that long.
They don't need your retina. They just need whatever big integer your retina digests to.
In case the conclusion isn't obvious: if they can get you to authenticate using a compromised scanner you'll only be able to handle that breach exactly once - assuming you have a second eye.
Then you use your retina along with your fingerprint.
Sure identity theft is always going to be possible but it would be much harder if they had to get your retina than if they just had to memorize your digits and crack a password.
They don't need your retina. They just need whatever big integer your retina digests to.
Intel's low-power Atom is also RISC (the complex x86 instruction set is emulated using "micro-ops") and seems to do OK with power consumption.
Dynamic translation of a CISC instruction flow onto a pipeline adds complexity. The other big difference is that x86 has a fully coherent N-way associative cache which neither ARM or MIPS provide - at quite a bit of complexity. Heck, ARM prior to Cortex didn't even support unaligned memory access, and I'm not sure if MIPS does in any of their recent designs. Complexity requires die real estate. A bigger die is more expensive, and with more gates involved in just about every instruction fetch, decode, pipeline step, and memory bus cycle power consumption is going to be higher. Without the need to execute legacy code there's no way Intel can make it work with an inferior product at lower margin. They're going to shovel a few hundred million into the furnace to test the waters, then decide they can't make it work and exit. Undoubtedly they expect to lose money on every unit sold, to spend their way to a dominant market position where they can manage costs and eke out a margin through sheer volume, similar to their desktop and server processor business. I doubt management will have the guts to take the hemorrhaging and stick it out; it's a company used to seeing more or less instant returns from its other business.
Re:Theoretical performance vs real-world performan
on
Knuth Got It Wrong
·
· Score: 1
No, it assumes memory for which all access are *at worst* a certain equal cost - in other words, memory for which accesses have an upper bound. Knuth's analysis still holds if certain pieces of memory are FASTER than the norm. If we take memory access going through swap as the normal, worst case, upper-bound cost of memory access, and RAM and cache hits as special better cases, the analysis still applies to the real world.
Well, both atan() and += are "operations". But clearly a loop of N atans, while of the same polynomial complexity, will be orders of magnitude slower than a loop of N additions. This is because += is O(1) while atan() is O(nlogn) for some accuracy n iirc. Similarly, there's a complexity difference between accessing L1/L2 cache and accessing main memory over a shared bus. The two will never converge or cross; there is no data set size where atan() will outperform +=. To say that the former is a worst-case cost estimate for the latter is only correct in a discrete mathematical sense, for all practical purposes it's meaningless. An algorithm that can rely on addition will always be superior to one relying on atans, even if of the same complexity. In fact, even an O(nlogn) addition-based algorithm will be superior to an O(n) atan based one - the two will converge at infinity (assuming I'm correct about the complexity of atan), so for any finite (real-world) data set the former will be faster.
Slashdot Mirror
Sweden is not a member of NATO, according to Wikipedia, and they never were one. They decided to sit the Cold War out and remain neutral.
Yeah, but the EU has lots of NATO members. It's somewhat plausible the EU parliament might vote to waive inviolability for that reason. If the PP gets into Riskdagen that changes since it's not illegal to divulge NATO secrets in Sweden (or the other way around), although generally the governments of friendly states try to avoid that. They usually also pressure local media not to divulge foreign secrets. In the case of some random server they could have the police seize it under some vague pretext (like to determine whether any Swedish secrets might be kept on them) - but with the inviolability of Riksdagen this would be impossible. It would require a 5/6 majority to waive inviolability, which just isn't going to happen. MPs handle classified information all the time, and without evidence of actual treason or espionage the vote would never pass; every MP would be asking themselves if they were next...
Odd to think of even being investigated on such boasts much less convicted.
You won't be arrested without evidence of a crime (no judge would issue a warrant), however it's enough for search warrants and holding you for questioning. Most likely this driver shot and posted his own video - dumb criminals and all that.
Now we just have to wait for the generation of programmers educated/weaned on MS tools to die off, and we're good :)
Which will probably happen sooner than many think if Windows loses market share. PCs and Windows is something you program not because you like working with them, but because someone pays you to do so and you're in need of money. People who aren't very adept and productive programmers tend to escape to engineering management or bizdev at first opportunity, and these tend to be people who have only experience with Windows and don't really think of programming as a craft. They will simply be unable to change and for this reason Windows programming will be as quickly forgotten among mainstream programmers as cobol.
They have every right to not be particularly happy. The programmers, these economics PhDs, know their worth and it isn't 125k a year.
People who build complex systems that are core to a business tend to also know their worth, and will eventually leave and start their own business building a competing system - only better. Then they sell it back for what it's actually worth. The funny thing is a lot of the dumbf*ck PHBs still won't get it, they're so utterly stuck in their trivial mental model.
I think with displays the machines cost about $4K each, just for Photoshop and Illustrator. Why?
Because in the developed world people cost a lot more than machines. If you think $4000 is a lot money, wait til you see payroll...
But lets say I blended that price in with something else that they also wanted, and they'd pay for that over time as well? How about a phone contract?
Even if you jailbreak or unlock the phone you'll still have a contract. You can connect an unlocked iPhone to T-Mobile, but you'll still have a legal commitment to keep paying AT&T, who will keep billing you. If you want out of your original contract there's an early termination fee to cover the device cost.
The discussion of interest is that once the contract reaches the end of its term (two years or whatever) you 1) no longer have to keep paying to subsidize a device already amortized, and 2) you have the option to terminate your contract and change carriers if you feel the one you have is so poor it's worth the termination fee. In fact, the device should automatically be unlocked at the end of the contract term (since all obligations are met at that point), or the termination fee has been paid and the contract canceled.
For people who have paid off their devices in full we need a free market for services that reflect network cost, with zero device cost.
What sort of a choice is this?
What? We give you democracy and now you ask for choices?! What an attitude! You're not a team player!
We all remember smart kids from high school that wasted their lives on pot.
Repeat after me: HS age teenagers are children, not adults. The social dynamics of high school don't apply to society at large. Children are DUMB - you were too, trust me. Lots of people wasted away their high school years, yet they're perfectly fine responsible adults. The A student might be the one who turns out to have an addictive or criminal make-money-fast personality. You can't fit adult behavior into HS age stereotypes, it simply doesn't apply. Life is not that trivial, because people change when they grow up.
In fact, it will likely be less detrimental than alcohol, and that's already legal.
A lot of people get belligerent and violent when drunk. I'd rather have them stoned. For the rest of us normal people, I don't know why the government would or even should care if I have a drink or a couple of puffs of pot. They should just mind their own business and go find something useful to occupy themselves with. To be frank I think the illicit market for dealing in contraband is far more detrimental than the contraband itself - at least for things like booze and pot. People get killed in turf wars to control the illicit trade and to show off their third-world peasant machismo, not from smoking pot.
None of the Mac toolbox APIs are available
QuickDraw is available from the same download page. It's also a lot more interesting than MacPaint. But at 17k lines of 68k assembly it has no real practical value; it's probably easier to reimplement it in C with an abstract hardware layer than try to port the original code to a different graphics system.
Example: When you were signed into your gmail account last, you searched for golf clubs, pricey restaurants and luxury spa getaways. You're probably more likely to care less about the price of your ticket than the guy who searched for hostels, cheap rental cars and other "budget" websites.
So you'd create an account solely for the purpose of making budget travel arrangements - then seed it with a history to yield the best possible deals. Works for me.
It's all part of the 'Knots per hour' and 'Watts per day' malaise that all journalists are infected with.
None of them* can use units correctly, leaving us to try to interpret what the scientist, who wrote the notes that were mismassaged into a press release which was misinterpreted by the journalist, was trying to say.
*unjustified absolute. YHBT
They should just stick to "high rate of speed". The reader gets to guess all the same, anyway.
There will be a few growing pains, but once APIs get straightened out and programmers get used to it (which shouldn't take too long for the ones writing HPC code), this is going to be a huge win for scientific computing.
All you say makes sense, but I for one don't understand the market for this. Today, if you need a compute server that's good for stream (e.g. SIMD) workloads you get a dozen 1U/2U rackmounts and fill them up with as many GPU boards as they'll take. You put a work scheduler on them that accepts tasks from a dispatch server, and hook them up to a NAS box (or just rsync data sets and results from a storage subsystem). Then you put a transaction server in front it all that exposes a job manager.
Throwing a couple of GPU boards in a transaction server will add some computational punch, but not enough to make it a real compute server. It'll be too expensive to rack and stack. It'll cost more than a plain old transaction server. The market clearly is whoever needs a little bit of computational power on their backend - but does it really exist?
Your rounding is incorrect. 1 ft = 0.305 m (exact), 1 m^2 = 1/(0.305)^2 ft^2 = 10.7497884... = 10.7 if you round it to the nearest 0.1.
0.305 has three significant digits, and as a rule of thumb you can't really get a result more accurate than this. 10.7 is as precise as 0.305 will afford you. With 0.3048 you'd get 10.76.
So somebody's free speech is being violated because a government agency doesn't want people reading Slashdot when they're supposed to be working? Are you utterly against workplaces trying to get their employees to not browse the web? Or just the government?
As long as they get their job done, who cares what they read, what radio station they listen to, podcasts, whatever. As long as it doesn't interfere with their job. People also have and need breaks. As an employer myself I only block sites that are illegal or can be construed as workplace harassment - like porn, and only tell my employees they can't click through EULAs or pretend to represent the company. Otherwise I recognize they are human beings who exist in a social context, and that context is to a large extent mediated by the net. If you treat people like mindless drones, surprise - that's what you get. Treat your employees like shit and you'll be hiring from the bottom of the talent pool. Of course, the TSA practically wrote the book on that...
Why should the employer care whether I read Slashdot on a break or when waiting for something? Remove the coffee machine and newspapers if you want people to stare at the roof when they're not actively doing anything.
You're talking about trust and integrity here. Clearly, the TSA doesn't think highly of its own staff in this department.
They used the "write zero" disk erase method, which in fact un-erase every NAND block of the disk, which in turn forces the disk to erase each block again as it writes. Thats why they see such consistency of results : they are measuring the worst possible case where the disk is forced to the slow path for each block.
To erase NAND, you need to erase it by block, and the resulting block is full of 1's. Writing to NAND is a question of writing zeros in places, you can't write 1's on NAND unless you erase it.
So in a way, their test is showing that OSX is much, much better than windows when the disk is dirty, but that apple hasn't implemented the "trim" that allows the disk to re-erase nand 'free' blocks.
While erased flash is all 1's, it's complemented before the host sees it, so the host reads erased flash as all zeros. So filling a drive with zeros is pretty much the same as erasing it. And if the drive controller is remotely intelligent it will detect that a sector is written with all 1's and not bother even allocating it; it can store a magic value in the sector map to indicate this logical sector is blank and has no physical allocation. The first non-zero write to it would then cause it to be allocated. The fact that at least one common OS tool writes zeros (and DU can't be the only one) to erase drive space is pretty good motivation to handle this correctly at the allocation level. I think it's pretty safe to assume that any firmware engineer who works on SSDs is perfectly aware of all these aspects - and more.
So actually giving classified secrets to a foreign power is a very serious crime. It is the kind of thing that can earn life in prison, or even death if it is done during war time. Working as an agent for a foreign government and trying to get classified data is also illegal, though less so.
As far as I can tell there was no classified information involved. They did things like collect media information on how Russia and its politics are portrayed and perceived in the U.S. There is no harm to the U.S. - it's purely a 'procedural' charge, for failing to register. Every country engages in these activities in just about every foreign country, it's not considered harmful or hostile, just plain simple information gathering. In addition, transferring files over WiFi is not all that advanced these days, and encryption is pretty much standard in the intelligence community and doesn't really mean a whole lot. To the 'spies' it was probably just contract work - they got paid like any other journalist or writer.
Of course, there could have been something else going on, but on the face of it as explained it seems pretty frivolous to bother arresting them. Instead of, say, reminding them to file paperwork and pay their fees.
duress password
There can also be a self-destruct password - which gives access to the decoy volume but begins a hard erase of the real data in background. You'd write that on a piece of paper and keep it in your wallet for a hostile to find. I'd expect the FBI to be a bit too savvy to bumble into traps like that.
They only have to look at the code to see what algorithms are used. Breaking two ciphers takes twice as long as breaking one, which is more or less equivalent to adding one bit to the key - i.e. not much difference. Also, it doesn't have to be two different ciphers, it could be one encrypted twice using one cipher with two different keys.
Men don't have to be passionate about computers and programming to do well in our field.
You don't have to be "passionate" in the sense that you're gadget obsessed, can assemble PCs, operate Linux to keep it from falling over, and know all the latest Intel processor code names, but you do have to be self-motivated. If you haven't learned programming basics on your own before you sit down for your first college CS class you're in the wrong field. Similarly, if you haven't wired up simple circuits and understand what basic discrete components do before your first EE class it's almost certainly a waste of effort. If the purpose of the education isn't a means to understanding but merely the start of a career path, then this will lead to a very common pattern: 1) poor to mediocre engineer, 2) gets very little done, 3) plays politics and takes credit for others' work, 4) bails to management at first opportunity because it's "more interesting" than engineering. It takes time and concentrated effort to become a good programmer - a recent MS graduate is barely employable, and doesn't start to become strong until after about ten years' experience. In another ten they're solid. If they're not motivated to do this but just see it as a career they'll never last that long.
They don't need your retina. They just need whatever big integer your retina digests to.
In case the conclusion isn't obvious: if they can get you to authenticate using a compromised scanner you'll only be able to handle that breach exactly once - assuming you have a second eye.
Then you use your retina along with your fingerprint.
Sure identity theft is always going to be possible but it would be much harder if they had to get your retina than if they just had to memorize your digits and crack a password.
They don't need your retina. They just need whatever big integer your retina digests to.
Intel's low-power Atom is also RISC (the complex x86 instruction set is emulated using "micro-ops") and seems to do OK with power consumption.
Dynamic translation of a CISC instruction flow onto a pipeline adds complexity. The other big difference is that x86 has a fully coherent N-way associative cache which neither ARM or MIPS provide - at quite a bit of complexity. Heck, ARM prior to Cortex didn't even support unaligned memory access, and I'm not sure if MIPS does in any of their recent designs. Complexity requires die real estate. A bigger die is more expensive, and with more gates involved in just about every instruction fetch, decode, pipeline step, and memory bus cycle power consumption is going to be higher. Without the need to execute legacy code there's no way Intel can make it work with an inferior product at lower margin. They're going to shovel a few hundred million into the furnace to test the waters, then decide they can't make it work and exit. Undoubtedly they expect to lose money on every unit sold, to spend their way to a dominant market position where they can manage costs and eke out a margin through sheer volume, similar to their desktop and server processor business. I doubt management will have the guts to take the hemorrhaging and stick it out; it's a company used to seeing more or less instant returns from its other business.
No, it assumes memory for which all access are *at worst* a certain equal cost - in other words, memory for which accesses have an upper bound. Knuth's analysis still holds if certain pieces of memory are FASTER than the norm. If we take memory access going through swap as the normal, worst case, upper-bound cost of memory access, and RAM and cache hits as special better cases, the analysis still applies to the real world.
Well, both atan() and += are "operations". But clearly a loop of N atans, while of the same polynomial complexity, will be orders of magnitude slower than a loop of N additions. This is because += is O(1) while atan() is O(nlogn) for some accuracy n iirc. Similarly, there's a complexity difference between accessing L1/L2 cache and accessing main memory over a shared bus. The two will never converge or cross; there is no data set size where atan() will outperform +=. To say that the former is a worst-case cost estimate for the latter is only correct in a discrete mathematical sense, for all practical purposes it's meaningless. An algorithm that can rely on addition will always be superior to one relying on atans, even if of the same complexity. In fact, even an O(nlogn) addition-based algorithm will be superior to an O(n) atan based one - the two will converge at infinity (assuming I'm correct about the complexity of atan), so for any finite (real-world) data set the former will be faster.