No they cannot. French ISPs voluntarily elect to take the ICANN monitored root servers as authoratative. Furthermore, the "root servers" are not centrally run. They are owned and operated by several different companies (not the US government or ICANN). Those companies voluntarily agree to accept changes that result from ICANN decisions.
The US government has no direct or indirect control. It can, at most, stop giving money to ICANN and withdraw its blessing.
At the end of the day, the DNS is a cooperative venture among many private individuals.
People don't seem to realize that implementing the EU/UN plan would likely have required those governments to pass laws mandating and restricting which domain name servers could be used--whether this would even be "constitutional" in most of those countries is severely questionable.
It certainly is not constitutional in the United States.
And I was reacting to:
"I am waiting for the day where someone invents a computer that doesn't need to
be cooled or generate excess heat."
But to take your question on the surface, let me tell you that performance must come at the cost of energy efficiency at least so long as we are using semiconductors!
The reason for this is primarily that feature size scaling (as follows moore's law) does not imply clock frequency scaling at all (and I mean a normalized metric such as the rate at which a 5 inverter ring can oscillate, not issues relating to pipeline granularity, etc). Clock frequency scaling has been acheived by continually reducing what is known as the threshold voltage. Unfortunately, this means that linear improvements in clock frequency come at exponential costs in leakage effects.
This is the true 'red-brick wall' that has confronted the semiconductor industry. Moreover, this problem cannot be escaped merely by changing to a higher k dielectric as was done to combat the gate-oxide tunneling leakage. This is one is fundamental.
Thus, the turn toward multicore CPUs. The feature sizes are scaling down but we can't pay the costs of lowering Vt further. So we have to learn to do more with concurrency (which we can get from mere feature size scaling).
So yes, if we learn better and better concurrent approaches there is a way to dig ourselves out of the power hole
This is essentially impossible. Unless you consider so called "reversible computing". But reversible computing must be adiabatic, and thus very slow. Basically, as you slow a computation down you begin to approach ideal efficiency.
Fast computing is made possible by destroying information (that's all computers do really, they destroy information). That destruction process entails an entropy cost that must be paid in heat.
It's all about taxes. They want to levy taxes on domain registrations to pay for laying fiber in Africa.
Its all about money, money, money; also about sex because the clamoring for this really only got loud after ICANN approved the.sex domain.
And its being cloaked in stories about the evil dictatorial "government control" that now exists. There is no government control of Internet. The US government certainly does not control the DNS system--perhaps it does nominally, but right now the entire system is based on voluntary consent. People around the world are voluntarily deciding to use the ICANN monitored servers as the root.
What is so disgusting here is that these governments (including the EU) are attempting to abolish a voluntary system to institute something based on involuntary compulsion so that they can collect rent payments.
They are trying to claim they are just transfering a "power" that already exists but that's simply untrue.
Further, their desire to depose the IETF and give the ITU control over internet standards is also suspicious. First we might ask why? Then we might notice that China chairs the ITU. Then we might notice that the ITU has stated they want to introduce stronger point-of-origin guarantees to make it easier to track down individuals. Its obvious why they want this: you just need to watch the Chinese efforts to crack-down on dissent via the Internet.
Haha just a prototype, it will get smaller? Think again. We're not talking about lithography here. The nanotube structures are what they are. Density improvements will come only if they actually figure out how to make transistors and the like at the nanoscale level. Right now, they have effective bond out to traditional silicon technology for those functions. That process is costly in terms of area.
Mind you these densities are only been achieved in small quantities in labs. If we compare to traditional approaches, small quantities in labs are already being made with traditional silicon that have just as small if not smaller feature sizes.
So a better question is: "why bother"
Silicon/Lithography approaches are not within a factor of two of what is believed to be achievable with carbon nanotubes. Meanwhile carbon nanotubes have very low yield and very poor device reliability.
The reason the first nanotube application is shaping up to be memory is because that is only field in which we have well-developed ideas on dealing with these defects.
In closing: had this field matured 10 years ago, it would have been extremely important. Now it does not have much to offer. No one thought silicon would get to the nanoscale back then. It has.
Indeed! What we are really talking about here is that the UN/EU are interested in levying a tax on domain names to pay for network infrastrcture in who knows where.
Like hell that we're going to agree to use their root servers and pay their tax.
Actually this is untrue. If you have been following the recommendations of the FSF your license statement is setup to automatically upgrade to the new GPL, and so code already in use will be subject to the new terms.
No. No. NCQ is bad in a RAID setup. The reason is that NCQ means that the head is in unpredicable place. This means that on a RAID5, each write operation will complete in a time determined by the worst seek. Conversely, most RAID controllers can reorder their own write operations. Moreover, they usually have their own caches--and battery backed ones at that.
Therefore, NCQ tends to offer no performance advantages, needless cost, and can occasionally do screwy things.
I don't even think the mainline SATA raid controllers 3ware, LSI MegaRaid, support NCQ on their array...
Your single-point of failure analysis is bogus. RAID setups do have a single point of failure: the RAID card for instance. Its all about probabilities. Disk crashes happen to be likely, other failure modes much less so.
The thing is this comparison is extremely misleading.
Look, they averaged the costs overall production. They excluded development.
As a engineer ing this field let me tell you development costs are *huge*.
Moreover, some processors might cost $637 but those process cost a hell of a lot more than the average to manufacture... and that is the price you pay for the processors that come off the line with performance in the second or third standard deviation from the mean. There are not many of those processors (hard to make) + lots of demand => no shortages require high prices. The point being those those applications that can justify the cost are the ones that get the chip. This is about not wasting those chips on grandma's email computer while some scientist needs them--and to make that allocation in keeping with liberalism, i.e., without coercing people + corruption.
The old Bell System had something like this before it was smashed apart by overzealous anti-trust enforcement:
"If it isn't measured it doesn't get done. If it's not measured right, it doesn't get done right." -- Robert Gryb
It was called the Green Book, published monthly:
"It contained more that forty pages of charts and tables, recording everthing from job injuries to unkept installation appointments.
Each local company and each area within the companies (there were 95 areas nationwide) were compared and ranked in dozens of categories. An area was usually headed by a Level 6 manager who seized the Green Book the instant it came in to see how his area ranked in the overall summary and to see if any performance category had slipped.
Even dropping... to 94[%] was considered a serious problem. These were labeled weakspots though in truth they could be career threatening crises. It was believed that performances below 90 were noticeable to the public and led to customer complaints."
"I would not call the forensic quality write-blockers on the market "a kludge". They perform a basic role to a level that is accepted by the highest courts and experts (the real ones). They are very simple, yet vital. They go a long way to preventing human error."
Being fairly well versed in the SATA specifications, I can tell you that having such a device in the middle that behaves as that device does, is not a part of the specification. Therefore I labeled it a kludge. I concede that might be to make a nit-picky point that plays on the emotional impact of kludge to discredit something which is merely absent from the specification rather than prohibited by it.
"The software in question is extremely complex and has to be driven by an error prone human. The write-blocker on the other hand, is a very simple device dedicated for one thing and is simply plugged into the drive to be captured."
I'm sure it seems that way, but in reality it some hardware running a program stepping through a state-machine and acting man-in-the-middle to filter the requests.
I agree it is dedicated. Still I think the Honda analogy might apply. Honda sells inexpensive, reliable cars in large volumes. Hondas last, Hondas perform as expected, etc. They do these things much better than competing products because their economies of scale can command the commitment of substantial engineering services. A car such as a Jaguar would have to cost 10x as much as its already inflated price + sell at the same volume as it does now to pay for the engineering that Honda can get for the accord.
These devices might be very simple, but in terms of relevant man-hours that doesn't make them better designed.
I think its a very difficult calculus without a certain result a priori to determine which actually has the greater risk, if you had to select one method over the other.
Of course, I'm sure we can agree that the *right* answer is that you should do it both ways.
The mere use of economic terminology does not for a sound argument make. Without even evaluating the piece for logical fallacies, lets deal with the material fallacy: the proposition that BSD Licenses results in less code being contributed back under a BSD license (to the OSS community).
Proof please?
This is especially ironic in an article that is about commercial entities contributing development time (code) ultimately under the BSD license to enhance an existing BSD project.
speaking of apples and oranges lets get something straight here. you're right 65nm says something about the gate length, but the measurements in the nanoscale technology also only covers the gate length. So while 65nm does not say much about entire area cost of a single transistor, nor does their 10s of nanometers say much about the entire area cost of a single nanotube transistor.
At the end of the day--even the article--admits that nanotechnology is only about a factor of two density improvement, but at the moment that is a useless factor because of the severe reliability issues and the failure of anyone to ever make something with the number of components (whether or not they work) comparable to a modern CPU.
And though they mention that the future nanotube devices will be smaller--so will silicon.
A P4 already incurs approximately a month of manufacturing from purification to bonding--that is with a manufacturing process that runs continuously 24 hours a day. Much of the process is highly parallel (lithography, oxide growth, implantation all occur over the entire wafer at once). Meanwhile... nanotubes are assembled in a mostly serial fashion.
Anyone seriously entertaining the idea that nanotubes are viable replacements for silicon techiques for CPUs from an engineering standpoint is seriously lacking awareness of the economics.
Only if you have a plan for incrementally changing FreeBSD. I don't think it likely that they'll accept sweeping changes from someone without a reputation yet.
One of the consequence of having a reputation for stability is a reluctance to betray the user community with inadequately reviewed ideas.
Equally important: he advocated "hands-off" in part because of businesses' tendancy to influence government in to doing the wrong things. Remember: Adam Smith's target was mercantialism which as a policy was in fact that darling of merchants among others.
DARPA is one of the principle funding sources for cold fusion research. Though this amounts to only about 25M (order of magnitude smaller than other US gov funding for hot fusion).
More anything, it's the academic community generally, the NSF, etc that ridicules this work, not the government per se.
And ironically... having airbags and not wearing a seat-belt is more deadly than just not wearing a seat-belt (because not only do you fly-forward but now you have a rapidly expanding explosion as well).
That may well be, but we gave lots of people aid--not just Israel.
Top three USAID recipients (2001): Israel 2.8B--a good chunk of which is loans Egypt 1.9B Jordan.2B
You're engaging in pure speculation as regards the consequences of the Iraq war. Hussein was a war criminal who slaughtered millions--he had to go. This preoccupation with present possession of WMD was wrong when Bush first mentioned it, and it is wrong now. The fact remains that the man was a long-term threat reviled by the citizens of Iraq. Perhaps you should remember that the headquarters of the Iraqi Olympic committee was in fact a front for a large torture facility?
Sorry, I can't help but think that we did the moral thing in Iraq--even if some people opted to do it for the wrong reasons.
I am not of the opinion that America stands for "unfettered majority rule". We've never accepted the idea that a society can pursue any illiberal policies it wishes. No, we're a fundamentally liberal country--albeit in the British, not the French sense of the word, and as a consequence can you really justify turning a blind eye to authoritarianism around the world?
I agree, we're being attacked because we're in the way, but it's because we're standing in the way of authoritarianism.
No, I believe in doing both. I firewall and then NAT. This creates a security onion.
But this was an article about IP address space, and I hold to the principle that in general devices should not be addressable--even if their software is well engineered, the onion principle again tells you that you shouldn't implement needless functionality.
Still, I agree, it depends on the device.
I'd *HATE* it if my ISP NATed me instead of giving me at least one real address.
Nono, I understand the difference. I merely said that after we NATed we realized that we didn't have to make everything publicly accessible and that making not things addressable was also reasonable.
Remember: we're talking about the security onion here.
It remains though that at home, I'm quite content to have one public machine and lots of NATed clients. In fact I did so despite being entitled to more IP addresses from my ISP. Further, I do so on top of running a firewall in front of the NAT.
Substantial numbers of IPs are used in managed computing environments, classrooms, labs, kiosks, etc.
Anyways, you can't implement a firewall in a non-evil way with FTP either. FTP is just a broken protocol.
NAT has many flaws, but it does 90% of the job very cost effectively. It's simple cost-benefit analysis.
Slashdot Mirror
No they cannot. French ISPs voluntarily elect to take the ICANN monitored root servers as authoratative. Furthermore, the "root servers" are not centrally run. They are owned and operated by several different companies (not the US government or ICANN). Those companies voluntarily agree to accept changes that result from ICANN decisions.
The US government has no direct or indirect control. It can, at most, stop giving money to ICANN and withdraw its blessing.
At the end of the day, the DNS is a cooperative venture among many private individuals.
People don't seem to realize that implementing the EU/UN plan would likely have required those governments to pass laws mandating and restricting which domain name servers could be used--whether this would even be "constitutional" in most of those countries is severely questionable.
It certainly is not constitutional in the United States.
And I was reacting to:
"I am waiting for the day where someone invents a computer that doesn't need to
be cooled or generate excess heat."
But to take your question on the surface, let me tell you that performance must come at the cost of energy efficiency at least so long as we are using semiconductors!
The reason for this is primarily that feature size scaling (as follows moore's law) does not imply clock frequency scaling at all (and I mean a normalized metric such as the rate at which a 5 inverter ring can oscillate, not issues relating to pipeline granularity, etc). Clock frequency scaling has been acheived by continually reducing what is known as the threshold voltage. Unfortunately, this means that linear improvements in clock frequency come at exponential costs in leakage effects.
This is the true 'red-brick wall' that has confronted the semiconductor industry. Moreover, this problem cannot be escaped merely by changing to a higher k dielectric as was done to combat the gate-oxide tunneling leakage. This is one is fundamental.
Thus, the turn toward multicore CPUs. The feature sizes are scaling down but we can't pay the costs of lowering Vt further. So we have to learn to do more with concurrency (which we can get from mere feature size scaling).
So yes, if we learn better and better concurrent approaches there is a way to dig ourselves out of the power hole
This is essentially impossible. Unless you consider so called "reversible computing". But reversible computing must be adiabatic, and thus very slow. Basically, as you slow a computation down you begin to approach ideal efficiency.
See http://en.wikipedia.org/wiki/Reversible_computing
Fast computing is made possible by destroying information (that's all computers do really, they destroy information). That destruction process entails an entropy cost that must be paid in heat.
A little embellishment for taste, but overall the point is right:
Java is "write once, run nowhere"
It's all about taxes. They want to levy taxes on domain registrations to pay for laying fiber in Africa.
.sex domain.
Its all about money, money, money; also about sex because the clamoring for this really only got loud after ICANN approved the
And its being cloaked in stories about the evil dictatorial "government control" that now exists. There is no government control of Internet. The US government certainly does not control the DNS system--perhaps it does nominally, but right now the entire system is based on voluntary consent. People around the world are voluntarily deciding to use the ICANN monitored servers as the root.
What is so disgusting here is that these governments (including the EU) are attempting to abolish a voluntary system to institute something based on involuntary compulsion so that they can collect rent payments.
They are trying to claim they are just transfering a "power" that already exists but that's simply untrue.
Further, their desire to depose the IETF and give the ITU control over internet standards is also suspicious. First we might ask why? Then we might notice that China chairs the ITU. Then we might notice that the ITU has stated they want to introduce stronger point-of-origin guarantees to make it easier to track down individuals. Its obvious why they want this: you just need to watch the Chinese efforts to crack-down on dissent via the Internet.
Haha just a prototype, it will get smaller? Think again. We're not talking about lithography here. The nanotube structures are what they are. Density improvements will come only if they actually figure out how to make transistors and the like at the nanoscale level. Right now, they have effective bond out to traditional silicon technology for those functions. That process is costly in terms of area.
Mind you these densities are only been achieved in small quantities in labs. If we compare to traditional approaches, small quantities in labs are already being made with traditional silicon that have just as small if not smaller feature sizes.
So a better question is: "why bother"
Silicon/Lithography approaches are not within a factor of two of what is believed to be achievable with carbon nanotubes. Meanwhile carbon nanotubes have very low yield and very poor device reliability.
The reason the first nanotube application is shaping up to be memory is because that is only field in which we have well-developed ideas on dealing with these defects.
In closing: had this field matured 10 years ago, it would have been extremely important. Now it does not have much to offer. No one thought silicon would get to the nanoscale back then. It has.
Indeed! What we are really talking about here is that the UN/EU are interested in levying a tax on domain names to pay for network infrastrcture in who knows where.
Like hell that we're going to agree to use their root servers and pay their tax.
Actually this is untrue. If you have been following the recommendations of the FSF your license statement is setup to automatically upgrade to the new GPL, and so code already in use will be subject to the new terms.
No. No. NCQ is bad in a RAID setup. The reason is that NCQ means that the head is in unpredicable place. This means that on a RAID5, each write operation will complete in a time determined by the worst seek. Conversely, most RAID controllers can reorder their own write operations. Moreover, they usually have their own caches--and battery backed ones at that.
Therefore, NCQ tends to offer no performance advantages, needless cost, and can occasionally do screwy things.
I don't even think the mainline SATA raid controllers 3ware, LSI MegaRaid, support NCQ on their array...
Your single-point of failure analysis is bogus. RAID setups do have a single point of failure: the RAID card for instance. Its all about probabilities. Disk crashes happen to be likely, other failure modes much less so.
It would be more precise to say that it is based on Mach code and BSD code--especially the FreeBSD userland.
However, substantial work went in during the NeXT era and subsequently from Apple directly.
This is a minor quibble with the meaning of "based" as you use it.
OSX is decidedly not a BSD variant though--most notably it is not developed under a BSD license.
As a point of fact, Windows is BSD based because microsoft forked their tcp stack off an early BSD one...
But we can still distinguish it as being in another class entirely.
You can read an article on his denial here: http://www.planetark.com/dailynewsstory.cfm/newsid /32513/story.htm
The thing is this comparison is extremely misleading.
Look, they averaged the costs overall production. They excluded development.
As a engineer ing this field let me tell you development costs are *huge*.
Moreover, some processors might cost $637 but those process cost a hell of a lot more than the average to manufacture... and that is the price you pay for the processors that come off the line with performance in the second or third standard deviation from the mean. There are not many of those processors (hard to make) + lots of demand => no shortages require high prices. The point being those those applications that can justify the cost are the ones that get the chip. This is about not wasting those chips on grandma's email computer while some scientist needs them--and to make that allocation in keeping with liberalism, i.e., without coercing people + corruption.
Anyone who was moved by this article should read
"Economic Calculation In The Socialist Commonwealth"
http://www.mises.org/econcalc/econcalc.pdf
The old Bell System had something like this before it was smashed apart by overzealous anti-trust enforcement:
"If it isn't measured it doesn't get done. If it's not measured right, it doesn't get done right." -- Robert Gryb
It was called the Green Book, published monthly:
"It contained more that forty pages of charts and tables, recording everthing from job injuries to unkept installation appointments.
Each local company and each area within the companies (there were 95 areas nationwide) were compared and ranked in dozens of categories. An area was usually headed by a Level 6 manager who seized the Green Book the instant it came in to see how his area ranked in the overall summary and to see if any performance category had slipped.
Even dropping... to 94[%] was considered a serious problem. These were labeled weakspots though in truth they could be career threatening crises. It was believed that performances below 90 were noticeable to the public and led to customer complaints."
"I would not call the forensic quality write-blockers on the market "a kludge". They perform a basic role to a level that is accepted by the highest courts and experts (the real ones). They are very simple, yet vital. They go a long way to preventing human error."
Being fairly well versed in the SATA specifications, I can tell you that having such a device in the middle that behaves as that device does, is not a part of the specification. Therefore I labeled it a kludge. I concede that might be to make a nit-picky point that plays on the emotional impact of kludge to discredit something which is merely absent from the specification rather than prohibited by it.
"The software in question is extremely complex and has to be driven by an error prone human. The write-blocker on the other hand, is a very simple device dedicated for one thing and is simply plugged into the drive to be captured."
I'm sure it seems that way, but in reality it some hardware running a program stepping through a state-machine and acting man-in-the-middle to filter the requests.
I agree it is dedicated. Still I think the Honda analogy might apply. Honda sells inexpensive, reliable cars in large volumes. Hondas last, Hondas perform as expected, etc. They do these things much better than competing products because their economies of scale can command the commitment of substantial engineering services. A car such as a Jaguar would have to cost 10x as much as its already inflated price + sell at the same volume as it does now to pay for the engineering that Honda can get for the accord.
These devices might be very simple, but in terms of relevant man-hours that doesn't make them better designed.
I think its a very difficult calculus without a certain result a priori to determine which actually has the greater risk, if you had to select one method over the other.
Of course, I'm sure we can agree that the *right* answer is that you should do it both ways.
This says much more about the courts and peoples inability to offer the right explanation at the right time.
That is there is a clear answer at "Then WHY did you not use it?!"
"Because such a device is Kludge. It is a black-box that cannot be verified and as such as is no better than the "black-box" of the operating system.
Moreover, the latter is used and effectively tested by millions whereas only a handful of people purchase such "write blockers"
Ultimately such a device merely relocates the nexus of trust and fails to actually improve the surety of the evidence."
So says the expert with confidence.
The mere use of economic terminology does not for a sound argument make. Without even evaluating the piece for logical fallacies, lets deal with the material fallacy: the proposition that BSD Licenses results in less code being contributed back under a BSD license (to the OSS community).
Proof please?
This is especially ironic in an article that is about commercial entities contributing development time (code) ultimately under the BSD license to enhance an existing BSD project.
speaking of apples and oranges lets get something straight here. you're right 65nm says something about the gate length, but the measurements in the nanoscale technology also only covers the gate length. So while 65nm does not say much about entire area cost of a single transistor, nor does their 10s of nanometers say much about the entire area cost of a single nanotube transistor.
At the end of the day--even the article--admits that nanotechnology is only about a factor of two density improvement, but at the moment that is a useless factor because of the severe reliability issues and the failure of anyone to ever make something with the number of components (whether or not they work) comparable to a modern CPU.
And though they mention that the future nanotube devices will be smaller--so will silicon.
A P4 already incurs approximately a month of manufacturing from purification to bonding--that is with a manufacturing process that runs continuously 24 hours a day. Much of the process is highly parallel (lithography, oxide growth, implantation all occur over the entire wafer at once). Meanwhile... nanotubes are assembled in a mostly serial fashion.
Anyone seriously entertaining the idea that nanotubes are viable replacements for silicon techiques for CPUs from an engineering standpoint is seriously lacking awareness of the economics.
Only if you have a plan for incrementally changing FreeBSD. I don't think it likely that they'll accept sweeping changes from someone without a reputation yet.
One of the consequence of having a reputation for stability is a reluctance to betray the user community with inadequately reviewed ideas.
Equally important: he advocated "hands-off" in part because of businesses' tendancy to influence government in to doing the wrong things. Remember: Adam Smith's target was mercantialism which as a policy was in fact that darling of merchants among others.
DARPA is one of the principle funding sources for cold fusion research. Though this amounts to only about 25M (order of magnitude smaller than other US gov funding for hot fusion).
More anything, it's the academic community generally, the NSF, etc that ridicules this work, not the government per se.
"more emphasis on (mathematics) basics"
See this wonderful essay by the famous mathematician V.I. Arnold
He explains how mathematicians' fetish for abstract formalism has negatively impacted the teaching of mathematics (specifically in France).
And ironically... having airbags and not wearing a seat-belt is more deadly than just not wearing a seat-belt (because not only do you fly-forward but now you have a rapidly expanding explosion as well).
Airbags can only be used safely with seat-belts.
That may well be, but we gave lots of people aid--not just Israel.
.2B
Top three USAID recipients (2001):
Israel 2.8B--a good chunk of which is loans
Egypt 1.9B
Jordan
You're engaging in pure speculation as regards the consequences of the Iraq war. Hussein was a war criminal who slaughtered millions--he had to go. This preoccupation with present possession of WMD was wrong when Bush first mentioned it, and it is wrong now. The fact remains that the man was a long-term threat reviled by the citizens of Iraq. Perhaps you should remember that the headquarters of the Iraqi Olympic committee was in fact a front for a large torture facility?
Sorry, I can't help but think that we did the moral thing in Iraq--even if some people opted to do it for the wrong reasons.
I am not of the opinion that America stands for "unfettered majority rule". We've never accepted the idea that a society can pursue any illiberal policies it wishes. No, we're a fundamentally liberal country--albeit in the British, not the French sense of the word, and as a consequence can you really justify turning a blind eye to authoritarianism around the world?
I agree, we're being attacked because we're in the way, but it's because we're standing in the way of authoritarianism.
No, I believe in doing both. I firewall and then NAT. This creates a security onion.
But this was an article about IP address space, and I hold to the principle that in general devices should not be addressable--even if their software is well engineered, the onion principle again tells you that you shouldn't implement needless functionality.
Still, I agree, it depends on the device.
I'd *HATE* it if my ISP NATed me instead of giving me at least one real address.
Nono, I understand the difference. I merely said that after we NATed we realized that we didn't have to make everything publicly accessible and that making not things addressable was also reasonable.
Remember: we're talking about the security onion here.
It remains though that at home, I'm quite content to have one public machine and lots of NATed clients. In fact I did so despite being entitled to more IP addresses from my ISP. Further, I do so on top of running a firewall in front of the NAT.
Substantial numbers of IPs are used in managed computing environments, classrooms, labs, kiosks, etc.
Anyways, you can't implement a firewall in a non-evil way with FTP either. FTP is just a broken protocol.
NAT has many flaws, but it does 90% of the job very cost effectively. It's simple cost-benefit analysis.