No, oh +5 Insightful but painfully incorrect one. There is no such law in the US that requires H1B visa applications to be denied even if there are many Americans capable of doing the job.
You certainly wouldn't have to prove it anyway, under any conceivable sensible legislation, since it's hard to prove a negative, and an exhaustive analysis of all 230 million + Americans would be rather cost-prohibitive.
What keeps the US from being "flooded (more so than now) with techs and doctors from all over asia" are the limits placed on the number of H1B visas granted, not any sort of rules for being eligible for an H1B visa.
(And yes, I do expect you to enumerate vulnerabilities that don't exist with gool ol' fashioned tumbler locks -- you did say "unique". Oh, and these should outweigh the cost and time savings afforded by instant reprogrammability too, of course.)
Given the overwhelming rate of adoption of these card-key locks by hotels, whether or not the advantages are "obvious" or not seems less relevant than the fact that the advantages exist.
The link you gave led to an article with the headline "Dems Threaten Suit in Florida as More Purported Ballot Boxes Turn Up". In the article itself, purported is further qualified as "what look to be five purported ballot boxes." One of the things which looked to be a purported ballot box turned into a dead end since it "apparently contained notebooks."
If it was "all over the news" at the time, surely you can come up with something a little more solid than something that looks to be an article purporting to describe some alleged missing boxes which in fact contained notebooks? All I remember in the news at the time was a lot of empty accusation, speculation, and whining.
Perhaps you can enlighten me with a nice, solid, conclusive link, especially given all the time that has elapsed to allow full investigations about anything which allegedly may slightly resemble a purported missing ballot box. Or are those investigations more like, oh, I don't know, OJ looking for the "real" killer?
Wait! Let's not dive into the good ideas just yet. First, someone needs to point out that the article author is a little confused on several key issues.
Got a hotel with perfectly good door locks and metal keys? Rip them out and replace them with computerized locks and swipe-cards.
There is nothing "perfectly good" about a lock whose keying needs to be changed every few days for liability and safety purposes. On-staff locksmitch or programmable locks? Hmmm.
These computer professionals say that accurately counted free elections are the bedrock of democracy. Voting, they claim, is too important to be done on a computer. The irony is delicious--it's sort of like group of doctors arguing for the return of leeches because the President of the United States is too important to be treated by modern medicine.
Oh boy. Even if this analogy were suitable (it isn't, obviously), there is still no irony here.
Because the voting is done on a large touch screen, they can use big fonts that are easier for the elderly to read.
Eh? How, exactly, is it easier to print big fonts on a screen than a piece of paper? I think the cost of paper varies less strongly with size than, say CRT and LCD technology.
They can even confirm the voter's choices on a second screen--which means that there would be no more elderly Jewish voters in Palm Beach accidentally casting their ballots for Pat Buchanan.
Oyve. Wow, two screens. Well, that sure represents carved-in-stone untemperable data to me. Regardless of the fact that the data could still be sitting in memory, not recorded permanently, and still quite subject to fraud or incompetence.
The [trojan or back-door] logic could be so well hidden that not even a careful review of the machine's source code would find it. This isn't as far-fetched as it might sound: Unauthorized features called "Easter eggs" are routinely hidden in commercial software, even software shipped by Microsoft.
(Emphasis mine). Bullshit! Careful review of source code finds as much as it wants to. And the example of "Easter eggs" in MS software is inappropriate since MS doesn't release source code.
paper is a fundamentally bad way of making and keeping accurate records. Paper is bulky and heavy. It can be hard to read something recorded on paper, no matter whether the marks were made by hand with pen-and-ink or by a computerized printer. Paper rips and gets jammed in machines. Paper dust gets everywhere. Eliminating paper, Selker explained to me, has the potential for dramatically improving elections.
WTF? And computers are less buggy than paper?!?! Help me.
"But what about all of the ways that you can hack the voting machines?" I asked him.
Selker laughed. Politicians, he told me, have been hacking elections in America for more than 200 years.
Hahah, haha. Good pun. Now, seriously, what about all the ways that you can hack the voting machines?
thousands of Democrats, many of them minorities, showed up at voting places and discovered that they were no longer registered. Why? Because it's illegal for convicted felons to vote unless that right is specifically restored. Florida had recently purged the voting roles against a computerized database of convicted felons; tens of thousands of people were removed, some apparently in error.
Oh no, the felons couldn't vote. Whatever shall we do? Jeebus, I think I know the case in question, and the "some apparently in error" were 2 people with repeatedly rejected appeals. Not pending appeals mind you, flat-out rejections for appeal -- though apparently the felons claimed that was unfair. this is not the sort of election hacking that worries me.
Other techniques for stealing an election, Selker told me, are stationing tow trucks outside the polls to intimidate voters; setting up po
We've gone from following the rules to playing the odds.
And if we do follow the rules and don't play the odds, then we are figured to be suckers.
Speak for yourself. Some of us just keep doing the right thing as best we can, no matter what everyone else is doing. We usually come out better in the long run.
You are probably right that hardware "glitches" (defined here as a temporary, non-fatal, non-locking error, just for clarity's sake) happen less often than software failures. My guess is that it's about one order of magnitude (10x). Your guess is more. I know people who adamantly argue that it's much less. Unfortunately for all of us, since there is rarely any way to recognize that an hardware glitch has occured, we can't really know for sure. How can you ever know for sure if the hardware retains no evidence of its "glitch"?
[Hardware errors] are typically of the form, "this program doesn't work at all", not of the form "This program works but under some circumstances, with just the right conditions, it finishes successfully but with the wrong result." If a hardware glitch flips a bit from 1 to 0, for example, the chances that the program will still run anyway but with just the wrong total is very small. Alter a random bit in the program and it's highly likely to cause it to die altogether.
I think you missed my attempt at distinguishing harware failure (which is what you seem to be describing) and a hardware "glitch" (as defined above). I assert that something on the order of 10% of all "crashes" are due to non-fatal (in that the chip doesn't fry), temporary, self-recovering chip failures, and the remaining 90% are software. More than most people acknowledge. Unfortunately, as explained above, there's no way to know how right or wrong either of us are (unless all software flawlessly trapped all errors with appropriate logging and reporting, which would make the question moot were it true).
Also, the failures aren't all 0->1 or 1->0 bitflips -- that really only applies to storage elements like RAM and flip flops. We have ways of making that redundant and really robust (starting way back when with parity, up to modern 3-bit ECC and RAM layouts designed to defend against cosmic radiation). It's the other stuff (interconnect resistive shorts, race conditions, excessive leakage, non-fatal punch-thorough, unexpected EMI, system-wide signal integrity, local hot-spots, On-Chip Variation, etc.) that we can't try to correct when they fail which cause the most problems these days.
Also, I think you usually have more data than executable code in memory, so it seems that a random bitflip in your RAM (assuming no ECC) would affect data, not program, and therefore (probably) not be fatal. Moreover, lots of opcodes with similar arguments have 1-bit difference, so you'd not necessarily get a coredump on changing "branch if greater than" to "branch if less than", would you?
Regarding knowing when a piece of hardware (chips, mostly) cause a "glitch": AFAIK, there is exactly one company trying to remedy this. Sun Microsystems. "RAS" (Reliability, Acessibility, Scalability, in that order) is the mantra at Sun, and the next generation stuff is insanely self-testable, to the point that the testability sometimes costs more dollars and time than the normal functionality.
Even with this expense, Sun is not even aiming at 100% -- it's impossible with the current knowledge base. Someone has to come up with some amazing new way of thinking as significant as Newtonian physics or relativity theory in order to change that. Until then, you just keep thinking it's all a software problem (except for obvious, permanent hardware failures), and I'll keep laughing when the programmers get flak for my bug:)
Has (or why hasn't) Apple considered developing a DirectX-compatible graphics API, or OpenGL wrapper, or similar? Seems to me that even moderate compatibility with the most popular, most heavily-developed (and by many accounts pretty good these days) graphic subsystem (DirectX 8+) would go a long way toward helping revive the flailing mac gaming market, which would be a good thing.
I, for one, would certainly love to be able to frag some of my mac-using buddies in Battlefield 1942 (though I'm afraid to consider the cost of a max that could run bf1942 at the resolution and detail settings I'm accustomed to (1280x1024x32bit, all options at max on an XP2400/1GB/GeFX5900U).
While legacy control systems are often UNIX-based ("Control-Alt-Delete scares power plant operators," Ahern said) and thus immune to MS worms and virii, their 10-megabit networking technologies can easily be overwhelmed. "Even the load from leading intrusion detection and monitoring systems can create a denial of service and shut these plants down," Ahern said.
It never ceases to amaze me how so many otherwise technically savvy people buy into the notion that hardware is infallible and software is always to blame for "glitches" or "crashes." I would concede that software causes more problems, but the notion that "computers don't make mistakes" is remarkably wrong.
No chip ever made is tested 100%. Test coverage of 99.0% is considered excellent in the ASIC and custom IC design worlds. Many go to fab with less than 95%, sometimes 90% test coverage. So you have 10 million gates on a hunk of stuff you grew and screenprinted with toxic chemicals, with a decent plan to make sure that 99.0% of them can be tested to work as they should. You do the math -- lots will fail, and worse, some will fail occasionally and then resume working. Moreover, we can't really test all posible sets of stimulus -- that would take an incredibly long time in an industry where tester time on billion-dollar testers is doled out in 5s increments (30s is considered unworkable by most fabs, and would still allow us to test less than 1% of all possible combinations of inputs and transitions).
The interconnect between chips is another problem that's hard to measure, but non-zero. Same with passive components (capacitors, resistors) -- they have non-zero non-fatal failure rates. Which is an obfuscatory way of saying they can "glitch" or "crash". Thank Ohm a resistor's reboot time is much faster than Windows or you'd really notice the hardware failures:)
I don't have time to go into system-wide signal integrity (intractable), fault-tolerance that isn't, metastability, radiative interference such as cosmic rays and alpha particles emitted from local metals, etc. There's a lot that can and does go wrong in hardware.
I'm really kind of reluctant to post this, since as a hardware designer it's cool that I never hear the "you're why my computer crashes" comments that my software engineers suffer. It's also fun to se MS take the brunt of most PC users reliability complaints. In truth, they probably deserve a lot of it, but not the 100% most believe -- hardware does sometimes fail for a microsecond and then recover nonchalantly, as if nothing happened, sort of like when a cat trips or runs into a wall.
I've googled my heart out on this one to no avail. Can you please give me some link or keywords to help me find more on this (names maybe)? I would love to have this ammo (if reliable) for my next discussion on the subject with a certain friend.
BTW, I'm not looking to be anti-organic. In general, most of the (fresh) organic fruits and veggies just look better to me, so I don't mind the small premium required to get them. However, some are of the bent that all organic is good and all non-organic is bad, and it's always handy to be able to point out at least one case that breaks the rule to instill a little healthy skepticism.
First, thank you for posting that. It truly was informative. I had no idea, and I've been paying attention to stories about how whacked the Recording Industry is.
Second, wow -- that seems as contrived, arbitrary, and complicated (note: not complex, complicated, the distinction is important here) as the US Income Tax Code.
To get the school's message across, all students were asked to sign a document confirming that their computers were updated with all the needed security upgrades. Not enough students confirmed that their machines were updated, prompting the GMU action today. Administrators said they would try later today to reconnect dorms, weeding out students with infected PCs. Students living off campus can continue to dial in to the campus computer network.
Looks like the kids are getting a decent deal on virus-removal and system updates too:
Students are being charged $30 if a university technician is called in to clean an infected machine, a school spokesman said. Students can go to off-campus experts for a fix but must certify that their computers are updated with the latest security fixes before being allowed to access the campus network.
Hmph, I can't find anything wrong here. Of course, there are a couple of choice quotes from the kids who, I believe, are our future:
Kimberly Borchert, a 19-year-old sophomore, said her computer "freaked out" as soon as she plugged it into the school's network last week.
Freshman Andrew Canose was one of several GMU students who encountered problems after installing the university-provided anti-virus software. Canose found the new program conflicted with an older anti-virus program already on his computer. "My computer is like at war with itself and won't work," he said.
But my favorite lines are from the admins, such as this gem:
"I think we really need to groom a new type of student who is responsible for their computer security," said Kathy Gillette, manager of George Mason University's beleaguered tech support center. "A lot of them lived at home and mom or dad took care of the computer so they've never learned how to fix them, but hopefully we'll be able to teach them that too."
And the classic:
"There were a certain percentage of students that wouldn't listen to us unless we hit them upside the head with a lockout," he said. "You simply can't deal with these problems until you've got your network under control."
I could imagine doing it for a lot less money, since you don't need the PVR features or the channel guide, etc. Just a webcam and a cheap PC with lots of HD space would do the trick just as easily (perhaps easier) and much more cheaply.
I'll go out on a limb and say no one will ever buy this to hook it up to a security camera.
Did you RTFA? I don't think so, since it's not in any way about developing a new processor architecture. Rather, it's about developing a new type of physical switch to replace a silicon transistor. If they got it to work, they could make an x86 or any other type of processor out of it -- but that's really irrelevant this early in the game.
"electrons jumping paths"? You're not in the semiconductor industry are you?:) Kindly accept the following reality check:
1. Double die size? Heat's not the problem (bigger die surface area makes it easier to dissipate heat, not harder, assuming same transistor density and switching rate). Manufacturability (and thus price) is the problem. Currently, the largest silicon die anyone can make at a price anyone will pay is just under 20mm on edge (400mm^2). Yield for such dice is around 20-30% (so, 70-80% of the dice made will be bad, and simply thrown away, but paid for in the price of the good dice). Doubling the die size to 40mm would cause an exponential increase in defect rates, resulting in yield around 0.01%, and prices around 1000x current.
2. Optical-based chipsets? I'm not sure what this means. If you mean optical processors (using photons instead of electrons) within a chip, that doesn't exist yet. If you mean inter-chip signaling using optics, this does exist (MEMS), but it doesn't help much to have your board-level interconnect have infinite bandwidth if your processors can't process it. Which they can't.
3. OK, efficiency is good, but hyperthreading is not really an example of increasing efficiency. A hyperthreaded chip is still doing the same amount of processing per gate, so it's not really any more efficient, it's just that some of the logic has been duplicated on chip to allow doing some operation twice (or more) at once. We can't just hyperthread everything, since adding more (duplicate copies) of logic requires a larger die to hold it all, which leads to the problems explained above in #1.
And the current crop of chips using copper interconnects propigate at maybe 1/2 or 2/3 that speed, meaning in a theoretical 1,000GHz chip (not too far off by Moore's law), electrical impulses can only travel about 50um.
Moore's Law says nothing about operating frequency. As pointed out already, Moore's law refers to the total number of components (switches) that can be fit on a die.
Checking a layout I'm working on now (0.13um all-Cu process with 0.28um wire pitch), I see that I have a 4mm wire with a nice 4x driver, and the delay on that wire is 182ps (MIN) to 423ps (MAX). So, MIN case (fastest process), that's about 45ps/mm, or 0.22mm/ps. 1000GHz gives a 1ps period, so I can move a signal a distance of about 0.22mm, or 220um, in the best case. Worst case (MAX process) I can do about 100um/ps. Your 50um is only off by a factor of 2-4 (not bad), but please note that this is today's process, and I could make it faster still (stronger driver, wider wire) if I wanted (needed) to. And, frankly, even 50um is plenty of distance to do a lot of work. Not really all that daunting, even now.
Finally, keep in mind that not every signal path in a chip has to make its destination in 1 clock cycle. Some are what we call "multicycle paths", and the system is designed to allow for 2 (or 3, or whatever) clocks to tick before using certain signal levels. Long-distance interconnect are usually MCP.
Er, there is no current process for making the stuff the article is about, other than some poor guy on his knees begging a CVD reactor to work right. The $3B mentioned is the cost for building a fab for a modern semicondoctor process which, while certainly a lot, is not a real problem since there are quite a few companies prepared to make such investements given the juicy returns (IBM, NEC, LSI, TSMC, . . . ).
I can't even squeeze 50MPG out of my V6 Jetta (which is several hundred pounds lighter than a Camry), even if I drive like a grandma afraid of acelleration. I usually get 20-30 in real life, according to the Jetta's computer thing, but I love to acellerate. My wife gets a litter better, but not 50.
Great idea! Except that, in general, cities wouldn't like that. Parking meters are a source of income (from the tickets issues when the expire), not a distribution system to be made more efficient. Why do you think it's illegal in many places to feed a meter that's not your own?
For future reference, here's how I could tell: (1) you created a false dichotomy, (2) the false dichotomy is based on nothing at all, not even the usual anecdotal evidence presented in such case and (3) you assail the war in Iraq in a silly way.
Me personally? A lot. Like 3-4 emails per day are covered by NDA, and the lawyers make us put those silly little sigs that do nothing. Oh, and we use Thawte certificates to encrypt and sign every one of them (well, even the non-NDA ones usually, out of habit).
In the ASIC design industry, I'd say this is pretty common. All discussions between vendor and customer are under NDA, sometimes even before any discussions at all happen there are multiple NDAs in place. And lots of the business (even exchange of IP in source code attachments) takes place by email.
Slashdot Mirror
What should I call you? "In-need of a logic or statistics class?"
Correlation does not imply causality.
Causality under certain conditions does not necessarily lead to the same results in all conditions.
Your logic is sorely flawed.
No, oh +5 Insightful but painfully incorrect one. There is no such law in the US that requires H1B visa applications to be denied even if there are many Americans capable of doing the job.
You certainly wouldn't have to prove it anyway, under any conceivable sensible legislation, since it's hard to prove a negative, and an exhaustive analysis of all 230 million + Americans would be rather cost-prohibitive.
What keeps the US from being "flooded (more so than now) with techs and doctors from all over asia" are the limits placed on the number of H1B visas granted, not any sort of rules for being eligible for an H1B visa.
Not "snake oil." That means it doesn't work. How about "a product"?
remember, without copyright law, there can be no GPL
You misspelled "need."
Hmm, I didn't know evil was the opposite of confidential. Interesting.
Such as?
(And yes, I do expect you to enumerate vulnerabilities that don't exist with gool ol' fashioned tumbler locks -- you did say "unique". Oh, and these should outweigh the cost and time savings afforded by instant reprogrammability too, of course.)
Given the overwhelming rate of adoption of these card-key locks by hotels, whether or not the advantages are "obvious" or not seems less relevant than the fact that the advantages exist.
The link you gave led to an article with the headline "Dems Threaten Suit in Florida as More Purported Ballot Boxes Turn Up". In the article itself, purported is further qualified as "what look to be five purported ballot boxes." One of the things which looked to be a purported ballot box turned into a dead end since it "apparently contained notebooks."
If it was "all over the news" at the time, surely you can come up with something a little more solid than something that looks to be an article purporting to describe some alleged missing boxes which in fact contained notebooks? All I remember in the news at the time was a lot of empty accusation, speculation, and whining. Perhaps you can enlighten me with a nice, solid, conclusive link, especially given all the time that has elapsed to allow full investigations about anything which allegedly may slightly resemble a purported missing ballot box. Or are those investigations more like, oh, I don't know, OJ looking for the "real" killer?
Wait! Let's not dive into the good ideas just yet. First, someone needs to point out that the article author is a little confused on several key issues.
Got a hotel with perfectly good door locks and metal keys? Rip them out and replace them with computerized locks and swipe-cards.
There is nothing "perfectly good" about a lock whose keying needs to be changed every few days for liability and safety purposes. On-staff locksmitch or programmable locks? Hmmm.
These computer professionals say that accurately counted free elections are the bedrock of democracy. Voting, they claim, is too important to be done on a computer. The irony is delicious--it's sort of like group of doctors arguing for the return of leeches because the President of the United States is too important to be treated by modern medicine.
Oh boy. Even if this analogy were suitable (it isn't, obviously), there is still no irony here.
Because the voting is done on a large touch screen, they can use big fonts that are easier for the elderly to read.
Eh? How, exactly, is it easier to print big fonts on a screen than a piece of paper? I think the cost of paper varies less strongly with size than, say CRT and LCD technology.
They can even confirm the voter's choices on a second screen--which means that there would be no more elderly Jewish voters in Palm Beach accidentally casting their ballots for Pat Buchanan.
Oyve. Wow, two screens. Well, that sure represents carved-in-stone untemperable data to me. Regardless of the fact that the data could still be sitting in memory, not recorded permanently, and still quite subject to fraud or incompetence.
The [trojan or back-door] logic could be so well hidden that not even a careful review of the machine's source code would find it. This isn't as far-fetched as it might sound: Unauthorized features called "Easter eggs" are routinely hidden in commercial software, even software shipped by Microsoft.
(Emphasis mine). Bullshit! Careful review of source code finds as much as it wants to. And the example of "Easter eggs" in MS software is inappropriate since MS doesn't release source code.
paper is a fundamentally bad way of making and keeping accurate records. Paper is bulky and heavy. It can be hard to read something recorded on paper, no matter whether the marks were made by hand with pen-and-ink or by a computerized printer. Paper rips and gets jammed in machines. Paper dust gets everywhere. Eliminating paper, Selker explained to me, has the potential for dramatically improving elections.
WTF? And computers are less buggy than paper?!?! Help me.
"But what about all of the ways that you can hack the voting machines?" I asked him. Selker laughed. Politicians, he told me, have been hacking elections in America for more than 200 years.
Hahah, haha. Good pun. Now, seriously, what about all the ways that you can hack the voting machines?
thousands of Democrats, many of them minorities, showed up at voting places and discovered that they were no longer registered. Why? Because it's illegal for convicted felons to vote unless that right is specifically restored. Florida had recently purged the voting roles against a computerized database of convicted felons; tens of thousands of people were removed, some apparently in error.
Oh no, the felons couldn't vote. Whatever shall we do? Jeebus, I think I know the case in question, and the "some apparently in error" were 2 people with repeatedly rejected appeals. Not pending appeals mind you, flat-out rejections for appeal -- though apparently the felons claimed that was unfair. this is not the sort of election hacking that worries me.
Other techniques for stealing an election, Selker told me, are stationing tow trucks outside the polls to intimidate voters; setting up po
Decent article, but regarding this part:
We've gone from following the rules to playing the odds.
And if we do follow the rules and don't play the odds, then we are figured to be suckers.
Speak for yourself. Some of us just keep doing the right thing as best we can, no matter what everyone else is doing. We usually come out better in the long run.
You are probably right that hardware "glitches" (defined here as a temporary, non-fatal, non-locking error, just for clarity's sake) happen less often than software failures. My guess is that it's about one order of magnitude (10x). Your guess is more. I know people who adamantly argue that it's much less. Unfortunately for all of us, since there is rarely any way to recognize that an hardware glitch has occured, we can't really know for sure. How can you ever know for sure if the hardware retains no evidence of its "glitch"?
:)
[Hardware errors] are typically of the form, "this program doesn't work at all", not of the form "This program works but under some circumstances, with just the right conditions, it finishes successfully but with the wrong result." If a hardware glitch flips a bit from 1 to 0, for example, the chances that the program will still run anyway but with just the wrong total is very small. Alter a random bit in the program and it's highly likely to cause it to die altogether.
I think you missed my attempt at distinguishing harware failure (which is what you seem to be describing) and a hardware "glitch" (as defined above). I assert that something on the order of 10% of all "crashes" are due to non-fatal (in that the chip doesn't fry), temporary, self-recovering chip failures, and the remaining 90% are software. More than most people acknowledge. Unfortunately, as explained above, there's no way to know how right or wrong either of us are (unless all software flawlessly trapped all errors with appropriate logging and reporting, which would make the question moot were it true).
Also, the failures aren't all 0->1 or 1->0 bitflips -- that really only applies to storage elements like RAM and flip flops. We have ways of making that redundant and really robust (starting way back when with parity, up to modern 3-bit ECC and RAM layouts designed to defend against cosmic radiation). It's the other stuff (interconnect resistive shorts, race conditions, excessive leakage, non-fatal punch-thorough, unexpected EMI, system-wide signal integrity, local hot-spots, On-Chip Variation, etc.) that we can't try to correct when they fail which cause the most problems these days.
Also, I think you usually have more data than executable code in memory, so it seems that a random bitflip in your RAM (assuming no ECC) would affect data, not program, and therefore (probably) not be fatal. Moreover, lots of opcodes with similar arguments have 1-bit difference, so you'd not necessarily get a coredump on changing "branch if greater than" to "branch if less than", would you?
Regarding knowing when a piece of hardware (chips, mostly) cause a "glitch": AFAIK, there is exactly one company trying to remedy this. Sun Microsystems. "RAS" (Reliability, Acessibility, Scalability, in that order) is the mantra at Sun, and the next generation stuff is insanely self-testable, to the point that the testability sometimes costs more dollars and time than the normal functionality.
Even with this expense, Sun is not even aiming at 100% -- it's impossible with the current knowledge base. Someone has to come up with some amazing new way of thinking as significant as Newtonian physics or relativity theory in order to change that. Until then, you just keep thinking it's all a software problem (except for obvious, permanent hardware failures), and I'll keep laughing when the programmers get flak for my bug
You just made me think of this:
Has (or why hasn't) Apple considered developing a DirectX-compatible graphics API, or OpenGL wrapper, or similar? Seems to me that even moderate compatibility with the most popular, most heavily-developed (and by many accounts pretty good these days) graphic subsystem (DirectX 8+) would go a long way toward helping revive the flailing mac gaming market, which would be a good thing.
I, for one, would certainly love to be able to frag some of my mac-using buddies in Battlefield 1942 (though I'm afraid to consider the cost of a max that could run bf1942 at the resolution and detail settings I'm accustomed to (1280x1024x32bit, all options at max on an XP2400/1GB/GeFX5900U).
Best part of the article, and hilarious:
While legacy control systems are often UNIX-based ("Control-Alt-Delete scares power plant operators," Ahern said) and thus immune to MS worms and virii, their 10-megabit networking technologies can easily be overwhelmed. "Even the load from leading intrusion detection and monitoring systems can create a denial of service and shut these plants down," Ahern said.
It never ceases to amaze me how so many otherwise technically savvy people buy into the notion that hardware is infallible and software is always to blame for "glitches" or "crashes." I would concede that software causes more problems, but the notion that "computers don't make mistakes" is remarkably wrong.
:)
No chip ever made is tested 100%. Test coverage of 99.0% is considered excellent in the ASIC and custom IC design worlds. Many go to fab with less than 95%, sometimes 90% test coverage. So you have 10 million gates on a hunk of stuff you grew and screenprinted with toxic chemicals, with a decent plan to make sure that 99.0% of them can be tested to work as they should. You do the math -- lots will fail, and worse, some will fail occasionally and then resume working. Moreover, we can't really test all posible sets of stimulus -- that would take an incredibly long time in an industry where tester time on billion-dollar testers is doled out in 5s increments (30s is considered unworkable by most fabs, and would still allow us to test less than 1% of all possible combinations of inputs and transitions).
The interconnect between chips is another problem that's hard to measure, but non-zero. Same with passive components (capacitors, resistors) -- they have non-zero non-fatal failure rates. Which is an obfuscatory way of saying they can "glitch" or "crash". Thank Ohm a resistor's reboot time is much faster than Windows or you'd really notice the hardware failures
I don't have time to go into system-wide signal integrity (intractable), fault-tolerance that isn't, metastability, radiative interference such as cosmic rays and alpha particles emitted from local metals, etc. There's a lot that can and does go wrong in hardware.
I'm really kind of reluctant to post this, since as a hardware designer it's cool that I never hear the "you're why my computer crashes" comments that my software engineers suffer. It's also fun to se MS take the brunt of most PC users reliability complaints. In truth, they probably deserve a lot of it, but not the 100% most believe -- hardware does sometimes fail for a microsecond and then recover nonchalantly, as if nothing happened, sort of like when a cat trips or runs into a wall.
I've googled my heart out on this one to no avail. Can you please give me some link or keywords to help me find more on this (names maybe)? I would love to have this ammo (if reliable) for my next discussion on the subject with a certain friend.
BTW, I'm not looking to be anti-organic. In general, most of the (fresh) organic fruits and veggies just look better to me, so I don't mind the small premium required to get them. However, some are of the bent that all organic is good and all non-organic is bad, and it's always handy to be able to point out at least one case that breaks the rule to instill a little healthy skepticism.
Thanks in advance.
First, thank you for posting that. It truly was informative. I had no idea, and I've been paying attention to stories about how whacked the Recording Industry is.
Second, wow -- that seems as contrived, arbitrary, and complicated (note: not complex, complicated, the distinction is important here) as the US Income Tax Code.
How did such craziness happen?
The action seems perfectly reasonable to me:
To get the school's message across, all students were asked to sign a document confirming that their computers were updated with all the needed security upgrades. Not enough students confirmed that their machines were updated, prompting the GMU action today. Administrators said they would try later today to reconnect dorms, weeding out students with infected PCs. Students living off campus can continue to dial in to the campus computer network.
Looks like the kids are getting a decent deal on virus-removal and system updates too:
Students are being charged $30 if a university technician is called in to clean an infected machine, a school spokesman said. Students can go to off-campus experts for a fix but must certify that their computers are updated with the latest security fixes before being allowed to access the campus network.
Hmph, I can't find anything wrong here. Of course, there are a couple of choice quotes from the kids who, I believe, are our future:
Kimberly Borchert, a 19-year-old sophomore, said her computer "freaked out" as soon as she plugged it into the school's network last week.
Freshman Andrew Canose was one of several GMU students who encountered problems after installing the university-provided anti-virus software. Canose found the new program conflicted with an older anti-virus program already on his computer. "My computer is like at war with itself and won't work," he said.
But my favorite lines are from the admins, such as this gem:
"I think we really need to groom a new type of student who is responsible for their computer security," said Kathy Gillette, manager of George Mason University's beleaguered tech support center. "A lot of them lived at home and mom or dad took care of the computer so they've never learned how to fix them, but hopefully we'll be able to teach them that too."
And the classic:
"There were a certain percentage of students that wouldn't listen to us unless we hit them upside the head with a lockout," he said. "You simply can't deal with these problems until you've got your network under control."
I could imagine doing it for a lot less money, since you don't need the PVR features or the channel guide, etc. Just a webcam and a cheap PC with lots of HD space would do the trick just as easily (perhaps easier) and much more cheaply.
I'll go out on a limb and say no one will ever buy this to hook it up to a security camera.
Did you RTFA? I don't think so, since it's not in any way about developing a new processor architecture. Rather, it's about developing a new type of physical switch to replace a silicon transistor. If they got it to work, they could make an x86 or any other type of processor out of it -- but that's really irrelevant this early in the game.
"electrons jumping paths"? You're not in the semiconductor industry are you? :) Kindly accept the following reality check:
1. Double die size? Heat's not the problem (bigger die surface area makes it easier to dissipate heat, not harder, assuming same transistor density and switching rate). Manufacturability (and thus price) is the problem. Currently, the largest silicon die anyone can make at a price anyone will pay is just under 20mm on edge (400mm^2). Yield for such dice is around 20-30% (so, 70-80% of the dice made will be bad, and simply thrown away, but paid for in the price of the good dice). Doubling the die size to 40mm would cause an exponential increase in defect rates, resulting in yield around 0.01%, and prices around 1000x current.
2. Optical-based chipsets? I'm not sure what this means. If you mean optical processors (using photons instead of electrons) within a chip, that doesn't exist yet. If you mean inter-chip signaling using optics, this does exist (MEMS), but it doesn't help much to have your board-level interconnect have infinite bandwidth if your processors can't process it. Which they can't.
3. OK, efficiency is good, but hyperthreading is not really an example of increasing efficiency. A hyperthreaded chip is still doing the same amount of processing per gate, so it's not really any more efficient, it's just that some of the logic has been duplicated on chip to allow doing some operation twice (or more) at once. We can't just hyperthread everything, since adding more (duplicate copies) of logic requires a larger die to hold it all, which leads to the problems explained above in #1.
And the current crop of chips using copper interconnects propigate at maybe 1/2 or 2/3 that speed, meaning in a theoretical 1,000GHz chip (not too far off by Moore's law), electrical impulses can only travel about 50um.
Moore's Law says nothing about operating frequency. As pointed out already, Moore's law refers to the total number of components (switches) that can be fit on a die.
Checking a layout I'm working on now (0.13um all-Cu process with 0.28um wire pitch), I see that I have a 4mm wire with a nice 4x driver, and the delay on that wire is 182ps (MIN) to 423ps (MAX). So, MIN case (fastest process), that's about 45ps/mm, or 0.22mm/ps. 1000GHz gives a 1ps period, so I can move a signal a distance of about 0.22mm, or 220um, in the best case. Worst case (MAX process) I can do about 100um/ps. Your 50um is only off by a factor of 2-4 (not bad), but please note that this is today's process, and I could make it faster still (stronger driver, wider wire) if I wanted (needed) to. And, frankly, even 50um is plenty of distance to do a lot of work. Not really all that daunting, even now.
Finally, keep in mind that not every signal path in a chip has to make its destination in 1 clock cycle. Some are what we call "multicycle paths", and the system is designed to allow for 2 (or 3, or whatever) clocks to tick before using certain signal levels. Long-distance interconnect are usually MCP.
Er, there is no current process for making the stuff the article is about, other than some poor guy on his knees begging a CVD reactor to work right. The $3B mentioned is the cost for building a fab for a modern semicondoctor process which, while certainly a lot, is not a real problem since there are quite a few companies prepared to make such investements given the juicy returns (IBM, NEC, LSI, TSMC, . . . ).
What, do you drive it only downhill? (Maybe the reverse of the route I took to school when I was a kid, which was uphill, both ways . . . )
If not I call bullshit. According to Toyota themselves, the V6 Camrys max out at 33MPG (highway), 23 (city):
Manual transmission 23/33 [6] 23/33 NA
Automatic transmission 23/32 23/32 23/32
I can't even squeeze 50MPG out of my V6 Jetta (which is several hundred pounds lighter than a Camry), even if I drive like a grandma afraid of acelleration. I usually get 20-30 in real life, according to the Jetta's computer thing, but I love to acellerate. My wife gets a litter better, but not 50.
Great idea! Except that, in general, cities wouldn't like that. Parking meters are a source of income (from the tickets issues when the expire), not a distribution system to be made more efficient. Why do you think it's illegal in many places to feed a meter that's not your own?
Here, let me help: you're trolling.
For future reference, here's how I could tell: (1) you created a false dichotomy, (2) the false dichotomy is based on nothing at all, not even the usual anecdotal evidence presented in such case and (3) you assail the war in Iraq in a silly way.
Hope that help!
Me personally? A lot. Like 3-4 emails per day are covered by NDA, and the lawyers make us put those silly little sigs that do nothing. Oh, and we use Thawte certificates to encrypt and sign every one of them (well, even the non-NDA ones usually, out of habit).
In the ASIC design industry, I'd say this is pretty common. All discussions between vendor and customer are under NDA, sometimes even before any discussions at all happen there are multiple NDAs in place. And lots of the business (even exchange of IP in source code attachments) takes place by email.
Other industries? I'm not sure.