No, I'm not concerned about that at all. I cannot envisage a situation where all of the following apply:
You have permission to distribute the file from the copyright holders.
You have created a.torrent from the original files, and can run a tracker 24/7.
You cannot run a seeder 24/7 with access to the original files, throttled to a rate no greater than you'd offer the files at over HTTP or FTP.
BitTorrent is designed to take over whenever someone would have otherwise offered the files over HTTP or FTP, but to use downloader's upstreams to increase other people's download speeds. I'm not bothered about other uses, since they're not what BitTorrent was designed for; it's no surprise that it sucks at things it wasn't supposed to do. Just because it's popularly abused doesn't mean it's bad that it doesn't work that way.
As soon as you remove the server, BitTorrent breaks completely; the tracker runs on a server, and unless you're aiming to do an end-run around copyright law, nothing stops the tracker server also acting as a slow BitTorrent seed.
Only people who can get to the tracker can grab them; there is no BitTorrent network. So, as you communicate with the tracker over HTTP, you can protect that as easily as a HTTP-based download, and you can stop users having access to the.torrent or equivalent.
There are no security or access issues with BitTorrent that are not present with HTTP based downloaders like iTunes. So, if iTunes is secure enough, so's the proposed BT based site.
Same way Apple does with iTunes Music Store, except instead of offering DRMed files for download via HTTP, offer them for download via BitTorrent, with the BT client hidden within your download application.
If you want individually watermarked files, use strong encryption on the downloadables, then have your download application decrypt them and apply the watermarking and individual decryption keys. If your download application is cracked, you're in the same position as if your DRMed player is cracked, so there's no extra risk in this method.
If I were implementing it, I'd give each user their own private key, and have the file encrypted with AES. Use BT to download the file, then prepend a small, customer specific file header with the file key encrypted using their public key, and a small amount of other file data, so that the AES'd data is useless without it, even if you could decrypt it. The customer specific download (which has to come from your servers) is then less than 16KByte, while the data (which is torrented) is still multiple mega- or gigabytes.
BitTorrent is just a file transfer method; like HTTP or FTP, it transfers files. DRM is applied at the file level, and is not related to the file transfer method, whether it be BitTorrent or HTTP (iTunes can use HTTP to download purchased music; I don't know if it uses it exclusively, or only when behind a strict firewall).
It's only a problem if you're looking at it from a piracy point of view. As a legitimate user, the server that in a pre-BT world would have been an FTP or HTTP server is now always seeding that torrent. The incentive to keep seeding is to ensure that your customers can always get the content at full speed via BT, so that they don't demand money back or switch to an alternative download method.
If you're doing it commercially, act as both a seeder and a tracker. For unpopular content, you simply seed the individual download it, giving them download speeds not short of HTTP. For popular content, they get the speed they'd have got from HTTP plus the benefit of other people's upstream.
Note that in this case, you closing your client as soon as the download completes reduces the benefit the seller gets, but does not negate it, as BitTorrent uploads and downloads simultaneously, even if the file is incomplete.
In Sun's case, Niagra is intended for big database stuff; the idea of Niagra is that, within certain limits, the time taken per access is unimportant. What matters is the number of queries per second.
Sun hopes that while Niagra may take (example numbers only) 10ms for a typical query, whereas an equivalently priced Opteron can do the same query in 1ms, the Niagra processor can do 1,000 of those queries in 500ms whereas the Opteron now takes 1200ms.
A processor's wait latency is the time it spends doing absolutely nothing while it waits for an external device to catch up. If your RAM latency is around 100 cycles, and context switching costs you 100 cycles, you're right in saying that efficiency goes down. On the other hand, if each context switch costs you 10 cycles, you can context switch nine times before you've started to lose efficiency.
Sun are putting in hardware to ensure that context switches are fast (possibly even one or two cycles); hopefully, this will result in the context switches costing less than waiting for memory accesses, and speed up the throughput of the system as a whole. So, benchmarking one thread of execution will show a slow system, whereas a group will hopefully show a big speedup.
A big webserver or database server is a highly parallel, memory-latency-bound system; each request is an individual thread, and in most database and web servers, locks are finegrained enough to allow many requests to proceed in parallel, subject to them being able to retrieve the data from RAM or disk in a timely fashion.
Third scenario; there are reports of attempted abuse, and that attempted abuse is then dealt with.
I'm not familiar with the PATRIOT Act, so I'm not going to comment on that law at all, but to reuse the psychiatric hospital as an example, there are three scenarios:
There are reports of attempted abuse; the management say that there is no abuse, and the reports must be false. In this scenario, the likelyhood of there being genuine abuse is high, as attempted abuse is being reported, but not dealt with, giving potential abusers the feeling that they can get away with it.
There are no reports of attempted abuse; management claim that this shows that there is no abuse. Chances are good that either reports are being supressed, or that people are intimidated into not reporting abuse. Again, the likelyhood of there being abuse is high.
There are reports of attempted abuse; management look into it, and report back that some staff have been disciplined for it. Chances of abuse in this scenario are low, as whenever it's attempted, it's cracked down upon.
The point is that human nature tends to lead to abuse of power; whenever people claim that no-one is trying to abuse powers, there's something odd going on (either lots of really moral employees, and no bad apples, or abuse being hidden). When abuse is attempted and cracked down on, it suggests that attempted abuse does occur, but is caught and prevented. Of course, I've assumed that penalties for successful abuse are high enough to act as a deterrent in the face of near-perfect detection; if this is untrue, abuse will be rampant regardless, as there's no reason for the bad apples to behave.
Some of the more obscure (and powerful) features of MPEG-2 are not used AFAIK. Things like hierarchical encoding, for example, (provide an SD lower stream, which an SD decoder uses, and an enhancement layer to go from SD to ED or HD).
Define porn in such a way that it's not going to cause trouble in one direction or the other. Bear in mind that if you classify a non-porn site as porn, you'll get sued for libel/slander as appropriate. If you classify a porn site as non-porn, you'll get sued for being negligent.
What's more, you're going to face action in multiple jurisidictions; a Utah court may slam you for classifying a site as not porn, and a California court may slam you for classifying that same site as porn.
PICS already exists to allow both the sites themselves, and an external classifying body, a software independent content description method (this site has total nudity, no violence, no sex, while this site has partial nudity, lots of violence, allusions to sexual activity). An admin can either run filtering software directly on each PC, or on a border web proxy (or both).
PICS has been a complete failure despite being integrated into Internet Explorer. The reason? No gatekeeper for access; sites weren't going to the effort of classifying themselves until it was a requirement to be seen, and ratings bureaux did not pop up to handle it, partly due to the liability issue, and partly due to the lack of demand for censorship software from consumers (at least, at the prices that make it worth doing).
So people do what I do. My username is (for example) bill-gates, so my password is bill-Gates051 at the moment; when the compulsory password change comes round in a few weeks' time, my password will become bill-Gates052. When I'm forced to change in Jan 06, it'll become bill-Gates061. Nice and simple for me to remember, but even my slashdot account has a better password.
A hypothetical situation I'd like you to consider: my mailserver explicitly informs anyone who connects to it that they may only use the mailserver to send solicited e-mail to my domains, and that unsolicited bulk mail is explicitly forbidden.
How does this interact with computer misuse laws? Further, how does this interact with those spammers who blanket-bomb "likely" addresses at my domain (e.g. webmaster@, info@, sales@)?
Where you're in a job which is making you unhappy (whatever the reason), it's inadvisable to leave before you've found something else to pay you.
If you've left, and don't find other work that you enjoy doing soon, you're at risk of ending up stuck doing stuff that you feel is a waste of your skills - something like flipping burgers, answering phones, whatever. You also have an issue getting back into your field later - saying that you quit because you didn't like the tools your employer was using is a potential red flag to a future employer, and may make it impossible to return to a field you enjoy.
More than that; bubble sort performs very well in the case where the list is normally sorted on input (O(n) scan to confirm that the list is sorted). Depending on the properties of your original dataset, you may find that bubble sort is actually the best choice. When implementing a packet output scheduler for an MPEG system, I implemented a bubble sort, then profiled the code, as it was a potentially serious hotspot. It turned out that as the common case was rapidly repeating small packets, the scheduler was best off with a bubble sort.
Also, whether or not the artist got any of the money seems irrelevant. People were still being charged for the copies. Sounds to me like when they "commissioned" the work what they were really doing was buying the copyright. Copyright law may not have existed, but it's the same idea. And how do you know that the great artists were happy with these sort of agreements? Maybe they just took what they could get? There wasn't a "music industry" like there is today, so they really had no other options except to just starve to death (or get another job).
It's the original which costs the money, the copying is peanuts - which was reflected in the pricing. It's only in the past few centuries that a business model based on restricting copying and selling every copy for the same price has existed
I don't know of any piece of art or music where the copies are being sold for the same price as the original. Again, sounds pretty much like what's going on today. The original recording of an album can cost anywhere from $10,000 to $10,000,000. Compared to that, $10 for a copy seems like peanuts to me.
You're missing the key difference; before copyright law, the standard pattern was as follows:
Rich man pays artist to produce a work.
Rich man sells copies to friends.
Rich man's friends sell copies to their friends.
Rich man's friends' friends sell copies to their friends.
and so on. Copyright law bans everything from 3 onwards without permission from the rich man (I'm assuming that he pays for transfer of copyright).
Copyright law was a reaction to the creation by Gutenberg and co. of an expensive system to produce cheap copies of certain classes of work. As such it was an industrial regulation, designed to ensure that those people profiting from copying a work did not do so at the expense of those who created the original, and I do not believe that it was intended to act as a restriction on your ability to give away copies of stuff you've got to your friends.
I would argue that morally, copying a friend's music at no charge (barter or money) isn't wrong; it's not different enough from lending them your copy without charging for it. As soon as there is a fee imposed (whether it be a penny per copying session, selling a derivative work such as a musical mash-up, I'll let you copy mine if I can copy yours, or simply insisting that you must buy CD-Rs from me to copy stuff onto), I believe that you have crossed the line into profiting unfairly from someone else's work, and should not be allowed to do so without permission from the copyright holder.
Note that in terms of P2P trading, this implies that systems that do not attempt to prevent "leeching" are morally OK. Systems which do attempt to restrict "leeching" are not, as they are insisting that you trade use of your bandwidth for copyrighted materials.
I would argue too that people do things not just because they are paid to do so, but because they enjoy their work. Even if I had one of your infinite replicators, I would still go to work, and create equipment used to produce television programs (my current job). I agree that we'd have to automate messy jobs that aren't fun (like sorting through rubbish), but I suspect that given time, people would be inspired to deal with that class of problem for their own personal benefit.
Oh, and it's nice to discuss this sort of thing sanely on Slashdot; too many people get caught up in the "it's illegal, and you're BAD if you're not sure it should be" mentality, or it's opposite on the illegal side of the fence.
What is a quality? How do you know that a computer that can pass the Turing Test does not have the experience of qualia? What makes the statement that a human is an unconscious process miming conscious behaviour a self-contradiction, but doesn't make the statement that a sufficently good AI is an unconcious process miming conscious behaviour a self-contradiction? Unless you're invoking mysticism, in which case no AI will ever count as conscious, you're not proving your case at all.
Rejections of Searle don't assume that there is no distinction between consciousness and the representation of output; we state that if an entity processes data in such a way that it appears to understand that data, then it does understand that data. There is no mystical quantity in understanding from this point of view; a machine can have a partial understanding of the data it sees.
We then follow on to define a conscious machine as one which can increase its understanding of any input without external changes (basically, any system which can learn from arbitrary input, and request additional input to help it understand).
By this definition, a cat is conscious; it can learn from its mistakes, and can do things with the intent of learning. A billing system isn't conscious; it understands enough about billing people to create bills, but no more, and even if it can learn from its input (this is a bad debtor), it's incapable of requesting extra information to allow it to learn more (e.g. how many children do my debtors have?).
The advantage of this view is that it is fully defined in terms of the inputs and outputs of the system; we can apply it to a human or an animal (a system whose construction is poorly understood), and see if it matches our views of what is and is not conscious. It can then be applied to a machine, and if the machine matches, it's effectively conscious. Searle's claim leaves us with no way to verify its truth value (as he claims that consciousness is a function of how a system is constructed), and no way to test whether a given machine is conscious.
Google's list of addresses is going to upset you then. They've already got tech centres in Bangalore and Hyderabad; presumably they didn't want to add Mumbai to the list.
The trouble with distributed computing is that not all problems can be suitably parallelised. For example, simulations are hard to parallelise; each time unit's calculations depend upon the results obtained in the previous time unit.
A supercomputer can be put to work partly because each serial element is fast, and partly because the results can be shared among nodes very quickly; think inter-node latencies in the 10s-100s of microseconds range, rather than broadband's 10s of milliseconds, and bandwidth measured in gigabits/s.
90% of your customers, or 90% of visitors to your website? The distinction is significant, especially if your website is IE-only.
I am technically savvy, and quite happy to buy online; I use a non-IE browser. My grandmother isn't happy to buy online, or even over the phone; she uses IE. If you count my grandmother towards your 90%, you're making a financially stupid decision; she'll never be a customer of your website, nor will she phone you and order stuff as a result of reading your site.
The PHB summary: for a multi-user machine, where different users speak different languages, there is no such thing as case-insensitivity. The rules for Danish are different to the rules for English are different to the rules for Japanese are different to the rules for Arabic etc. You can only get true case insensitivity by insisting that everyone names their files in the same language.
If the kernel makes filenames case insensitive, some filenames are different to (say) an English speaker, but identical to a (say) Japanese speaker. Conversely, a Japanese speaker can create two files that to him are completely different, but that to a Danish speaker differ only by case.
Given case-insensitivity, how do you tell them apart? One solution (taken by Windows 95/98/ME) is to insist that all on-disk filenames are in the same language. This means that with English Windows 98, I get case insensitivity for all my filenames. Kenji does not, and gets told that he can't use filenames that to him are different (in Japanese, using Shift-JIS), because once they map to English, they differ only by case.
Define case-sensitivity for all users, such that the case rules stay the same no matter what locale I use. I happen to be English, so I can cope with a US-ASCII filesystem, and US-ASCII collating rules. If you're Danish, you have a different set of collating rules, and the mechanisms for a case insensitive comparison are different; it gets even worse when we introduce Chinese, Japanese, Korean, Russian, Greek and other languages which don't use the Latin alphabet into the mix.
Putting Unicode rules into the kernel is also painful; the collation rules for Unicode are huge and messy, and this relies on all users selecting the same Unicode compliant encoding (note that CJK users tend to prefer 2-byte encodings like UTF-16, while Latin alphabet users tend to prefer single-byte encodings like UTF-8).
Far easier to have the kernel say that a given byte ('/') is a directory separator, another byte ('\0') is the end of file name character, and leave user apps to interpret the byte stream. Then, I can use UTF-8 freely, and it's fine; a Japanese user can use Shift-JIS, and while his filenames look wrong to me, I can select any file he can create. In a case-insensitive system, he can create files that have different names in Shift-JIS, but are differentiated only by case in UTF-8 or ISO 8859-1.
You've never looked at the UK system, as it doesn't work like that at all.
Firstly, the judge can refuse to award costs, or can award them such that the winner pays all; if a big company tries a trick like you're suggesting, a judge will probably use this flexibility. Note that under a loser pays system, the judge has to explain why they didn't award costs, or awarded them in a "winner pays" fashion.
Secondly, if you've got a strong case, you can get a good lawyer to work for you for minimal expense; typically, they demand an up-front payment of £500-£1000 (maybe as much as $2500) to touch the case, but then works in the hope of winning the case and getting a big costs award (courts normally award your standard fee schedule, plus credit-card rate interest).
The result is that anyone faced with a case they are likely to lose is going to settle. Where it's genuinely unclear, the courts revert to pay your own costs, and where you have an abusive but technically victorious litigant, they still pay everyone's costs.
Slashdot Mirror
BitTorrent is designed to take over whenever someone would have otherwise offered the files over HTTP or FTP, but to use downloader's upstreams to increase other people's download speeds. I'm not bothered about other uses, since they're not what BitTorrent was designed for; it's no surprise that it sucks at things it wasn't supposed to do. Just because it's popularly abused doesn't mean it's bad that it doesn't work that way.
As soon as you remove the server, BitTorrent breaks completely; the tracker runs on a server, and unless you're aiming to do an end-run around copyright law, nothing stops the tracker server also acting as a slow BitTorrent seed.
There are no security or access issues with BitTorrent that are not present with HTTP based downloaders like iTunes. So, if iTunes is secure enough, so's the proposed BT based site.
If you want individually watermarked files, use strong encryption on the downloadables, then have your download application decrypt them and apply the watermarking and individual decryption keys. If your download application is cracked, you're in the same position as if your DRMed player is cracked, so there's no extra risk in this method.
If I were implementing it, I'd give each user their own private key, and have the file encrypted with AES. Use BT to download the file, then prepend a small, customer specific file header with the file key encrypted using their public key, and a small amount of other file data, so that the AES'd data is useless without it, even if you could decrypt it. The customer specific download (which has to come from your servers) is then less than 16KByte, while the data (which is torrented) is still multiple mega- or gigabytes.
BitTorrent is just a file transfer method; like HTTP or FTP, it transfers files. DRM is applied at the file level, and is not related to the file transfer method, whether it be BitTorrent or HTTP (iTunes can use HTTP to download purchased music; I don't know if it uses it exclusively, or only when behind a strict firewall).
It's only a problem if you're looking at it from a piracy point of view. As a legitimate user, the server that in a pre-BT world would have been an FTP or HTTP server is now always seeding that torrent. The incentive to keep seeding is to ensure that your customers can always get the content at full speed via BT, so that they don't demand money back or switch to an alternative download method.
Note that in this case, you closing your client as soon as the download completes reduces the benefit the seller gets, but does not negate it, as BitTorrent uploads and downloads simultaneously, even if the file is incomplete.
Sun hopes that while Niagra may take (example numbers only) 10ms for a typical query, whereas an equivalently priced Opteron can do the same query in 1ms, the Niagra processor can do 1,000 of those queries in 500ms whereas the Opteron now takes 1200ms.
Sun are putting in hardware to ensure that context switches are fast (possibly even one or two cycles); hopefully, this will result in the context switches costing less than waiting for memory accesses, and speed up the throughput of the system as a whole. So, benchmarking one thread of execution will show a slow system, whereas a group will hopefully show a big speedup.
A big webserver or database server is a highly parallel, memory-latency-bound system; each request is an individual thread, and in most database and web servers, locks are finegrained enough to allow many requests to proceed in parallel, subject to them being able to retrieve the data from RAM or disk in a timely fashion.
I'm not familiar with the PATRIOT Act, so I'm not going to comment on that law at all, but to reuse the psychiatric hospital as an example, there are three scenarios:
The point is that human nature tends to lead to abuse of power; whenever people claim that no-one is trying to abuse powers, there's something odd going on (either lots of really moral employees, and no bad apples, or abuse being hidden). When abuse is attempted and cracked down on, it suggests that attempted abuse does occur, but is caught and prevented. Of course, I've assumed that penalties for successful abuse are high enough to act as a deterrent in the face of near-perfect detection; if this is untrue, abuse will be rampant regardless, as there's no reason for the bad apples to behave.
Some of the more obscure (and powerful) features of MPEG-2 are not used AFAIK. Things like hierarchical encoding, for example, (provide an SD lower stream, which an SD decoder uses, and an enhancement layer to go from SD to ED or HD).
What's more, you're going to face action in multiple jurisidictions; a Utah court may slam you for classifying a site as not porn, and a California court may slam you for classifying that same site as porn.
PICS already exists to allow both the sites themselves, and an external classifying body, a software independent content description method (this site has total nudity, no violence, no sex, while this site has partial nudity, lots of violence, allusions to sexual activity). An admin can either run filtering software directly on each PC, or on a border web proxy (or both).
PICS has been a complete failure despite being integrated into Internet Explorer. The reason? No gatekeeper for access; sites weren't going to the effort of classifying themselves until it was a requirement to be seen, and ratings bureaux did not pop up to handle it, partly due to the liability issue, and partly due to the lack of demand for censorship software from consumers (at least, at the prices that make it worth doing).
So people do what I do. My username is (for example) bill-gates, so my password is bill-Gates051 at the moment; when the compulsory password change comes round in a few weeks' time, my password will become bill-Gates052. When I'm forced to change in Jan 06, it'll become bill-Gates061. Nice and simple for me to remember, but even my slashdot account has a better password.
How does this interact with computer misuse laws? Further, how does this interact with those spammers who blanket-bomb "likely" addresses at my domain (e.g. webmaster@, info@, sales@)?
If you've left, and don't find other work that you enjoy doing soon, you're at risk of ending up stuck doing stuff that you feel is a waste of your skills - something like flipping burgers, answering phones, whatever. You also have an issue getting back into your field later - saying that you quit because you didn't like the tools your employer was using is a potential red flag to a future employer, and may make it impossible to return to a field you enjoy.
Good luck finding a new job!
More than that; bubble sort performs very well in the case where the list is normally sorted on input (O(n) scan to confirm that the list is sorted). Depending on the properties of your original dataset, you may find that bubble sort is actually the best choice. When implementing a packet output scheduler for an MPEG system, I implemented a bubble sort, then profiled the code, as it was a potentially serious hotspot. It turned out that as the common case was rapidly repeating small packets, the scheduler was best off with a bubble sort.
You're missing the key difference; before copyright law, the standard pattern was as follows:
- Rich man pays artist to produce a work.
- Rich man sells copies to friends.
- Rich man's friends sell copies to their friends.
- Rich man's friends' friends sell copies to their friends.
and so on. Copyright law bans everything from 3 onwards without permission from the rich man (I'm assuming that he pays for transfer of copyright).Copyright law was a reaction to the creation by Gutenberg and co. of an expensive system to produce cheap copies of certain classes of work. As such it was an industrial regulation, designed to ensure that those people profiting from copying a work did not do so at the expense of those who created the original, and I do not believe that it was intended to act as a restriction on your ability to give away copies of stuff you've got to your friends.
I would argue that morally, copying a friend's music at no charge (barter or money) isn't wrong; it's not different enough from lending them your copy without charging for it. As soon as there is a fee imposed (whether it be a penny per copying session, selling a derivative work such as a musical mash-up, I'll let you copy mine if I can copy yours, or simply insisting that you must buy CD-Rs from me to copy stuff onto), I believe that you have crossed the line into profiting unfairly from someone else's work, and should not be allowed to do so without permission from the copyright holder.
Note that in terms of P2P trading, this implies that systems that do not attempt to prevent "leeching" are morally OK. Systems which do attempt to restrict "leeching" are not, as they are insisting that you trade use of your bandwidth for copyrighted materials.
I would argue too that people do things not just because they are paid to do so, but because they enjoy their work. Even if I had one of your infinite replicators, I would still go to work, and create equipment used to produce television programs (my current job). I agree that we'd have to automate messy jobs that aren't fun (like sorting through rubbish), but I suspect that given time, people would be inspired to deal with that class of problem for their own personal benefit.
Oh, and it's nice to discuss this sort of thing sanely on Slashdot; too many people get caught up in the "it's illegal, and you're BAD if you're not sure it should be" mentality, or it's opposite on the illegal side of the fence.
Rejections of Searle don't assume that there is no distinction between consciousness and the representation of output; we state that if an entity processes data in such a way that it appears to understand that data, then it does understand that data. There is no mystical quantity in understanding from this point of view; a machine can have a partial understanding of the data it sees.
We then follow on to define a conscious machine as one which can increase its understanding of any input without external changes (basically, any system which can learn from arbitrary input, and request additional input to help it understand).
By this definition, a cat is conscious; it can learn from its mistakes, and can do things with the intent of learning. A billing system isn't conscious; it understands enough about billing people to create bills, but no more, and even if it can learn from its input (this is a bad debtor), it's incapable of requesting extra information to allow it to learn more (e.g. how many children do my debtors have?).
The advantage of this view is that it is fully defined in terms of the inputs and outputs of the system; we can apply it to a human or an animal (a system whose construction is poorly understood), and see if it matches our views of what is and is not conscious. It can then be applied to a machine, and if the machine matches, it's effectively conscious. Searle's claim leaves us with no way to verify its truth value (as he claims that consciousness is a function of how a system is constructed), and no way to test whether a given machine is conscious.
Google's list of addresses is going to upset you then. They've already got tech centres in Bangalore and Hyderabad; presumably they didn't want to add Mumbai to the list.
A supercomputer can be put to work partly because each serial element is fast, and partly because the results can be shared among nodes very quickly; think inter-node latencies in the 10s-100s of microseconds range, rather than broadband's 10s of milliseconds, and bandwidth measured in gigabits/s.
I am technically savvy, and quite happy to buy online; I use a non-IE browser. My grandmother isn't happy to buy online, or even over the phone; she uses IE. If you count my grandmother towards your 90%, you're making a financially stupid decision; she'll never be a customer of your website, nor will she phone you and order stuff as a result of reading your site.
If the kernel makes filenames case insensitive, some filenames are different to (say) an English speaker, but identical to a (say) Japanese speaker. Conversely, a Japanese speaker can create two files that to him are completely different, but that to a Danish speaker differ only by case.
Given case-insensitivity, how do you tell them apart? One solution (taken by Windows 95/98/ME) is to insist that all on-disk filenames are in the same language. This means that with English Windows 98, I get case insensitivity for all my filenames. Kenji does not, and gets told that he can't use filenames that to him are different (in Japanese, using Shift-JIS), because once they map to English, they differ only by case.
Putting Unicode rules into the kernel is also painful; the collation rules for Unicode are huge and messy, and this relies on all users selecting the same Unicode compliant encoding (note that CJK users tend to prefer 2-byte encodings like UTF-16, while Latin alphabet users tend to prefer single-byte encodings like UTF-8).
Far easier to have the kernel say that a given byte ('/') is a directory separator, another byte ('\0') is the end of file name character, and leave user apps to interpret the byte stream. Then, I can use UTF-8 freely, and it's fine; a Japanese user can use Shift-JIS, and while his filenames look wrong to me, I can select any file he can create. In a case-insensitive system, he can create files that have different names in Shift-JIS, but are differentiated only by case in UTF-8 or ISO 8859-1.
Firstly, the judge can refuse to award costs, or can award them such that the winner pays all; if a big company tries a trick like you're suggesting, a judge will probably use this flexibility. Note that under a loser pays system, the judge has to explain why they didn't award costs, or awarded them in a "winner pays" fashion.
Secondly, if you've got a strong case, you can get a good lawyer to work for you for minimal expense; typically, they demand an up-front payment of £500-£1000 (maybe as much as $2500) to touch the case, but then works in the hope of winning the case and getting a big costs award (courts normally award your standard fee schedule, plus credit-card rate interest).
The result is that anyone faced with a case they are likely to lose is going to settle. Where it's genuinely unclear, the courts revert to pay your own costs, and where you have an abusive but technically victorious litigant, they still pay everyone's costs.