Theoretically the attacker could throw away un-memorizable permutations and cut through tons of entropy, but it would be extremely difficult.
No, it's not extremely difficult. It's more or less what password-cracking software does. Here are some examples of the kinds of low-entropy passwords that are susceptible to attack.
So you're an attacker and you try all the English words. Users can just type in 1337 5P33K. So you throw in leet words. Users can type their word ROT1. [...]
All of those are perfectly reasonable things to do, and doing them will get you closer to the theoretical maximum entropy of 6.6 bits per character for printable ASCII. An 8-character password done using the techniques you're talking about is probably quite secure. However, most users don't understand enough to do that. Here is an analysis that shows that users vary widely in their sophistication.
Most systems have a "three strikes and you're out for 5 minutes". So that kind of makes 65 guesses a minute impossible. You'd have 3 every 5 minutes.
Hmm...I tested this on my internet-facing server, which is running a Debian with a default configuration of ssh, and it doesn't seem to do that. After three strikes it disconnects, but there doesn't seem to be any long waiting period before it will accept another connection and allow three more tries.
Last time I dug through the man pages for ssh, I remember that it was a long slog to get information I wanted. Is there anyone here who knows where to look in/etc/ssh/sshd_config for this kind of stuff, and who can suggest what parameters to look at or fiddle with?
The solution is not complexity. It is limiting the number of attempts and logging the process and having a HUMAN review the logs on a daily basis.
Hmm...I'd be interested in hearing more about exactly how you implement that. It doesn't seem easy to me. You can limit attempts by throttling them to, say, a maximum of 1/second on any particular account; but then any user who knows your login name can DoS you. You can lock an account after, say, 1000 failed attempts, but you get the same DoS issue. You can have certain accounts that can only be logged in to from a terminal that's physically located where the computer is, but that's not an option for many people, including me, who rent a box in a rack from someone far away. (I'd really prefer not to have to call my webhost and get them to let me back in to my own machine. They'd charge me money, for one thing.) Having a human review the logs every day might make a lot of sense for, e.g., a corporate system with a lot of users, but personally, for me, it's just not something I want to dedicate time to for every single day of my life. And I'm curious what you would actually do if you found a bad guy doing such an attack. Use iptables to block that ip address? But what if the attack is coming from a botnet? If it's a botnet, then out of all the addresses that are attempting to get in, there may be one that belongs to the actual user, and it may not be obvious which one it is. And if you don't have certain accounts that can only be logged in to from a terminal that's physically where the CPU is, then I don't see how you can even guarantee that you'll be able to log in and do your daily check of the logs; you could be locked out yourself because they're DoS'ing you.
It seems to me like the really clearcut answer probably is to pick a very high entropy password.
With 8 characters you have to make on the order of 10^15 guesses.
No, that's incorrect. You seem to be assuming about 6 bits worth of entropy per character. The author of TFA has a long discussion of this buried in one of the tabs of the spreadsheet. The classic reference, which he gives, seems to be C.E. Shannon, "Prediction and Entropy in Printed English." Shannon estimated about 2.3 bits of entropy per character of printed English. That's a worst case, if you make your password out of English words. ASCII has 95 printable characters, which corresponds to 6.6 bits, so that's pretty much the best case. You can only use the best-case scenario to estimate the strength of passwords if either (a) users choose passwords that really look like a cat walking across a keyboard, or (b) the cracker guesses randomly, rather than using attacks like dictionary attacks that exploit the low entropy content of real-world passwords. Neither a nor b is likely to apply in the real world, so the right estimate is definitely an entropy content of more than 2.3 and less than 6.6 bits per character.
So for 8 characters, we can be pretty sure that the actual entropy content is between 18 and 53 bits. On the low end, that's only 10^5 combinations; on the high end, it's about 10^16. If you can test one password per second, then in 90 days you can try 10^11 combinations, which is (logarithmically) right in the middle of the range of estimates. So if you have an internet-facing server that accepts incoming ssh connections, and it never locks out or throttles back a user who is making too many login attempts, then it's quite reasonable that someone could crack your 8-character password in 90 days. If they have access to the hashed values of the passwords, then they can presumably carry out thousands of attempts per second, so they can almost certainly crack an 8-character password in much less than 90 days.
You're clearly very well informed, and your post is rich in detailed factual information. But although you posted it as a response to my post, you don't address any of my points. I offered some reasons why I don't think the US should have a tax-funded, government-monopoly, crewed space program. You haven't offered any reasons why it should.
For example, you write, We'll have to beg/buy rides from Russia to get to the ISS. Should the Russians decide they're not in the mood, they can easily say: "And what are you going to do about it?". So what? Why is this a bad thing? You haven't offered any reason why it's a good thing to send Americans to the ISS.
There are a lot of factors that are contributing to this trend. Vocational programs can require expensive equipment. For example, the community college where I teach physics recently spent a large amount of money to upgrade its printing program to digital equipment. I'd also imagine that insurance would be more expensive for a machine shop class than for an English class. At the K-12 level in the U.S., they're so focused on standardized testing these days that everything else is going away: music, vocational education, etc. Another vocational program at my school is horticulture, and I think one of the problems they're having is that they can't teach large sections, because the students need a lot of personal supervision. Small sections are seen as less cost-effective. I think there's also a heavy layer of class and racial prejudice that affects the horticulture program negatively, because here in Southern California, gardening is supposed to be what you hire Mexican immigrants to do.
The rationalizations for the space shuttle program are all bogus.
TFA laments the loss of the ability to repair satellites in space, and states that the shuttles have carried out 10 repair missions, 5 of them on the Hubble. Okay, Wikipedia says that the total cost of the space shuttle program has been 170 billion (in 2008 dollars), which works out to 1.5 billion per flight over 124 flights. So if the shuttle has only carried out 10 repairs in 124 flights, repairs clearly aren't one of its major missions. These figures also show that the repairs aren't cost-effective. The entire Hubble program cost about 5-6 billion dollars. Clearly this figure can't include the five shuttle missions to repair the telescope, since their cost would come out to 5x1.5=7.5 billion dollars. Given the exorbitant cost of the repairs, it seems that NASA could have been more cost-efficient by simply launching 2 or 3 space telescopes over the last couple of decades rather than one. Or, as an alternative, we could take the 15 billion dollars spent on the shuttle's 10 repair missions, and simply spend it on insurance for satellites. 15 billion bucks will buy a lot of insurance.
So if repairing satellites isn't an economically viable justification for the shuttle, what about getting to the space station? The problem is that we then have to ask what the purpose of the space station is. Yes, yes -- I have the answer! The purpose of the shuttle is to fly to the space station, and the purpose of the space station is to give the shuttle somewhere to fly.
Of course there's always scientific research as a justification for the crewed space program. But that always makes me wonder, why doesn't the crewed space program have to submit proposals for peer review, competing with other science experiments that want funding? Could it be because the scientific returns are trivial, especially in relation to the vast costs?
The plain truth is that the U.S. crewed space program is nothing more than an exercise in nationalistic propaganda.
It's skewed logic to say "Let's replace flash with javascript and html 5." They're different beasts, for a variety of reasons.
Flash is more strictly sandboxed than js. That's a good thing when the flash app is someone's ad, but possibly a bad thing if you want to write a web app.
Flash was specifically designed so that it could be used for games. Javascript wasn't. It is possible to write games in js, but I really doubt that you'll be able to do really fancy games in js+html 5 any time soon.
Flash is less proprietary than it used to be. Part of its remaining proprietary nature is because Adobe wants to make sure there's no other first-class toolchain for developing flash. But part of the reason it's proprietary is that it's encumbered by a ton of patents, which Adobe doesn't even own. Doing things using html 5 doesn't magically cure all the patent issues. The audio part of the html 5 standard doesn't require browsers to implement any patent-free codec, so web developers will still have to use mp3. The situation is even worse for video codecs. There are open-source flash players (gnash) and development toolchains (haxe), but there are serious legal obstacles preventing them from becoming plug-in replacement for Adobe's proprietary tools. Not only do we have the patent issues, but Adobe has tons of libraries that flash developers have become dependent on. Those libraries are all totally proprietary, and Adobe has very carefully set their licenses up so that there's no way to obtain the API docs and write a competing tool.
Xhtml can't die, because it was never really alive in the first place. IE never supported it properly, and that made it a no-go.
The best reason to want xhtml is the ability to integrate xml languages like svg and mathml. The good news is that html 5 will implement all the svg and mathml tags. The bad news is they'll just be duct-taped on. That is, if you want to use some other kind of xml language inside html 5, you won't be able to; only svg and mathml are designed in.
But the credit removal requirement in even CC-BY might cause license incompatibility if a free program under a GNU license uses CC-BY images, audio, etc.
As far as I can tell, that was an issue with CC v2 licenses, but it's been fixed in v3:
The parent post makes some good points about what was undesirable about the GFDL. In addition, there's the issue of needless proliferation of licenses. What everybody originally intended here was to make a commons that everyone could draw from. If A makes an animation, and B writes a song, and C performs B's song, and A, B, and C all try their best to put their work in the commons, then D should be able to come along and make a video consisting of A's animation with a sound track consisting of C's performance of B's song. There shouldn't be artificial obstacles just because A, B, and C chose different licenses.
I'm not saying there should only be one free-as-in-speech license for written materials. We do need at least two, because
there are real philosophical differences between BSD-style licenses and GPL-style licenses. But there is not a real philosophical difference between the GFDL and CC-BY-SA.
I also think that most of the returns weren't just because they didn't have windows but because they were expecting a full-featured laptop that was small and cheap.
That's possible, although I don't think I saw that sentiment expressed very much in the amazon reviews. The thing is, if people are disappointed in the keyboard and screen (which is reasonable, since they do kind of suck), you would think they would return linux netbooks and windows netbooks at equal rates. That wouldn't explain why retailers no longer seem to be selling the linux version.
Actually another possible explanation for why retailers are no longer selling them with linux is simply that that's what customers want. Early on, linux was the only OS available on the eee. However, once Asus started selling it with the choice of windows and linux, it kind of makes sense that users would buy the two versions more or less in proportion to their established popularity on the desktop. If the population of people who are used to windows is 50x bigger than the population who are used to linux, then it's probably logical to expect that 98% of people, given the choice, would buy the eee with windows. If that's what's been happening, then it totally makes sense that retailers would stop stocking the version that wasn't selling worth beans.
ARM doesn't matter. When the chipset, screen and SSD chew more power than the CPU, the CPU is the least of your worries.
I guess I didn't make that point very clearly. The thing is that Windows is basically an Intel-only OS. Windows CE does support other CPU families, including ARM, but if you want to run full-fledged Windows, with all the usual Windows apps, Intel is really your only option. There's been a lot of speculation that the next wave of netbooks will use ARM rather than Intel, and that would presumably make it much more difficult for MS to compete. However, I'm willing to bet that MS will compete vigorously on ARM if ARM becomes popular for netbooks. Even if it costs them a lot of money, they still want to keep their monopoly from evaporating because the hardware market evolves in a new direction.
When the eee PC came out, Linux had a big opportunity. Unfortunately, Asus completely blew the details of the implementation. They picked a sucky distro, and they did a lousy job on quality control and integration. My wife uses linux on the desktop, and when she saw the eee at Target for $280 she asked me to get her one for her birthday. The model they were selling at Target was out of date and not very good, so I ordered a fancier model on amazon for $400. It came with its wifi misconfigured, and Asus tech support told me they couldn't fix it, and I'd have to return it. This was a few months ago. Yesterday I was making a trip to Fry's, so my wife suggested I just buy one while I was there. Well, Fry's is now selling the eee only with Windows, and Amazon's site also doesn't have the linux version available. AFAICT retailers were just getting too many returns of the linux ones. You can pretty much tell what was going on based on the amazon reviews. Some, like mine, were being shipped misconfigured. In other cases, you had people buying the linux version and not understanding that it wasn't windows. And in still other cases, people were buying them with linux and then trying to install a (presumably illegal) copy of Windows, and failing. (None of this is new, either. All this stuff happened in the past with the Great Quality linux boxes they used to sell at Fry's. The hardware was in fact great quality, but Fry's was getting too many returns, so they stopped carrying them.)
I think the basic problem here is that it's expensive and difficult to do a good job integrating hardware and software for a consumer computer. That's the kind of thing Apple is famous for doing well. Apple puts a lot of money and effort into it, and they charge for it when you buy a mac. I just don't see how anyone is going to do anything like that in the netbook market, which is an ultra-low-margin market. It would have been especially difficult for East Asian manufacturers like Asus and Great Quality, which have a language barrier to deal with. (At one point, Great Quality was shipping their machines with a linux distro that didn't even have an English-language web site.)
Meanwhile, MS can afford to do what it takes to maintain dominance in all sectors of the market. MS doesn't even have to do a good job on netbooks. They just have to avoid doing such a horrible job that it becomes painfully obvious to people who have never used anything but Windows before.
It's possible that ARM-based netbooks will change the equation, but I wouldn't be surprised if MS jumps in and starts competing vigorously on ARM, simply to maintain their monopoly.
They've always been pretty hostile toward their customers.
Yep. One of the most important experiences that made me stop using proprietary software was with Mathematica. I bought a copy of Mathematica and installed it on my mac. This was back in the 90's, so I was running MacOS 6 or something. I upgraded to MacOS 7, and Mathematica would no longer run. Called Wolfram, and they told me I'd have to pay for a new copy of Mathematica.
They're not filling one of the traditional purposes of a website either, which is to present content in a highly-portable markup language that is readable in all web browsers.
Instead, they used Flash to embed PDFs into a little window on a website.
There are several different issues here.
First there's the fact that PDF is almost completely nonproprietary (i.e., the functionality required by the vast majority of users for reading and writing PDFs is nonproprietary), whereas flash has a lot more proprietary stuff in the mix. However, the proprietary aspects of flash mostly have to do with patented audio and video codecs, which are irrelevant in the case of scribd. A lot more of flash used to be proprietary. What Adobe's been doing is to open up flash a lot more, while still keeping enough of the tasty bits proprietary so that nobody else can really compete effectively with them to sell development tools. There are, however, open-source apps for both creating (haxe) and running (gnash) flash. If the vast majority of flash developers don't see haxe as a viable alternative to Adobe's toolchain, and the vast majority of users don't see gnash as a viable replacement for Adobe's plugin, I think that basically just shows that the OSS community hasn't made it a priority.
Another issue is that there are certain types of documents for which html simply isn't the right tool. For example, if you want to make a printer-friendly, illustrated physics textbook, pdf is the right tool for the job, not html.
As far as the relative merits of displaying PDFs in a flash application or a PDF plugin, I think for the average user scribd's flash app may indeed be the more convenient and useful choice. Most people use Adobe Reader as their PDF plugin. AR is slow to load, and has a horrible history of security problems. Most people aren't propeller-heads, and don't know that there is any alternative to AR. For a large document, scribd's flash app also has the advantage that it loads without having to download megabytes worth of PDF data. For someone on a dialup connection, this is a big deal. Although it's true that the PDF standard has advanced features designed to allow the user to see individual pages without having to download the whole file first, in reality I don't think that's implemented properly in the majority of PDF documents you see out there in the wild, and I don't think it's implemented properly in open-source readers like Evince, either. (Evince has truly horrible performance on PDFs consisting of many scans of bitmaps.)
Keep in mind that scribd's killer app is, frankly, copyright violation. I'm sure they don't intend it to be used that way, but that's the big thing that users are accomplishing with scribd that they can't accomplish in any other way. For instance, search scribd for "feynman lectures," and you'll get a whole bunch of PDFs consisting of pages scanned from a copyrighted book. The performance of scribd's flash interface on this type of PDF file is much, much better than what you'd get with straight PDF.
There's a tiny little text box taking up like 6 cm by 5 cm of space with a scrollbar... I have multiple monitors, huge space on my desktop, and they're cramming all the content into this tiny little unreadable scrollable space.
On my machine, it's 22 cm x 16 cm.
A lot of people are posting about how much they hate scribd's UI, but I don't see that as the big problem with scribd.
People have posted some of my books on scribd, and that's fine with me, because the books are free-as-in-speech. However, their system has some problems. For instance, if you search on scribd for "Newtonian Physics," which is the title of one of my books, the first 8 hits consist of 8 different uploaded copies of my book. Seems like a lot of scribd users don't bother checking to see whether something is already on scribd before they upload it. Now if I type in some text from my book as a search, only a few of the books come up, not all 8 -- don't ask me why. And when I click on the #1 search result, it's a version of the book from 2001, with an incorrect description and an incorrect license listed for it.
I think the fundamental problem here is that they're not serving one of the traditional purposes of a publisher, which is to act as a filter. Filters can be good or bad. A filter doesn't have to be all-or-nothing, and it doesn't have to be elitist or authoritarian. Google page rank is a filter. Slashdot's moderation system is a filter. Scribd doesn't seem to have enough useful filtering mechanisms. It just seems to act as a huge dumping ground, where anyone can put anything. The trouble is that finding anything there is like saying, "Huh, I need a new cartridge for my antique fountain pen. Maybe I'll go down to the town dump and dig around for one."
No, not really. One reason it's not big news is that scribd is currently too small a commercial entity to make any difference in this big marketplace. Another reason it's not big news is that other people are already selling digital books without DRM. Fictionwise and Baen are two examples that come to mind.
So, really, writers have absolutely no incentive to deal with Amazon anymore as their "bookstore," especially since the next generation of ebook readers will surely be touchscreen netbooks, making the Kindle look like a Tiger handheld next to the future's Game Boy.
Well, no. Amazon is a huge, profitable business that readers know about. Scribd isn't. That's a pretty strong incentive for writers to deal with Amazon -- or, more accurately, it's a pretty strong incentive for their publishers to. The author generally doesn't make any decisions about the distribution channels through which a book gets to the public.
TFA is wrong in three different ways: morally, politically, and practically.
Its proposals are morally wrong because they trample on individual freedom. For instance, #1 says "Any individual shall be able to choose anonymity when posting to Internet sites." Okay, I run a site that catalogs free books, and accepts user-submitted reviews. When new users register, it's made very clear to the new user that he needs to provide his real name. The reason is that I don't want people reviewing their own books. No, I'm not under the illusion that 100% of people are truthful about this. However, the author of TFA seems to be proposing that the government prohibit me from even setting this as a policy on my site. That's absurd. If users don't like the fact that I require them to give their real names, I can't force them to visit my site.
Politically it's wrong, because he seems to be completely clueless about the nature of individual rights in a constitutional democracy. Individual rights are not about telling A that he has to give B something, or forcing A to associate with B, or telling A that if he wants to do business he needs to act in a certain way. Individual rights have to do with being left alone, especially with being left along by the government. You can believe that the government has other legitimate functions besides safeguarding individual rights, but that doesn't mean that those other functions qualify as individual rights. If you think government should be inspecting cuts of beef to make sure they won't make people sick, then that's a perfectly legitimate, mainstream political opinion in the U.S. -- but it has absolutely nothing to do with any supposed individual right not to get sick.
And finally, he's wrong in practice, because these ideas are all just stupid in the context of, say, the United States in the year 2009. Taking his items one by one:
No anonymous posting. Stupid, as discussed above.
Net neutrality. The basic problem is a problem with monopolies, not a problem with a lack of individual rights. In my neighborhood, for example, I have only one broadband provider I can choose. Because of that, I can't vote with my feet if they start imposing loathsome restrictions on how much bandwidth I can use for different things. Get rid of the monopoly, which is the real problem. The U.S. has antitrust legislation already on the books. If that legislation already covers local broadband providers, then let's get it enforced. If it doesn't, then let's pass stronger antitrust laws. If people had a choice, they wouldn't choose a broadband provider that didn't offer net neutrality.
Users not responsible for malware. We already have laws about this kind of thing. We have employment law. We have laws about negligence. If I build a killer robot in my front yard, and it eats small children on the way to school, the law already discusses to what extent I'm in civil and/or criminal trouble. If someone else puts a killer robot in my front yard without my knowledge, we have laws about that as well. Likewise if I'm at work, and someone puts a killer robot in my cubicle, and it eats my PHB -- which is esentially what happened in the case he refers to. Can my employer fire me because the killer robot ate my PHB without my knowledge? I dunno, check you state's employment laws, and check your employment contract.
Closed-source software houses responsible for security flaws. We already have laws about this kind of thing, e.g., implied warranties of merchantability and suitability for a particular purpose. These may or may not apply to software in your state -- I dunno, I'm not a lawyer. Want to change the law? Go ahead, but there are obvious reasons not to go too far with this kind of thing. For instance, I don't think car manufacturers should be responsible for drunk drivers, and I don't think Microsoft should be responsible for people who double-click on malware that was sent to them attached to penis-enlargement spams.
This kind of thing is a minefield, and very hard to prove. To see what I mean, do a google search on "cannibalism anasazi." People get emotional about certain scientific issues, and often the reason they're so emotional is that there's painful history involved, and/or a history of the misuse of science. For instance, it's theoretically a reasonable scientific topic to look for correlations between race and intelligence -- but if you try study it, you'll unleash such a shitstorm that you'll wish you hadn't. Part of this is because the topic isn't PC, but part is also because of history (eugenics, Nazism, Cyril Burt).
Cannibalism has historically been one of these scientific issues that are just hard to study because emotions run too high. For instance, you have the history of Europeans portraying Africans as savage cannibals (which made it easier for Americans to justify slavery, and for the Belgians to justify cutting people's arms off in Congo).
Some archaeologists and anthropologists have gone so far as to claim that cannibalism simply doesn't exist, and never has. Others have found physical evidence that they interpret as evidence of widespread cannibalism in certain societies. Still others say that it exists, but only in a ritualized form.
I'm not convinced that the chances are very good of coming to a definite conclusion about cannibalism that might have happened hundreds of thousands of years ago, when we can't even study the more recent cases.
the axioms of ZF set theory are sufficient to construct real numbers, whereas you need a conservative extension of ZFC to be able to model NSA.
That's incorrect. Standard ZFC is sufficient for NSA. You may be thinking of Internal Set Theory, which is one way of approaching NSA, but postdates NSA by several decades.
It would be useful to understand just what the difficulty was that the failing or dropped-out students had - was it the teacher, the book, or something more intrinsic to NSA?
One story I've heard is that Keisler's course was disorganized. The book was handed out on purple mimeograph sheets, and the first 20 minutes of every lecture was filled with a grad student reading off errata, which the students were supposed to correct on their copies.
And while rote manipulations with hyperreal numbers aren't too hard to learn, to understand them rigorously involves abstract math and set theory much deeper than that needed for the real numbers and limits of standard calculus (see the Epilogue of the book).
I disagree. Actually, if you want to understand the plain old real number system deeply, it involves quite a bit of abstract math and set theory. For example, if you look in a freshman calculus book, they never prove things like the intermediate value theorem, because it requires a deeper understanding of the foundations of the real number system than a freshman in college is going to have. For an example of how deep and strange the theory of the reals can be, see Chaitin's number.
Nonstandard analysis isn't harder or deeper than calculus using limits. It's really just a different language for talking about all the same ideas.
I also disagree with you about "rote manipulations." Keisler actually does quite a good job of teaching the conceptual side of the hyperreals. Euclid went to his grave without knowing Cartesian coordinates; that doesn't mean he didn't understand geometry. Gauss and Euler used infinitesimals without ever learning about the hyperreal numbers; that doesn't mean they didn't understand infinitesimals. The Keisler book gives a very nice, self-contained axiomatic treatment of the hyperreals, in much the same way that the Elements gives a self-contained axiomatic treatment of geometry.
As states lower their legal limits to the point where they intersect with non-impaired drinking drivers, especially with a.01 or more margin of error,
A.08 BAC is nowhere near non-impaired. For instance, this DOT page states that
"At a BAC of.10, the probability of a fatal or serious-injury crash was estimated to be 6 to 12 times that of a driver with no alcohol."
Like most first time offenders, I took a plea deal to avoid significant jail time and paid the ridiculous fine and took alcohol awareness classes. The whole thing was a farce, intended to make money.
When you use words like "farce" and "ridiculous," it makes it sound like you don't want to take responsibility for your own actions. I don't think DUI laws are "a farce, intended to make money." I think they're intended to protect people like me from getting killed by people like you.
Slashdot Mirror
No, it's not extremely difficult. It's more or less what password-cracking software does. Here are some examples of the kinds of low-entropy passwords that are susceptible to attack.
All of those are perfectly reasonable things to do, and doing them will get you closer to the theoretical maximum entropy of 6.6 bits per character for printable ASCII. An 8-character password done using the techniques you're talking about is probably quite secure. However, most users don't understand enough to do that. Here is an analysis that shows that users vary widely in their sophistication.
Hmm...I tested this on my internet-facing server, which is running a Debian with a default configuration of ssh, and it doesn't seem to do that. After three strikes it disconnects, but there doesn't seem to be any long waiting period before it will accept another connection and allow three more tries.
Last time I dug through the man pages for ssh, I remember that it was a long slog to get information I wanted. Is there anyone here who knows where to look in /etc/ssh/sshd_config for this kind of stuff, and who can suggest what parameters to look at or fiddle with?
Hmm...I'd be interested in hearing more about exactly how you implement that. It doesn't seem easy to me. You can limit attempts by throttling them to, say, a maximum of 1/second on any particular account; but then any user who knows your login name can DoS you. You can lock an account after, say, 1000 failed attempts, but you get the same DoS issue. You can have certain accounts that can only be logged in to from a terminal that's physically located where the computer is, but that's not an option for many people, including me, who rent a box in a rack from someone far away. (I'd really prefer not to have to call my webhost and get them to let me back in to my own machine. They'd charge me money, for one thing.) Having a human review the logs every day might make a lot of sense for, e.g., a corporate system with a lot of users, but personally, for me, it's just not something I want to dedicate time to for every single day of my life. And I'm curious what you would actually do if you found a bad guy doing such an attack. Use iptables to block that ip address? But what if the attack is coming from a botnet? If it's a botnet, then out of all the addresses that are attempting to get in, there may be one that belongs to the actual user, and it may not be obvious which one it is. And if you don't have certain accounts that can only be logged in to from a terminal that's physically where the CPU is, then I don't see how you can even guarantee that you'll be able to log in and do your daily check of the logs; you could be locked out yourself because they're DoS'ing you.
It seems to me like the really clearcut answer probably is to pick a very high entropy password.
No, that's incorrect. You seem to be assuming about 6 bits worth of entropy per character. The author of TFA has a long discussion of this buried in one of the tabs of the spreadsheet. The classic reference, which he gives, seems to be C.E. Shannon, "Prediction and Entropy in Printed English." Shannon estimated about 2.3 bits of entropy per character of printed English. That's a worst case, if you make your password out of English words. ASCII has 95 printable characters, which corresponds to 6.6 bits, so that's pretty much the best case. You can only use the best-case scenario to estimate the strength of passwords if either (a) users choose passwords that really look like a cat walking across a keyboard, or (b) the cracker guesses randomly, rather than using attacks like dictionary attacks that exploit the low entropy content of real-world passwords. Neither a nor b is likely to apply in the real world, so the right estimate is definitely an entropy content of more than 2.3 and less than 6.6 bits per character.
So for 8 characters, we can be pretty sure that the actual entropy content is between 18 and 53 bits. On the low end, that's only 10^5 combinations; on the high end, it's about 10^16. If you can test one password per second, then in 90 days you can try 10^11 combinations, which is (logarithmically) right in the middle of the range of estimates. So if you have an internet-facing server that accepts incoming ssh connections, and it never locks out or throttles back a user who is making too many login attempts, then it's quite reasonable that someone could crack your 8-character password in 90 days. If they have access to the hashed values of the passwords, then they can presumably carry out thousands of attempts per second, so they can almost certainly crack an 8-character password in much less than 90 days.
You're clearly very well informed, and your post is rich in detailed factual information. But although you posted it as a response to my post, you don't address any of my points. I offered some reasons why I don't think the US should have a tax-funded, government-monopoly, crewed space program. You haven't offered any reasons why it should.
For example, you write, We'll have to beg/buy rides from Russia to get to the ISS. Should the Russians decide they're not in the mood, they can easily say: "And what are you going to do about it?". So what? Why is this a bad thing? You haven't offered any reason why it's a good thing to send Americans to the ISS.
There are a lot of factors that are contributing to this trend. Vocational programs can require expensive equipment. For example, the community college where I teach physics recently spent a large amount of money to upgrade its printing program to digital equipment. I'd also imagine that insurance would be more expensive for a machine shop class than for an English class. At the K-12 level in the U.S., they're so focused on standardized testing these days that everything else is going away: music, vocational education, etc. Another vocational program at my school is horticulture, and I think one of the problems they're having is that they can't teach large sections, because the students need a lot of personal supervision. Small sections are seen as less cost-effective. I think there's also a heavy layer of class and racial prejudice that affects the horticulture program negatively, because here in Southern California, gardening is supposed to be what you hire Mexican immigrants to do.
The rationalizations for the space shuttle program are all bogus.
TFA laments the loss of the ability to repair satellites in space, and states that the shuttles have carried out 10 repair missions, 5 of them on the Hubble. Okay, Wikipedia says that the total cost of the space shuttle program has been 170 billion (in 2008 dollars), which works out to 1.5 billion per flight over 124 flights. So if the shuttle has only carried out 10 repairs in 124 flights, repairs clearly aren't one of its major missions. These figures also show that the repairs aren't cost-effective. The entire Hubble program cost about 5-6 billion dollars. Clearly this figure can't include the five shuttle missions to repair the telescope, since their cost would come out to 5x1.5=7.5 billion dollars. Given the exorbitant cost of the repairs, it seems that NASA could have been more cost-efficient by simply launching 2 or 3 space telescopes over the last couple of decades rather than one. Or, as an alternative, we could take the 15 billion dollars spent on the shuttle's 10 repair missions, and simply spend it on insurance for satellites. 15 billion bucks will buy a lot of insurance.
So if repairing satellites isn't an economically viable justification for the shuttle, what about getting to the space station? The problem is that we then have to ask what the purpose of the space station is. Yes, yes -- I have the answer! The purpose of the shuttle is to fly to the space station, and the purpose of the space station is to give the shuttle somewhere to fly.
Of course there's always scientific research as a justification for the crewed space program. But that always makes me wonder, why doesn't the crewed space program have to submit proposals for peer review, competing with other science experiments that want funding? Could it be because the scientific returns are trivial, especially in relation to the vast costs?
The plain truth is that the U.S. crewed space program is nothing more than an exercise in nationalistic propaganda.
It's skewed logic to say "Let's replace flash with javascript and html 5." They're different beasts, for a variety of reasons.
Flash is more strictly sandboxed than js. That's a good thing when the flash app is someone's ad, but possibly a bad thing if you want to write a web app.
Flash was specifically designed so that it could be used for games. Javascript wasn't. It is possible to write games in js, but I really doubt that you'll be able to do really fancy games in js+html 5 any time soon.
Flash is less proprietary than it used to be. Part of its remaining proprietary nature is because Adobe wants to make sure there's no other first-class toolchain for developing flash. But part of the reason it's proprietary is that it's encumbered by a ton of patents, which Adobe doesn't even own. Doing things using html 5 doesn't magically cure all the patent issues. The audio part of the html 5 standard doesn't require browsers to implement any patent-free codec, so web developers will still have to use mp3. The situation is even worse for video codecs. There are open-source flash players (gnash) and development toolchains (haxe), but there are serious legal obstacles preventing them from becoming plug-in replacement for Adobe's proprietary tools. Not only do we have the patent issues, but Adobe has tons of libraries that flash developers have become dependent on. Those libraries are all totally proprietary, and Adobe has very carefully set their licenses up so that there's no way to obtain the API docs and write a competing tool.
Xhtml can't die, because it was never really alive in the first place. IE never supported it properly, and that made it a no-go.
The best reason to want xhtml is the ability to integrate xml languages like svg and mathml. The good news is that html 5 will implement all the svg and mathml tags. The bad news is they'll just be duct-taped on. That is, if you want to use some other kind of xml language inside html 5, you won't be able to; only svg and mathml are designed in.
As far as I can tell, that was an issue with CC v2 licenses, but it's been fixed in v3:
Wikipedia has a useful FAQ about the relicensing.
The parent post makes some good points about what was undesirable about the GFDL. In addition, there's the issue of needless proliferation of licenses. What everybody originally intended here was to make a commons that everyone could draw from. If A makes an animation, and B writes a song, and C performs B's song, and A, B, and C all try their best to put their work in the commons, then D should be able to come along and make a video consisting of A's animation with a sound track consisting of C's performance of B's song. There shouldn't be artificial obstacles just because A, B, and C chose different licenses.
I'm not saying there should only be one free-as-in-speech license for written materials. We do need at least two, because there are real philosophical differences between BSD-style licenses and GPL-style licenses. But there is not a real philosophical difference between the GFDL and CC-BY-SA.
That's possible, although I don't think I saw that sentiment expressed very much in the amazon reviews. The thing is, if people are disappointed in the keyboard and screen (which is reasonable, since they do kind of suck), you would think they would return linux netbooks and windows netbooks at equal rates. That wouldn't explain why retailers no longer seem to be selling the linux version.
Actually another possible explanation for why retailers are no longer selling them with linux is simply that that's what customers want. Early on, linux was the only OS available on the eee. However, once Asus started selling it with the choice of windows and linux, it kind of makes sense that users would buy the two versions more or less in proportion to their established popularity on the desktop. If the population of people who are used to windows is 50x bigger than the population who are used to linux, then it's probably logical to expect that 98% of people, given the choice, would buy the eee with windows. If that's what's been happening, then it totally makes sense that retailers would stop stocking the version that wasn't selling worth beans.
I guess I didn't make that point very clearly. The thing is that Windows is basically an Intel-only OS. Windows CE does support other CPU families, including ARM, but if you want to run full-fledged Windows, with all the usual Windows apps, Intel is really your only option. There's been a lot of speculation that the next wave of netbooks will use ARM rather than Intel, and that would presumably make it much more difficult for MS to compete. However, I'm willing to bet that MS will compete vigorously on ARM if ARM becomes popular for netbooks. Even if it costs them a lot of money, they still want to keep their monopoly from evaporating because the hardware market evolves in a new direction.
When the eee PC came out, Linux had a big opportunity. Unfortunately, Asus completely blew the details of the implementation. They picked a sucky distro, and they did a lousy job on quality control and integration. My wife uses linux on the desktop, and when she saw the eee at Target for $280 she asked me to get her one for her birthday. The model they were selling at Target was out of date and not very good, so I ordered a fancier model on amazon for $400. It came with its wifi misconfigured, and Asus tech support told me they couldn't fix it, and I'd have to return it. This was a few months ago. Yesterday I was making a trip to Fry's, so my wife suggested I just buy one while I was there. Well, Fry's is now selling the eee only with Windows, and Amazon's site also doesn't have the linux version available. AFAICT retailers were just getting too many returns of the linux ones. You can pretty much tell what was going on based on the amazon reviews. Some, like mine, were being shipped misconfigured. In other cases, you had people buying the linux version and not understanding that it wasn't windows. And in still other cases, people were buying them with linux and then trying to install a (presumably illegal) copy of Windows, and failing. (None of this is new, either. All this stuff happened in the past with the Great Quality linux boxes they used to sell at Fry's. The hardware was in fact great quality, but Fry's was getting too many returns, so they stopped carrying them.)
I think the basic problem here is that it's expensive and difficult to do a good job integrating hardware and software for a consumer computer. That's the kind of thing Apple is famous for doing well. Apple puts a lot of money and effort into it, and they charge for it when you buy a mac. I just don't see how anyone is going to do anything like that in the netbook market, which is an ultra-low-margin market. It would have been especially difficult for East Asian manufacturers like Asus and Great Quality, which have a language barrier to deal with. (At one point, Great Quality was shipping their machines with a linux distro that didn't even have an English-language web site.)
Meanwhile, MS can afford to do what it takes to maintain dominance in all sectors of the market. MS doesn't even have to do a good job on netbooks. They just have to avoid doing such a horrible job that it becomes painfully obvious to people who have never used anything but Windows before. It's possible that ARM-based netbooks will change the equation, but I wouldn't be surprised if MS jumps in and starts competing vigorously on ARM, simply to maintain their monopoly.
Yep. One of the most important experiences that made me stop using proprietary software was with Mathematica. I bought a copy of Mathematica and installed it on my mac. This was back in the 90's, so I was running MacOS 6 or something. I upgraded to MacOS 7, and Mathematica would no longer run. Called Wolfram, and they told me I'd have to pay for a new copy of Mathematica.
There are several different issues here.
First there's the fact that PDF is almost completely nonproprietary (i.e., the functionality required by the vast majority of users for reading and writing PDFs is nonproprietary), whereas flash has a lot more proprietary stuff in the mix. However, the proprietary aspects of flash mostly have to do with patented audio and video codecs, which are irrelevant in the case of scribd. A lot more of flash used to be proprietary. What Adobe's been doing is to open up flash a lot more, while still keeping enough of the tasty bits proprietary so that nobody else can really compete effectively with them to sell development tools. There are, however, open-source apps for both creating (haxe) and running (gnash) flash. If the vast majority of flash developers don't see haxe as a viable alternative to Adobe's toolchain, and the vast majority of users don't see gnash as a viable replacement for Adobe's plugin, I think that basically just shows that the OSS community hasn't made it a priority.
Another issue is that there are certain types of documents for which html simply isn't the right tool. For example, if you want to make a printer-friendly, illustrated physics textbook, pdf is the right tool for the job, not html.
As far as the relative merits of displaying PDFs in a flash application or a PDF plugin, I think for the average user scribd's flash app may indeed be the more convenient and useful choice. Most people use Adobe Reader as their PDF plugin. AR is slow to load, and has a horrible history of security problems. Most people aren't propeller-heads, and don't know that there is any alternative to AR. For a large document, scribd's flash app also has the advantage that it loads without having to download megabytes worth of PDF data. For someone on a dialup connection, this is a big deal. Although it's true that the PDF standard has advanced features designed to allow the user to see individual pages without having to download the whole file first, in reality I don't think that's implemented properly in the majority of PDF documents you see out there in the wild, and I don't think it's implemented properly in open-source readers like Evince, either. (Evince has truly horrible performance on PDFs consisting of many scans of bitmaps.)
Keep in mind that scribd's killer app is, frankly, copyright violation. I'm sure they don't intend it to be used that way, but that's the big thing that users are accomplishing with scribd that they can't accomplish in any other way. For instance, search scribd for "feynman lectures," and you'll get a whole bunch of PDFs consisting of pages scanned from a copyrighted book. The performance of scribd's flash interface on this type of PDF file is much, much better than what you'd get with straight PDF.
On my machine, it's 22 cm x 16 cm.
A lot of people are posting about how much they hate scribd's UI, but I don't see that as the big problem with scribd.
People have posted some of my books on scribd, and that's fine with me, because the books are free-as-in-speech. However, their system has some problems. For instance, if you search on scribd for "Newtonian Physics," which is the title of one of my books, the first 8 hits consist of 8 different uploaded copies of my book. Seems like a lot of scribd users don't bother checking to see whether something is already on scribd before they upload it. Now if I type in some text from my book as a search, only a few of the books come up, not all 8 -- don't ask me why. And when I click on the #1 search result, it's a version of the book from 2001, with an incorrect description and an incorrect license listed for it.
I think the fundamental problem here is that they're not serving one of the traditional purposes of a publisher, which is to act as a filter. Filters can be good or bad. A filter doesn't have to be all-or-nothing, and it doesn't have to be elitist or authoritarian. Google page rank is a filter. Slashdot's moderation system is a filter. Scribd doesn't seem to have enough useful filtering mechanisms. It just seems to act as a huge dumping ground, where anyone can put anything. The trouble is that finding anything there is like saying, "Huh, I need a new cartridge for my antique fountain pen. Maybe I'll go down to the town dump and dig around for one."
The Fiction Circus article linked to from the slashdot summary isn't as good as the NY Times article that it links to.
No, not really. One reason it's not big news is that scribd is currently too small a commercial entity to make any difference in this big marketplace. Another reason it's not big news is that other people are already selling digital books without DRM. Fictionwise and Baen are two examples that come to mind.
Well, no. Amazon is a huge, profitable business that readers know about. Scribd isn't. That's a pretty strong incentive for writers to deal with Amazon -- or, more accurately, it's a pretty strong incentive for their publishers to. The author generally doesn't make any decisions about the distribution channels through which a book gets to the public.
TFA is wrong in three different ways: morally, politically, and practically.
Its proposals are morally wrong because they trample on individual freedom. For instance, #1 says "Any individual shall be able to choose anonymity when posting to Internet sites." Okay, I run a site that catalogs free books, and accepts user-submitted reviews. When new users register, it's made very clear to the new user that he needs to provide his real name. The reason is that I don't want people reviewing their own books. No, I'm not under the illusion that 100% of people are truthful about this. However, the author of TFA seems to be proposing that the government prohibit me from even setting this as a policy on my site. That's absurd. If users don't like the fact that I require them to give their real names, I can't force them to visit my site.
Politically it's wrong, because he seems to be completely clueless about the nature of individual rights in a constitutional democracy. Individual rights are not about telling A that he has to give B something, or forcing A to associate with B, or telling A that if he wants to do business he needs to act in a certain way. Individual rights have to do with being left alone, especially with being left along by the government. You can believe that the government has other legitimate functions besides safeguarding individual rights, but that doesn't mean that those other functions qualify as individual rights. If you think government should be inspecting cuts of beef to make sure they won't make people sick, then that's a perfectly legitimate, mainstream political opinion in the U.S. -- but it has absolutely nothing to do with any supposed individual right not to get sick.
And finally, he's wrong in practice, because these ideas are all just stupid in the context of, say, the United States in the year 2009. Taking his items one by one:
This kind of thing is a minefield, and very hard to prove. To see what I mean, do a google search on "cannibalism anasazi." People get emotional about certain scientific issues, and often the reason they're so emotional is that there's painful history involved, and/or a history of the misuse of science. For instance, it's theoretically a reasonable scientific topic to look for correlations between race and intelligence -- but if you try study it, you'll unleash such a shitstorm that you'll wish you hadn't. Part of this is because the topic isn't PC, but part is also because of history (eugenics, Nazism, Cyril Burt).
Cannibalism has historically been one of these scientific issues that are just hard to study because emotions run too high. For instance, you have the history of Europeans portraying Africans as savage cannibals (which made it easier for Americans to justify slavery, and for the Belgians to justify cutting people's arms off in Congo).
Some archaeologists and anthropologists have gone so far as to claim that cannibalism simply doesn't exist, and never has. Others have found physical evidence that they interpret as evidence of widespread cannibalism in certain societies. Still others say that it exists, but only in a ritualized form.
I'm not convinced that the chances are very good of coming to a definite conclusion about cannibalism that might have happened hundreds of thousands of years ago, when we can't even study the more recent cases.
That's incorrect. Standard ZFC is sufficient for NSA. You may be thinking of Internal Set Theory, which is one way of approaching NSA, but postdates NSA by several decades.
One story I've heard is that Keisler's course was disorganized. The book was handed out on purple mimeograph sheets, and the first 20 minutes of every lecture was filled with a grad student reading off errata, which the students were supposed to correct on their copies.
The information here seems to indicate that NSA is educationally beneficial: http://en.wikipedia.org/wiki/Influence_of_non-standard_analysis#Education
I disagree. Actually, if you want to understand the plain old real number system deeply, it involves quite a bit of abstract math and set theory. For example, if you look in a freshman calculus book, they never prove things like the intermediate value theorem, because it requires a deeper understanding of the foundations of the real number system than a freshman in college is going to have. For an example of how deep and strange the theory of the reals can be, see Chaitin's number.
Nonstandard analysis isn't harder or deeper than calculus using limits. It's really just a different language for talking about all the same ideas.
I also disagree with you about "rote manipulations." Keisler actually does quite a good job of teaching the conceptual side of the hyperreals. Euclid went to his grave without knowing Cartesian coordinates; that doesn't mean he didn't understand geometry. Gauss and Euler used infinitesimals without ever learning about the hyperreal numbers; that doesn't mean they didn't understand infinitesimals. The Keisler book gives a very nice, self-contained axiomatic treatment of the hyperreals, in much the same way that the Elements gives a self-contained axiomatic treatment of geometry.
Here is a catalog of a few hundred free math books.
A drunk driver could kill me. I don't think $10,000 is excessive as a deterrent. My life is worth a lot more to me than $10,000.
A .08 BAC is nowhere near non-impaired. For instance, this DOT page states that
"At a BAC of .10, the probability of a fatal or serious-injury crash was estimated to be 6 to 12 times that of a driver with no alcohol."
When you use words like "farce" and "ridiculous," it makes it sound like you don't want to take responsibility for your own actions. I don't think DUI laws are "a farce, intended to make money." I think they're intended to protect people like me from getting killed by people like you.