Also, the paper is written in a "scientific manner". That is, although he may be trying to show that we should have copyrights are too short, he did his best to prove himself wrong. For *every* simplification, he shows that the bias it introduces tilts things against himself. For the numbers he chooses, he chooses quite conservative ones, that would imply longer copyrights.
If you are surprised by the 14 year number he gives as a point estimate, really, you just don't understand exponentials. The value of longer copyright terms goes down fast, and even with relatively low exponents, you reach break-even quite fast. Just because a small handful of works maintain their value so well as to warrant massive lobbying, the vast majority of creative works aren't worth the bother to continue printing after just a few years. Consider books; except for bestsellers, they have a really low self-life (pun intended)--only 1% of books ever had their copyright renewed at the 14 year mark (back when that was required). Just 1%!
He also gives a range of estimates, including some ridiculously low discount rates (2% is obscenely low...). The *highest* copyright term he comes up with is 51.51 years. Now, it seems to me that 50 years is quite a bit shorter than life+70...
3. But generally, all models simplify, often over-simplify, the thing they describe. While true, the author was quite thorough in trying to limit the effects of his simplifications. He lists quite a few complications, and then demonstrates that a) they have small effects, and b) any effect they do give is a bias towards longer copyrights. That is, for the inaccuracies that he considered, the simplification makes his math give *longer* terms of copyright. For example, he uses a power law estimation of the distribution of sale of works. This is a quite good approximation, but it is known to somewhat overstate the importance of less popular works. However, such overstatement leads to higher copyright estimates. His simplifications cannot be used to argue that his estimates are too short; instead, they are tilted towards too long.
He was also a good boy in providing more than just a point estimate: for the hardest to estimate numbers (the discount rates), he provides several examples on both extremes. Note that the most extreme example (with a ridiculously low discount rate of 4% and cultural discount rate of merely 2%) he gets 51.51 years--this is well under life+70. Even the point estimates he gives use quite conservative numbers, and are tilted towards media with low cultural discount rates--he uses song data (songs "last" a long time & don't have much cultural decay), while simultaneously mentioning that written works have much much higher cultural discount rates (which would argue towards quite short copyrights).
Overall, the author dotted his i's and crossed his t's. Although a 14 year point estimate is certainly controversial, just because it is controversial doesn't make it inaccurate. Further, the paper provides a very persuasive argument that current copyright terms are too long, and that anything over 50 years is socially suboptimal.
which is the point of the article. That is, society as a whole was better off for Spiderman being made into a movie. Sony & Marvel, of course, got money. It is fairly clear that they feel that they are better off, since not only did they make a Spiderman movie, but then they made two more.
Now, you may say "well, for every dollar Sony got, Joe Public had to spend one", and that it really just balances out. But that is a fallacy. Joe Public saw that the price of a ticket was $10, or that you could rent it for $4, or buy it for $15, or whatever. Joe Public did so. And, although some people wish they hadn't, most people feel that they got some good out of seeing it, or owning a copy. When you buy a $10 movie ticket, you admit that you think you will be better off with $10 fewer and having seen the movie. Of course, you may be dissapointed, but judging by the number of people who went and saw the second and third movies, it looks like people got a bargain.
The article talks about this in economics terms: it's called surplus. If I am willing to buy something at a certain price, that is because I think it is more valuable to me than the money. The difference between what it actually costs and what I feel it is worth is the "consumer surplus". For some people, it is low. For a product of somewhat unknown quality, it may even end up being negative. But, for the most part, the surplus is positive. The "producer surplus" is similar, although you are probably familiar with it by a more mundane word: profit. These two surpluses are the only reason why people buy or sell anything: because both sides feel that they will be better off for making the transaction.
Now, this doesn't mean that society would have been better off if the original Spiderman copyrights had expired, and the movie wasn't made. Perhaps the combined surpluses would have been larger if we'd all gotten cheap copies of the old comic books. But, given that the movie was made, it *did* leave society better off.
Although, if Spiderman was public domain, Sony still could have made a Spiderman movie; they just wouldn't have had to pay Marvel as much. All other things the same, with lower costs they likely would have done a combination of lowering prices (to increase demand) and pocketing some of it. Which would have increased the wellbeing of *both* Sony and moviegoers.
There are other reasons to consider payola immoral, but there is a straightforward reason: if the DJ's only spin songs they've been payed to play, the those who can't pay won't get paid.
Simply put, payola keeps small artists and those without the backing of a well-monied party at a distinct disadvantage. The major labels certainly form an oligopoly, and, cartel or not, they have maintained their oligopoly through 1) control of the distribution chain, 2) buying out the supply of new talent, and 3) through squeezing small players from the most effective publicity channels. #1 is threatened by the internet, and is their largest problem right now. #2 is the fault of bands stupidly signing disadvantageous contracts; to a mild extend newer bands are wising up, though. #3 is still an issue. Payola is the direct way of doing it, and gave the majors their initial dominance. Nowadays, it is a little more discreet; "independent promoters" get money from the majors, and then they in turn turn over "stuff" to radio stations (stuff ranging from blatant cash bribes to concert tickets to give away through on-air contests). Direct or not, payola floods playlists with songs from well-funded labels, at the expense of smaller labels or self-produced bands which do not have the resources to buy their way onto playlists.
There is an exception; a record label can straight out pay to get a song played, but the radio station has to disclaim that it is a pay-for-play, and the amount of airtime devoted to pay-for-play is limited by law (I believe it may be by considering such to be advertising; and radio stations are limited in the fraction of airtime which is advertising). This sort of payment is probably unproblematic from a legal and a moral standpoint, unless playlists are influenced by who is buying advertising (which would essentially be old-skool payola again).
Well, a few (honest) questions, & some comments...
Hello - have you used Office 2007?
Well, it's only been out 4 months. I based everything I said off of previous incarnations of Office, from 2003 (the second latest-n-greatest) to 95 (when WordPerfect started sucking). Given I don't want to replace my shiny & new copy of Office 2003 so soon after I got it, I won't be purchasing 2007 for some time. And nor will I be using to submit to Nature or Science, apparently, even if it were free.
1. PDF publishing is supported (free)
That's good to hear, and long long missing. Questions: is the output of "Print to PDF, then print the PDF" the same as just printing? I've had issues with such software producing different output that way (even sometimes products made by Adobe!). Also, how "good" is the PDF output. That is, are the files sizes quite small, is it embedding proper scaleable fonts, and does it print fast? A big problem with the old "print to.PRN, change extension to.PS, then ps2pdf" way of going from Word to PDF was getting bloated, poor quality, and complex PDF files; even Acrobat sometimes will non-sensibly make a crappy PDF.
2. Citations and Bibliographies are both supported under Word
Having used BibTeX, I will never go back. There are hugedatabases of freely available BibTeX format citations. The second runner up, EndNote, hasn't nearly the amount of citations available (although importing BibTeX into EndNote isn't hard). I have used EndNote, and it is not nearly as good as BibTeX. Other people (read the comments) seem to feel that EndNote works better than Word 2007's support; I simply don't believe that a mouse-driven interface for adding citations can ever beat a text-based one.
Office 2007 documents can be saved to document managment servers for sharing
I think you're missing the point. LaTeX easily allows you (and encourages you) to split your documents up into multiple files. So, regardless of what collaboration service you use, anything from emailing files back and forth to something overly complicated like SourceVault, you can have multiple people editing the same document simultaneously. They just work on different sub-files. I've done it with email, but usually do use RCS to automate the locking support. Unless Office 2007 vastly changes things, Word documents are still monolithic files. That makes it quite difficult to support simultaneous editing; you *need* a concurrent versioning system. And, forgoing large changes in Office 2007,.doc files are still stored as BLOBS, which makes automated commit/merging difficult to impossible.
4. LaTeX has style files; Word has templates. What's the difference?
As best I can tell (I've never found it, and I've tried...) you can't apply a template to a document after the fact (and get the expected results). If my paper is rejected from one place, I can reformat the entire document by changing which style file I include. Style files pretty much guarantee that all of the final product will have a consistent look, and that said look is easy to change across the board. Journals that are either done all in LaTeX or those that hire separate typesetters (for mucho $$$) have the most consistent appearance. Others look a bit like just a bunch of papers glued together.
I can't claim Office 2007 is better than LaTeX, since I've not used the latter extensively,
Except for those on the Office 2007 team, *nobody* has used Office 2007 extensively:)
From talking with people who have used 2007, there is quite a learning jump to go from 2003 to 2007. Especially since 2007 breaks math support for journals, it makes sense to consider moving to LaTeX just a
I could use LaTeX like he does, but I don't really have a need for it. This smells of the 80/20 rule in Office: 80% of the people only use (or even know about) 20% of the functionality. Being good at formatting math is just part of the functionality of LaTeX. There are lots of things that LaTeX is good at (and Office stinks at), which would probably be quite useful to you.
PDF Publishing. Nowadays, if your paper isn't published on the web as a PDF, it isn't nearly as influential. Some people won't even bother reading it. Now, if you're on a Mac, publishing to PDF is pretty good, but no such luck for Windows. You have to buy Acrobat separately.
Citations and bibliographies. Hmm, LaTeX comes with BibTeX; I haven't manually generated a citation or a bibliography entry since forever. I don't even have to write the BibTeX; just Google "paper title + bibtex" or check one of the standard online sources. Office has minimal support for bibliographies; guess you'll have to buy EndNote.
Support for concurrent editing. LaTeX lets you split up your sections into multiple chunks; indeed it encourages you to do so. So I can be working on one section and my coauthors can work on others. Now, truthfully, we *do* use RCS to synchronize, but we've done it without it. Or you could be fancy and use CVS or SVN and have concurrent editing of the same file. Word, um, can't do that. At all.
Automatic formatting. It always amazes me that people are willing to fight with Word to get their document to meet a formatting requirement. In LaTeX, I just download the style file provided by the conference. Now, Word does have templates, but they seem rather fragile to me, and while Science and Nature may give templates (I don't know), some other journals do not.
Really, Office isn't nearly as good as people make it out to be. And LaTeX isn't nearly as hard; especially if you use one of the WYSIWYG editors.
As other posters have pointed out, the editing that goes into a digital file is worth something; but the print itself is worth very little. People know that they can get a digital print from Walmart* for under a quarter; four bucks from your wedding photographer, which is sixteen times as much, is clearly obscene. If I want a lot of prints, your traditional wedding photographer is clearly crap crap crap for value.
Now, even after charging for your time massaging the files, the prices you charge should be lower than they were 20 years ago (in inflation-adjusted terms). Think; 20 years ago you *had* to be shooting film. Your cost of producing prints included the film, the development, carrying many cameras with different speed film, etc. If you wanted to be sure to catch those perfect moments in pictures, you had to blow a *lot* of film, and a lot of processing. Nowadays, it's just the incremental cost of a larger memory card; that's *much* cheaper than film. The massaging that you can apply to a digital file used to be done in the darkroom, in a difficult labor intensive process. Nowadays, Photoshop makes it a lot easier, faster, etc. *Your* costs of producing good pictures have gone down; it makes sense that your prices should drop too.
I've noticed that the older, more established photographers still do silly things like charge extra for B&W & sepia-toned prints; they also refuse to release digital files. Now, the really silly of them are still shooting with film (and frankly, deserve to be shot**). They are a dying breed. But for those charging large differentials for B&W when it takes all of 30 seconds to toggle it in Photoshop, please, tell me how that makes any sense at all. Really, they're just still clinging to a dying business model.
Really, do yourself a favor and switch to charing for the time you actually spend. Charge for the labor of generating an album, but charge what it actually costs you to get it bound for the album itself. Charge for touching up photos, but charge what it costs you to get prints for the prints. By charging what reflects your actual costs, you'll be able to attract more business. If you are artistically good (either in the shooting or the editing or both), you can charge a premium for those services, and people will make lots of prints of your beautiful pictures (at no cost to you). Your customers will be happy, they'll tell their friends (if they can get cheap prints of your pictures, *they* will pay to give them to their friends--free advertising), you can charge ever more for your time... it's a virtuous cycle. Or, you can continue charging obscene amounts per print, and people will feel that you're ripping them off. Other photographers will offer what people feel to be more fair price structures, and you'll start loosing business.
Remember, in the end you can't force people to accept your terms. If you *seem* unfair, they'll balk, and then they'll walk.
*You make fun of Walgreens, but the quality of their prints has been steadily increasing. Reputations for bad quality are a legacy of poor film processing; their digital stuff is not bad. And Walmart, disreputable as they may be in many areas, actually produces really nice prints, with consistent color, clarity, contrast, etc. Their film processing leaves a bit to be desired (dammit, stop pushing my low-light shots!), but their digital is great. Kodak does good digital prints online for cheap too. If the raw files are put through even basic color correction, the prints will look good almost no matter who does them (Snapfish excluded because they suck).
**I still use film myself, but not professionally. I do it because it's fun; anyone shooting film professionally is either an idiot or has special needs (e.g. IR, or ultra high/low ISO).
This does at least insulate us from the rising cost of oil for the time being, since oil is traded in US$, and US$ is falling almost as fast as the price of oil is rising.
Well, consider what percentage of a liter of petroleum costs in either EUR or GBP is caused by flat taxes; the end price won't change much at all compared to the price of the underlying commodity.
Further, that the dollar is falling and that the price of oil (as denominated in USD) is rising is really the same thing. For a fairly fungible commodity like oil, rather than thinking of the actual number, it makes more sense to think in terms of how much it will buy. Since oil is priced in USD, and the real purchasing power of the USD is falling, it makes sense for the price to increase. Notice that since the US CPI is still quite low (under 2%); in terms of real goods, a barrel of oil will get you more stuff. The price really is rising.
Another way to look at it is in terms of Germany. Germany is 1) terrified of inflation, and 2) dominates the EU economically. The Euro, therefore, is tilted towards deflation if anything else (this is in general an argument against using the same currency in a such a large and diverse area). So, the purchasing power of the Euro should likely be increasing slightly (or, in terms of exchange rates, it should be rising). But the price of oil in terms of Euros is rather flat; so in terms of other goods, the price of oil is still increasing. That is, what a barrel of oil can actually buy you is increasing.
Or we can look at the Yen; Japan has been experiencing near zero inflation to deflation for quite a while; yet the price of oil in terms of Yen is still increasing. Although not nearly as dramatic as the increases compared to USD, one can still make an argument for the actual value of a barrel of oil increasing.
If the price a commodity is nearly flat in terms of a currency, then what is to blame is that the currency is gaining in strength; not that any specific currency is falling. You are still paying more for that commodity in real terms (that is, other goods); but that the numeric value is flat is good at fooling you into thinking that it's "price" is constant.
The sad thing is they wouldn't have even needed to buy the machines. You can "buy" them, use them for a short while, then return them, minus some small depreciation amount. Big retailers (e.g. Macy's) do this for the Christmas shopping season--and the big machine vendors (e.g. IBM, HP) are happy to oblige, since if the machines are used for a sufficiently short amount of time, they still count as "new", so they get to sell them again, full price, to someone else.
Or, another cheap and easy way is software-disabled hardware. Buy a blade system and have a large percentage of disabled (but present) blades; pay for them and turn them on only if you need them. Or you can do the same thing with most mainframe systems. Nearly every mainframe system ships with additional processors installed but disabled; IBM even has this thing where they skip every other clock cycle until you pay for the upgrade. Intuit was really quite short sighted about this. Obviously there is going to be a huge load spike right before the deadline. They could have easily allotted themselves a large over provision amount for not that much money.
Top of page 9 has a chart comparing native Linux, Xen, VMWare, and UML for different workloads. They show VMWare degrading performance by over 70% for SPECWEB 99.
Web applications are OS intensive; while VMWare is quite good at pure CPU-bound tasks, it has to perform a lot of emulation whenever you are running inside the OS. So it will stink at anything with lots of small IO, lots of metadata operations, or lots of process creation/switching. For example, VMWare shows a whopping 90% slowdown for OLTP database workloads, according to the Xen paper, and it really isn't surprising. The OS microbenchmarks in the above paper (page 10) show that VMWare has abysmal performance for things like fork(), exec(), mmap(), page faults, and context switches.
Basically, Xen doesn't have to emulate the OS, because they make modifications to the OS. VMWare does dynamic binary rewriting (think fancy emulation) to run an unmodified OS; they therefore pay through the nose in performance overhead for OS-intensive workloads.
And even there, there is a workaround that can be employed with the use of a 3rd party that doesn't block incoming connections (though I haven't heard of any P2P protocols currently use this method in the wild).
Skype (which, coincidentally, was written by the same people who wrote Kazaa) uses some of those workarounds to punch through NAT firewalls. I do not know if Kazaa uses them, but the authors of Kazaa could have certainly done so.
The point of all this being, you can share files, without accepting inbound connections. You can download files from others without accepting inbound connections. And you can participated in the P2P network (communications, searches, etc) and all of the above, without your P2P program knowing your public IP address.
But P2P works better if it has access to your public IP address, and you can accept inbound connections. Hence some P2P applications will complain if they detect that they are NATed and ask for your public IP. Some will auto-detect you public IP. Others will not only autodetect your public IP, but if you have a UPNP-capable router, will automatically detect or otherwise set up appropriate holes in you NAT firewall (later versions of Azeurus do this, I believe) to forward inbound connections. And, as aforementioned, Skype uses NAT-busting techniques to bypass setting up proper forwarding rules altogether. Skype's ability to get past firewalls is actually somewhat frightening...
Although the original design of the internet was based on the assumption of a static one-to-one mapping of computers to IP addresses, this is not the case today. DHCP means that the mappings are not static, and NAT means that the mapping isn't one-to-one (indeed, a sufficiently sophisticated NAT setup could be many-to-many, although such would be unusual). Even MAC addresses aren't really unique--it is quite common to set up interface failover by spoofing the MAC address of the failed NIC. Identifying a computer uniquely is a very tricky process--the common means of doing so rely on these broken assumptions. The uncommon means (specifically, searching for evidence of clock drift in timing parameters) are, well, not commonly used, and have higher false positives (due to sensitivities to temperature and the low precision of clock drift measurements). And none of this can be used to show that a particular person was doing anything at any point in time.
From my limited experience with expert testimony, many expert witnesses, although experts in their field, are not experts at being witnesses. It's a way for a university professor to pick up more money on the side with easy consulting work, especially if hired by a petitioner under the expectation of a weak defense by the respondent. In such a case, speed and cheapness are prized above thoroughness and accuracy, and actually being deposed by a lawyer who has been prepped on the sorts of questions to ask would be quite the surprise. Dr. Jacobson appears to have been caught with his pants down, giving a slap-dash report which is clearly biased in favor of the side which hired him. Although he isn't a member of any regulatory body, I would be surprised if he wasn't a member of the ACM or the IEEE Computer Society, and in violation of their respective codes of ethics (specifically, ACM 1.2, 1.3, and 2.5, and IEEE 2, 3, 7, and 9).
At least on/., the Google study is getting all of the press. But there were two large studies on disk failures this year at FAST: the one by Google, and the other by the Carnegie Mellon Parallel Data Lab. It won best paper. You can find it at http://www.pdl.cmu.edu/PDL-FTP/Failure/failure-fas t07_abs.html
Although the CMU group didn't have environmental data to see what correlations there were, they get results similar to Google in other areas; e.g. that drive "vintage" matters, the failures are not a Poisson distribution, etc. Since the two studies agree, it implies that both are reliable. Also, most people *don't* have a heat chamber to weed out flaky firmware behavior--the drives that almost everybody runs are drives that may have such firmware games enabled. The studies are on disk drives *as they are shipped from the manufacturer*; the correlations found, whatever their reasons, do exist.
Yay! I get to recycle an old comment, just like drug manufacturers recycle old compounds! Unlike the drug manufactures, I include extra content, and all at no extra charge.
Some other examples of "bad" patents include:
1) Obvious compounding. A good example is pain medication. Acetaminophen (Tylenol) has an unusual method of action which is synergistic with nearly every other analgesic, and rarely interacts with other drugs. So, the drug company will file a patent on their new painkiller, and then (just before the patent is made public/the drug is approved), they'll patent mixing it with acetaminophen. Doctors prefer prescribing the mixture because it has a perceived lower risk of abuse (due to the liver toxicity of acetaminophen), so the generic unmixed version isn't used so much. A specific example is Ultram vs. Ultracet (although both are now available generic, there was a period when Ultracet was not).
Going back to Claratin, Claratin-D is just Claratin plus pseudo-ephedrine (Sudafed). Yes, lets compound an antihistamine with a decongestant... never would have thought of that one.
2) Racemic mixtures. Many drugs have left handed and right handed versions. Often, one version or the other is more effective/safer. Especially since the thalidomide incident (anti-nausia drug where one version (left?) caused birth defects) testing both versions is standard. Yet the drug companies can get separate patents on the left, right, and mixture versions. Sometimes, the patent on the left or right can be used to control the mixture, especially if it is difficult to make just one version or the other. Regardless, it gives the company a "new" drug to market and to compete with the generics. Prilosec and Nexium are an example of this; Prilosec is mixed and Nexium is not. There is little evidence that Nexium is any different from a 2x dose of Prilosec (which have the same amount of the active handedness).
3) Particle size patents. Hmm, it just so happens that a certian size granule is "better" than others, and the standard manufacturing technique (whose patent is expiring) makes that particle size (or at least contains it)... Provigil is currently being protected under a particle size patent, and will for 5 more years, even though the formulation patent expired last March.
4) Time release/enteric versions. Coating something (with a standard, commonly used coating) to make it time released or gentle on the stomach isn't obvious, for some silly reason. Sorry, no example of this one.
5) Dosage. Yes, taking your pill in a different dose can afford protection against generics, although I believe this is at an FDA regulatory level rather than as a patent (finally! We found an obvious patent). Flexeril is a muscle relaxant, and the usual dose is 10mg. If you cut that to 5mg, you get a much lower incidence of side effects, and it is just about as effective. For a while you couldn't get a 5mg generic, although it seems that you can now. Frankly, rather than spend $52 for 30 brand name 5mg pills, I'd rather pay $8 for 30 generic 10mg pills and $5 for a pill cutter. That is $.21 per dose for the generic and $1.73 for the brand name.
Even now--with this rigorous testing--we find that some drugs should never have been sold. Vioxx comes to mind.
And what is so bad about Vioxx? If you are allergic to sulfa, need to avoid gastrointestinal side-effects, and are at a low risk for heart disease anyway, Vioxx is an excellent choice of drug (Celebrex has a sulfa group, NSAIDs and aspirin cause stomach bleeding). Especially if you are going to be taking it for a medium term (12 months or less) there is little risk. The studies that show statistically significant increases in adverse cardiovascular events show them for long-term high-dose use; many of them are only borderline significant. Both the US FDA and Health Canada have recommended it be returned to market, and it appears that aside from the lower bleeding risk, Vioxx is no riskier than ibuprofen.
There are lots of drugs which are far more dangerous than Vioxx (say, anything with lithium in it), yet continue to be sold because they serve a therapeutic niche. I am quite disappointed in Merck for two reasons: first because they took an unreasonable and unwise approach to the evidence of cardiovascular risk in Vioxx, and second because now they refuse to manufacture Vioxx and sell it with appropriate warnings.
Don't forget all of the other stuff you get from spalling, like high energy X-rays. Actually, at 300MeV, I'd wager on getting a fairly decent gamma ray beam. Without a purpose built collimator, I'd guess that there'd be a good amount of "spray" all over the place. So you'd probably get a hole in a radioactive sandwich, plus a good dose of radiation just for standing nearby. Yes, a healthy dose life-giving radiation.
Remember, the Therac-25 system was quite lethal when it malfunctioned, and it "only" used a 25 MeV beam. 300MeV is a LOT of punch per particle, and if the intensity is high enough all sorts of nasty things will happen.
They sell *most* of the parts from old Thinkpads. They discontinued selling replacement batteries for T4x series machines; rather irritating because the T43 model was only introduced early last year, and now you can't get batteries for them straight from Lenovo.
On the other hand, you can get most of the parts, and the batteries are still in supply in the retail chain; I just ordered one from Newegg. IBM was absolutely stellar in terms of support; Lenovo seems to be not quite as good. Much much better than the other big vendors (like, >shudder, Dell), but not quite the same as big blue was.
wouldn't this sort of thinking result in the entire hard drive ending up as 0.5--or some random noise--rather than a meaningful data signal?
If "old value" starts out between 0 and 1 and "new value" is always either 0 or 1 (plus or minus minor noise), things will work out. If I write a 1, then you get (.9 * 1 +.1 * something_between_0_and_1) =.9 + something_small = something_between_1_and_.9, while writing a 0 will get you something inbetween 0 and.1. Even if, somehow (say, lots of unluckily correlated noise), the "old value" is corrupted to something out of range, (e.g. 1.1 or -.1), the average will pull it very quickly back into range. It isn't a linear average; the most recent result matters most, and dominates enough that you'll have something distinguishable one way or the other. If, somehow, the data was written to be.5, then you'd get (1 *.9 +.5 *.1) =.95 for writing 1, and (0 *.9 +.5 *.1) =.05 for writing 0, which is actually closer to a perfect 1 or 0 than if the previous write was the opposite.
Although, do note that rather than using actual 1's and 0's, we have analog field polarities which we let stand for 1's and 0's... the idea is the same, though.
I've always wondered, if this were really true, why we don't see random errors cropping up constantly especially on heavily used portions of hard drives.
The very error correction that makes it easier to recover. The writing will flip the magnetization most of the way there, so you can tell if it was a 1 or a 0, it just won't be exactly 1 or 0. Think of writing as "old value *.1 + new value *.9"; you never really destroy the old values, but you make them weaker and weaker every time. Plus, for the cases where the read head is mistaken, the ECC saves your butt.
Is there a similiar random misalignment with the read head and, if so, why again do we not observe daily errors on heavily used portions of hard drives? If not then how does the read head compensate for the misalignment of the write head?
The write head is narrow (to avoid accidentally stomping on adjacent tracks). The read head is wide, to catch what was written even if they heads aren't lined up exactly where they were last time. You won't be off by a whole track, just a fraction of a track. So you don't completely overwrite the width of what was written last time, but you'll overwrite the vast bulk of what the read head will be picking up next time.
The magnetic field on the drive is what it is--it has no way of knowing if it is being read for recovery purposes or for standard reading.
But the specialized reads head/drive electronics do know that it is for recovery. A normal read head tells you "1" or "0". Forensic electronics save the analog information which a normal read head discards. Then they "subtract" what the ECC was telling them was written most recently, to get a weaker version of the old data. Then you see what the ECC tells you what that should have been (although your overall noise is higher), and recurse. Eventually the old signal is so drowned out by random noise that the ECC no longer is enough, and the data is effectively lost. Although it *could* be used for everyday use, it would be much slower and you would have a much higher data error rate; it is a simple speed/reliability vs. capacity tradeoff. People get cranky when their hard drive corrupts data, and they get cranky when it is slow, so the manufacturers have made their choice.
Well, that you can recover data doesn't mean you can reliably recover it, or do it fast. Hence no 40TB drives.
As for how, the short version is that if you write a one and then a zero, you end up with.1; the old value leaks in a bit. When combined with the error correction codes built in to the drive, you stand a decent chance of recovering the overwritten data (slowly, with special read heads/drive electronics, and somewhat error prone). If you overwrite with the same fixed pattern repeatedly, you really don't improve the situation much (diminishing returns on removing the residual bit), so an 80x "wipe" of all zeros may be recoverable (although 80x is a bit much, even for all zeros). But overwriting to DoD spec is probably sufficient in this case, though. So is beating up the physical platters.
For the really determined attacker, then can use high resolution magnetic force microscopy (MFM). See, when you overwrite, the write head doesn't exactly line up with the old stuff--so you'll have little bits of the old data sticking out from above or below the track. MFM can resolve very localized magnetic fields, far smaller than your disk read head, and can see the misalignment. These misaligned pieces allow peaking back a few generations of overwrite, which allows you to subtract out the newer things (and hence clarify the older). Plus, it doesn't use rapidly spinning the disk, so it can work on beat up platters. Even without misalignment in the generations of overwrite, good MFM can resolve better analog detail as well, so the whole 1 then 0 =.1 thing works better. But it is far too expensive to bother with in most cases.
If the drive still works, Eraser (http://www.heidi.ie/eraser/) is a great choice. Wipe one file or every block device on the machine. Nothing will be able to get it back after a Gutmann wipe.
If the drive doesn't work, disassemble the drive, remove the magnets (to play with later), and apply a combination of sandpaper and a hammer to the platters. Yeah, magnetic force microscopy might still get it back, but who'd pay that much for your mostly worthless data anyway?
If you're sitting down to read through Knuth all at once, well, gads that is painful to try. They really aren't books to learn the concepts from, but rather serve as a reference for a wide variety of basic topics. Although there are better books on all of the individual topics, Knuth covers a lot in just three volumes. True, volume 1 is of limited practical use, but I've often referenced 2 and 3, for when I needed custom random number generation, custom hash tables, had to deal with unusual sorts, and as a general data structure reference. I feel confident that I'll use it for other things in the future. Knuth probably is of limited use to somebody who finds that they can always use the standard libraries, but if the STL Map classes are just too slow for your particular application, or if you need to sort a 10 GB set of data (whups, guess the system quicksort() isn't going to work... now what's the best way to do an on-disk sort?), Knuth is invaluable. As for being too dense, well, all of the big-O analysis and whatnot explain to you why the various algorithms and data structures work the way they do. If you don't understand how things work and why they behave the way they do, you are likely to make poor choices.
Second on the Knuth and K&R. Especially the Knuth -- the level of detail is beyond belief. If Knuth has anything about it, he has everything you ever wanted to know about it; like the 15 jillion different ways to do a hash table. K&R unfortunately is covering a fairly simple topic, so it doesn't get used as much (actually, use it for lending more than anything).
Besides Knuth, the book I use the most is "Computer Architecture, A Quantitative Approach", by Hennessy & Patterson, although it isn't adorning my shelf so much as much as my desk. If you care about the architecture side of things at all, Hennessy & Patterson is "the" book; plus the later editions cover relatively modern processors, like P4 Netburst. Not to be confused with Patterson & Hennessy, which is a more introductory book, and not nearly as well written.
A few other goodies are Numerical Recipes in C (for the rare cases when Knuth is lacking), and a lot of books by W. Richard Stevens, William Stallings, and Andrew Tanenbaum.
then one can extend it to a variety of areas which might be more useful than movie recomendations
I'd say, the odds are that this is going the other way. They had an existing technique, and then they extended it to movie recomendations. You don't need to offer researchers in data mining a price to get them to advance the state of the art in data mining; that's what they're interested in, and what they're payed for anyway. The prize just got them to apply it to movie recomendations.
The only thing to see here is that Netflix hadn't kept up to date on the latest and greatest in the academic literature. Otherwise, just move along.
It is the other way around. Let me tell a story...
If you die, and somebody is responsible (say, negligence, or recklessness), then your next of kin are entitled to compensation for your death. There's the whole "pain & suffering" bit, which is quite abstract, but there are also economic losses. That is, you, being dead, are no longer being paid, and that is money out of their pockets. Figuring this out is dreary economics and statistics, but it is quite measurable, and well presented economic damages are really hard to argue with, as opposed to those fuzzy pain & suffering numbers.
Of course, there are lots of complicating factors. You could have been fired next year. You aren't doing any housework either (and so someone must either be working harder, or hiring somebody; both are losses). Of course, dead people don't eat, don't wear clothes, need less housing, and don't drink or smoke; this money not spent is subtracted from the damages. Now, actually measuring how much the deceased ate, or that they were in solid with the boss, etc. are difficult things. So, instead, there are tables to look these things up in. Hundreds of tables, computed from census information, surveys, etc. Tables on everything. What the odds are, given age and occupation, of entering or leaving the workforce. What your raise would have been. And how much you eat, drink, and smoke.
The tables are split by family size and income. Richer people, for example, spend more on food; they eat out more, and buy nicer food. They of course spend more on travel, clothes, and housing. Most of these expenses start to level off after a bit, though. Interestingly, smoking is a decreasing expense; the poorest bracket spends less on smoking than the second poorest bracket (guess the poorest really just can't afford it), but spending decreases from there.
The expense that increases most linearly with income is money spent on alcohol. People with twice as much income sure don't spend twice as much on food, but they spend twice as much on booze. Cheers!
Slashdot Mirror
Also, the paper is written in a "scientific manner". That is, although he may be trying to show that we should have copyrights are too short, he did his best to prove himself wrong. For *every* simplification, he shows that the bias it introduces tilts things against himself. For the numbers he chooses, he chooses quite conservative ones, that would imply longer copyrights.
If you are surprised by the 14 year number he gives as a point estimate, really, you just don't understand exponentials. The value of longer copyright terms goes down fast, and even with relatively low exponents, you reach break-even quite fast. Just because a small handful of works maintain their value so well as to warrant massive lobbying, the vast majority of creative works aren't worth the bother to continue printing after just a few years. Consider books; except for bestsellers, they have a really low self-life (pun intended)--only 1% of books ever had their copyright renewed at the 14 year mark (back when that was required). Just 1%!
He also gives a range of estimates, including some ridiculously low discount rates (2% is obscenely low...). The *highest* copyright term he comes up with is 51.51 years. Now, it seems to me that 50 years is quite a bit shorter than life+70...
He was also a good boy in providing more than just a point estimate: for the hardest to estimate numbers (the discount rates), he provides several examples on both extremes. Note that the most extreme example (with a ridiculously low discount rate of 4% and cultural discount rate of merely 2%) he gets 51.51 years--this is well under life+70. Even the point estimates he gives use quite conservative numbers, and are tilted towards media with low cultural discount rates--he uses song data (songs "last" a long time & don't have much cultural decay), while simultaneously mentioning that written works have much much higher cultural discount rates (which would argue towards quite short copyrights).
Overall, the author dotted his i's and crossed his t's. Although a 14 year point estimate is certainly controversial, just because it is controversial doesn't make it inaccurate. Further, the paper provides a very persuasive argument that current copyright terms are too long, and that anything over 50 years is socially suboptimal.
which is the point of the article. That is, society as a whole was better off for Spiderman being made into a movie. Sony & Marvel, of course, got money. It is fairly clear that they feel that they are better off, since not only did they make a Spiderman movie, but then they made two more.
Now, you may say "well, for every dollar Sony got, Joe Public had to spend one", and that it really just balances out. But that is a fallacy. Joe Public saw that the price of a ticket was $10, or that you could rent it for $4, or buy it for $15, or whatever. Joe Public did so. And, although some people wish they hadn't, most people feel that they got some good out of seeing it, or owning a copy. When you buy a $10 movie ticket, you admit that you think you will be better off with $10 fewer and having seen the movie. Of course, you may be dissapointed, but judging by the number of people who went and saw the second and third movies, it looks like people got a bargain.
The article talks about this in economics terms: it's called surplus. If I am willing to buy something at a certain price, that is because I think it is more valuable to me than the money. The difference between what it actually costs and what I feel it is worth is the "consumer surplus". For some people, it is low. For a product of somewhat unknown quality, it may even end up being negative. But, for the most part, the surplus is positive. The "producer surplus" is similar, although you are probably familiar with it by a more mundane word: profit. These two surpluses are the only reason why people buy or sell anything: because both sides feel that they will be better off for making the transaction.
Now, this doesn't mean that society would have been better off if the original Spiderman copyrights had expired, and the movie wasn't made. Perhaps the combined surpluses would have been larger if we'd all gotten cheap copies of the old comic books. But, given that the movie was made, it *did* leave society better off.
Although, if Spiderman was public domain, Sony still could have made a Spiderman movie; they just wouldn't have had to pay Marvel as much. All other things the same, with lower costs they likely would have done a combination of lowering prices (to increase demand) and pocketing some of it. Which would have increased the wellbeing of *both* Sony and moviegoers.
There are other reasons to consider payola immoral, but there is a straightforward reason: if the DJ's only spin songs they've been payed to play, the those who can't pay won't get paid.
Simply put, payola keeps small artists and those without the backing of a well-monied party at a distinct disadvantage. The major labels certainly form an oligopoly, and, cartel or not, they have maintained their oligopoly through 1) control of the distribution chain, 2) buying out the supply of new talent, and 3) through squeezing small players from the most effective publicity channels. #1 is threatened by the internet, and is their largest problem right now. #2 is the fault of bands stupidly signing disadvantageous contracts; to a mild extend newer bands are wising up, though. #3 is still an issue. Payola is the direct way of doing it, and gave the majors their initial dominance. Nowadays, it is a little more discreet; "independent promoters" get money from the majors, and then they in turn turn over "stuff" to radio stations (stuff ranging from blatant cash bribes to concert tickets to give away through on-air contests). Direct or not, payola floods playlists with songs from well-funded labels, at the expense of smaller labels or self-produced bands which do not have the resources to buy their way onto playlists.
There is an exception; a record label can straight out pay to get a song played, but the radio station has to disclaim that it is a pay-for-play, and the amount of airtime devoted to pay-for-play is limited by law (I believe it may be by considering such to be advertising; and radio stations are limited in the fraction of airtime which is advertising). This sort of payment is probably unproblematic from a legal and a moral standpoint, unless playlists are influenced by who is buying advertising (which would essentially be old-skool payola again).
Hello - have you used Office 2007?
Well, it's only been out 4 months. I based everything I said off of previous incarnations of Office, from 2003 (the second latest-n-greatest) to 95 (when WordPerfect started sucking). Given I don't want to replace my shiny & new copy of Office 2003 so soon after I got it, I won't be purchasing 2007 for some time. And nor will I be using to submit to Nature or Science, apparently, even if it were free.
1. PDF publishing is supported (free)
That's good to hear, and long long missing. Questions: is the output of "Print to PDF, then print the PDF" the same as just printing? I've had issues with such software producing different output that way (even sometimes products made by Adobe!). Also, how "good" is the PDF output. That is, are the files sizes quite small, is it embedding proper scaleable fonts, and does it print fast? A big problem with the old "print to .PRN, change extension to .PS, then ps2pdf" way of going from Word to PDF was getting bloated, poor quality, and complex PDF files; even Acrobat sometimes will non-sensibly make a crappy PDF.
2. Citations and Bibliographies are both supported under Word
Having used BibTeX, I will never go back. There are huge databases of freely available BibTeX format citations. The second runner up, EndNote, hasn't nearly the amount of citations available (although importing BibTeX into EndNote isn't hard). I have used EndNote, and it is not nearly as good as BibTeX. Other people (read the comments) seem to feel that EndNote works better than Word 2007's support; I simply don't believe that a mouse-driven interface for adding citations can ever beat a text-based one.
Office 2007 documents can be saved to document managment servers for sharing
I think you're missing the point. LaTeX easily allows you (and encourages you) to split your documents up into multiple files. So, regardless of what collaboration service you use, anything from emailing files back and forth to something overly complicated like SourceVault, you can have multiple people editing the same document simultaneously. They just work on different sub-files. I've done it with email, but usually do use RCS to automate the locking support. Unless Office 2007 vastly changes things, Word documents are still monolithic files. That makes it quite difficult to support simultaneous editing; you *need* a concurrent versioning system. And, forgoing large changes in Office 2007, .doc files are still stored as BLOBS, which makes automated commit/merging difficult to impossible.
4. LaTeX has style files; Word has templates. What's the difference?
As best I can tell (I've never found it, and I've tried...) you can't apply a template to a document after the fact (and get the expected results). If my paper is rejected from one place, I can reformat the entire document by changing which style file I include. Style files pretty much guarantee that all of the final product will have a consistent look, and that said look is easy to change across the board. Journals that are either done all in LaTeX or those that hire separate typesetters (for mucho $$$) have the most consistent appearance. Others look a bit like just a bunch of papers glued together.
I can't claim Office 2007 is better than LaTeX, since I've not used the latter extensively,
Except for those on the Office 2007 team, *nobody* has used Office 2007 extensively :)
From talking with people who have used 2007, there is quite a learning jump to go from 2003 to 2007. Especially since 2007 breaks math support for journals, it makes sense to consider moving to LaTeX just a
Really, Office isn't nearly as good as people make it out to be. And LaTeX isn't nearly as hard; especially if you use one of the WYSIWYG editors.
As other posters have pointed out, the editing that goes into a digital file is worth something; but the print itself is worth very little. People know that they can get a digital print from Walmart* for under a quarter; four bucks from your wedding photographer, which is sixteen times as much, is clearly obscene. If I want a lot of prints, your traditional wedding photographer is clearly crap crap crap for value.
Now, even after charging for your time massaging the files, the prices you charge should be lower than they were 20 years ago (in inflation-adjusted terms). Think; 20 years ago you *had* to be shooting film. Your cost of producing prints included the film, the development, carrying many cameras with different speed film, etc. If you wanted to be sure to catch those perfect moments in pictures, you had to blow a *lot* of film, and a lot of processing. Nowadays, it's just the incremental cost of a larger memory card; that's *much* cheaper than film. The massaging that you can apply to a digital file used to be done in the darkroom, in a difficult labor intensive process. Nowadays, Photoshop makes it a lot easier, faster, etc. *Your* costs of producing good pictures have gone down; it makes sense that your prices should drop too.
I've noticed that the older, more established photographers still do silly things like charge extra for B&W & sepia-toned prints; they also refuse to release digital files. Now, the really silly of them are still shooting with film (and frankly, deserve to be shot**). They are a dying breed. But for those charging large differentials for B&W when it takes all of 30 seconds to toggle it in Photoshop, please, tell me how that makes any sense at all. Really, they're just still clinging to a dying business model.
Really, do yourself a favor and switch to charing for the time you actually spend. Charge for the labor of generating an album, but charge what it actually costs you to get it bound for the album itself. Charge for touching up photos, but charge what it costs you to get prints for the prints. By charging what reflects your actual costs, you'll be able to attract more business. If you are artistically good (either in the shooting or the editing or both), you can charge a premium for those services, and people will make lots of prints of your beautiful pictures (at no cost to you). Your customers will be happy, they'll tell their friends (if they can get cheap prints of your pictures, *they* will pay to give them to their friends--free advertising), you can charge ever more for your time... it's a virtuous cycle. Or, you can continue charging obscene amounts per print, and people will feel that you're ripping them off. Other photographers will offer what people feel to be more fair price structures, and you'll start loosing business.
Remember, in the end you can't force people to accept your terms. If you *seem* unfair, they'll balk, and then they'll walk.
*You make fun of Walgreens, but the quality of their prints has been steadily increasing. Reputations for bad quality are a legacy of poor film processing; their digital stuff is not bad. And Walmart, disreputable as they may be in many areas, actually produces really nice prints, with consistent color, clarity, contrast, etc. Their film processing leaves a bit to be desired (dammit, stop pushing my low-light shots!), but their digital is great. Kodak does good digital prints online for cheap too. If the raw files are put through even basic color correction, the prints will look good almost no matter who does them (Snapfish excluded because they suck).
**I still use film myself, but not professionally. I do it because it's fun; anyone shooting film professionally is either an idiot or has special needs (e.g. IR, or ultra high/low ISO).
This does at least insulate us from the rising cost of oil for the time being, since oil is traded in US$, and US$ is falling almost as fast as the price of oil is rising.
Well, consider what percentage of a liter of petroleum costs in either EUR or GBP is caused by flat taxes; the end price won't change much at all compared to the price of the underlying commodity.
Further, that the dollar is falling and that the price of oil (as denominated in USD) is rising is really the same thing. For a fairly fungible commodity like oil, rather than thinking of the actual number, it makes more sense to think in terms of how much it will buy. Since oil is priced in USD, and the real purchasing power of the USD is falling, it makes sense for the price to increase. Notice that since the US CPI is still quite low (under 2%); in terms of real goods, a barrel of oil will get you more stuff. The price really is rising.
Another way to look at it is in terms of Germany. Germany is 1) terrified of inflation, and 2) dominates the EU economically. The Euro, therefore, is tilted towards deflation if anything else (this is in general an argument against using the same currency in a such a large and diverse area). So, the purchasing power of the Euro should likely be increasing slightly (or, in terms of exchange rates, it should be rising). But the price of oil in terms of Euros is rather flat; so in terms of other goods, the price of oil is still increasing. That is, what a barrel of oil can actually buy you is increasing.
Or we can look at the Yen; Japan has been experiencing near zero inflation to deflation for quite a while; yet the price of oil in terms of Yen is still increasing. Although not nearly as dramatic as the increases compared to USD, one can still make an argument for the actual value of a barrel of oil increasing.
If the price a commodity is nearly flat in terms of a currency, then what is to blame is that the currency is gaining in strength; not that any specific currency is falling. You are still paying more for that commodity in real terms (that is, other goods); but that the numeric value is flat is good at fooling you into thinking that it's "price" is constant.
The sad thing is they wouldn't have even needed to buy the machines. You can "buy" them, use them for a short while, then return them, minus some small depreciation amount. Big retailers (e.g. Macy's) do this for the Christmas shopping season--and the big machine vendors (e.g. IBM, HP) are happy to oblige, since if the machines are used for a sufficiently short amount of time, they still count as "new", so they get to sell them again, full price, to someone else.
Or, another cheap and easy way is software-disabled hardware. Buy a blade system and have a large percentage of disabled (but present) blades; pay for them and turn them on only if you need them. Or you can do the same thing with most mainframe systems. Nearly every mainframe system ships with additional processors installed but disabled; IBM even has this thing where they skip every other clock cycle until you pay for the upgrade. Intuit was really quite short sighted about this. Obviously there is going to be a huge load spike right before the deadline. They could have easily allotted themselves a large over provision amount for not that much money.
It isn't surprising that VMWare would be bad at a web-app workload. See the original paper on Xen:
/ 2003-xensosp.pdf
http://www.cl.cam.ac.uk/research/srg/netos/papers
Top of page 9 has a chart comparing native Linux, Xen, VMWare, and UML for different workloads. They show VMWare degrading performance by over 70% for SPECWEB 99.
Web applications are OS intensive; while VMWare is quite good at pure CPU-bound tasks, it has to perform a lot of emulation whenever you are running inside the OS. So it will stink at anything with lots of small IO, lots of metadata operations, or lots of process creation/switching. For example, VMWare shows a whopping 90% slowdown for OLTP database workloads, according to the Xen paper, and it really isn't surprising. The OS microbenchmarks in the above paper (page 10) show that VMWare has abysmal performance for things like fork(), exec(), mmap(), page faults, and context switches.
Basically, Xen doesn't have to emulate the OS, because they make modifications to the OS. VMWare does dynamic binary rewriting (think fancy emulation) to run an unmodified OS; they therefore pay through the nose in performance overhead for OS-intensive workloads.
Skype (which, coincidentally, was written by the same people who wrote Kazaa) uses some of those workarounds to punch through NAT firewalls. I do not know if Kazaa uses them, but the authors of Kazaa could have certainly done so.
But P2P works better if it has access to your public IP address, and you can accept inbound connections. Hence some P2P applications will complain if they detect that they are NATed and ask for your public IP. Some will auto-detect you public IP. Others will not only autodetect your public IP, but if you have a UPNP-capable router, will automatically detect or otherwise set up appropriate holes in you NAT firewall (later versions of Azeurus do this, I believe) to forward inbound connections. And, as aforementioned, Skype uses NAT-busting techniques to bypass setting up proper forwarding rules altogether. Skype's ability to get past firewalls is actually somewhat frightening...
Although the original design of the internet was based on the assumption of a static one-to-one mapping of computers to IP addresses, this is not the case today. DHCP means that the mappings are not static, and NAT means that the mapping isn't one-to-one (indeed, a sufficiently sophisticated NAT setup could be many-to-many, although such would be unusual). Even MAC addresses aren't really unique--it is quite common to set up interface failover by spoofing the MAC address of the failed NIC. Identifying a computer uniquely is a very tricky process--the common means of doing so rely on these broken assumptions. The uncommon means (specifically, searching for evidence of clock drift in timing parameters) are, well, not commonly used, and have higher false positives (due to sensitivities to temperature and the low precision of clock drift measurements). And none of this can be used to show that a particular person was doing anything at any point in time.
From my limited experience with expert testimony, many expert witnesses, although experts in their field, are not experts at being witnesses. It's a way for a university professor to pick up more money on the side with easy consulting work, especially if hired by a petitioner under the expectation of a weak defense by the respondent. In such a case, speed and cheapness are prized above thoroughness and accuracy, and actually being deposed by a lawyer who has been prepped on the sorts of questions to ask would be quite the surprise. Dr. Jacobson appears to have been caught with his pants down, giving a slap-dash report which is clearly biased in favor of the side which hired him. Although he isn't a member of any regulatory body, I would be surprised if he wasn't a member of the ACM or the IEEE Computer Society, and in violation of their respective codes of ethics (specifically, ACM 1.2, 1.3, and 2.5, and IEEE 2, 3, 7, and 9).
At least on /., the Google study is getting all of the press. But there were two large studies on disk failures this year at FAST: the one by Google, and the other by the Carnegie Mellon Parallel Data Lab. It won best paper. You can find it at http://www.pdl.cmu.edu/PDL-FTP/Failure/failure-fas t07_abs.html
Although the CMU group didn't have environmental data to see what correlations there were, they get results similar to Google in other areas; e.g. that drive "vintage" matters, the failures are not a Poisson distribution, etc. Since the two studies agree, it implies that both are reliable. Also, most people *don't* have a heat chamber to weed out flaky firmware behavior--the drives that almost everybody runs are drives that may have such firmware games enabled. The studies are on disk drives *as they are shipped from the manufacturer*; the correlations found, whatever their reasons, do exist.
Yay! I get to recycle an old comment, just like drug manufacturers recycle old compounds! Unlike the drug manufactures, I include extra content, and all at no extra charge.
Some other examples of "bad" patents include:
1) Obvious compounding. A good example is pain medication. Acetaminophen (Tylenol) has an unusual method of action which is synergistic with nearly every other analgesic, and rarely interacts with other drugs. So, the drug company will file a patent on their new painkiller, and then (just before the patent is made public/the drug is approved), they'll patent mixing it with acetaminophen. Doctors prefer prescribing the mixture because it has a perceived lower risk of abuse (due to the liver toxicity of acetaminophen), so the generic unmixed version isn't used so much. A specific example is Ultram vs. Ultracet (although both are now available generic, there was a period when Ultracet was not).
Going back to Claratin, Claratin-D is just Claratin plus pseudo-ephedrine (Sudafed). Yes, lets compound an antihistamine with a decongestant ... never would have thought of that one.
2) Racemic mixtures. Many drugs have left handed and right handed versions. Often, one version or the other is more effective/safer. Especially since the thalidomide incident (anti-nausia drug where one version (left?) caused birth defects) testing both versions is standard. Yet the drug companies can get separate patents on the left, right, and mixture versions. Sometimes, the patent on the left or right can be used to control the mixture, especially if it is difficult to make just one version or the other. Regardless, it gives the company a "new" drug to market and to compete with the generics. Prilosec and Nexium are an example of this; Prilosec is mixed and Nexium is not. There is little evidence that Nexium is any different from a 2x dose of Prilosec (which have the same amount of the active handedness).
3) Particle size patents. Hmm, it just so happens that a certian size granule is "better" than others, and the standard manufacturing technique (whose patent is expiring) makes that particle size (or at least contains it)... Provigil is currently being protected under a particle size patent, and will for 5 more years, even though the formulation patent expired last March.
4) Time release/enteric versions. Coating something (with a standard, commonly used coating) to make it time released or gentle on the stomach isn't obvious, for some silly reason. Sorry, no example of this one.
5) Dosage. Yes, taking your pill in a different dose can afford protection against generics, although I believe this is at an FDA regulatory level rather than as a patent (finally! We found an obvious patent). Flexeril is a muscle relaxant, and the usual dose is 10mg. If you cut that to 5mg, you get a much lower incidence of side effects, and it is just about as effective. For a while you couldn't get a 5mg generic, although it seems that you can now. Frankly, rather than spend $52 for 30 brand name 5mg pills, I'd rather pay $8 for 30 generic 10mg pills and $5 for a pill cutter. That is $.21 per dose for the generic and $1.73 for the brand name.
And what is so bad about Vioxx? If you are allergic to sulfa, need to avoid gastrointestinal side-effects, and are at a low risk for heart disease anyway, Vioxx is an excellent choice of drug (Celebrex has a sulfa group, NSAIDs and aspirin cause stomach bleeding). Especially if you are going to be taking it for a medium term (12 months or less) there is little risk. The studies that show statistically significant increases in adverse cardiovascular events show them for long-term high-dose use; many of them are only borderline significant. Both the US FDA and Health Canada have recommended it be returned to market, and it appears that aside from the lower bleeding risk, Vioxx is no riskier than ibuprofen.
There are lots of drugs which are far more dangerous than Vioxx (say, anything with lithium in it), yet continue to be sold because they serve a therapeutic niche. I am quite disappointed in Merck for two reasons: first because they took an unreasonable and unwise approach to the evidence of cardiovascular risk in Vioxx, and second because now they refuse to manufacture Vioxx and sell it with appropriate warnings.
When it hits the sandwich.
Don't forget all of the other stuff you get from spalling, like high energy X-rays. Actually, at 300MeV, I'd wager on getting a fairly decent gamma ray beam. Without a purpose built collimator, I'd guess that there'd be a good amount of "spray" all over the place. So you'd probably get a hole in a radioactive sandwich, plus a good dose of radiation just for standing nearby. Yes, a healthy dose life-giving radiation.
Remember, the Therac-25 system was quite lethal when it malfunctioned, and it "only" used a 25 MeV beam. 300MeV is a LOT of punch per particle, and if the intensity is high enough all sorts of nasty things will happen.
They sell *most* of the parts from old Thinkpads. They discontinued selling replacement batteries for T4x series machines; rather irritating because the T43 model was only introduced early last year, and now you can't get batteries for them straight from Lenovo.
On the other hand, you can get most of the parts, and the batteries are still in supply in the retail chain; I just ordered one from Newegg. IBM was absolutely stellar in terms of support; Lenovo seems to be not quite as good. Much much better than the other big vendors (like, >shudder, Dell), but not quite the same as big blue was.
If "old value" starts out between 0 and 1 and "new value" is always either 0 or 1 (plus or minus minor noise), things will work out. If I write a 1, then you get (.9 * 1 + .1 * something_between_0_and_1) = .9 + something_small = something_between_1_and_.9, while writing a 0 will get you something inbetween 0 and .1. Even if, somehow (say, lots of unluckily correlated noise), the "old value" is corrupted to something out of range, (e.g. 1.1 or -.1), the average will pull it very quickly back into range. It isn't a linear average; the most recent result matters most, and dominates enough that you'll have something distinguishable one way or the other. If, somehow, the data was written to be .5, then you'd get (1 * .9 + .5 * .1) = .95 for writing 1, and (0 * .9 + .5 * .1) = .05 for writing 0, which is actually closer to a perfect 1 or 0 than if the previous write was the opposite.
Although, do note that rather than using actual 1's and 0's, we have analog field polarities which we let stand for 1's and 0's... the idea is the same, though.
Well, that you can recover data doesn't mean you can reliably recover it, or do it fast. Hence no 40TB drives.
.1; the old value leaks in a bit. When combined with the error correction codes built in to the drive, you stand a decent chance of recovering the overwritten data (slowly, with special read heads/drive electronics, and somewhat error prone). If you overwrite with the same fixed pattern repeatedly, you really don't improve the situation much (diminishing returns on removing the residual bit), so an 80x "wipe" of all zeros may be recoverable (although 80x is a bit much, even for all zeros). But overwriting to DoD spec is probably sufficient in this case, though. So is beating up the physical platters.
.1 thing works better. But it is far too expensive to bother with in most cases.
d ings/sec96/full_papers/gutmann/
As for how, the short version is that if you write a one and then a zero, you end up with
For the really determined attacker, then can use high resolution magnetic force microscopy (MFM). See, when you overwrite, the write head doesn't exactly line up with the old stuff--so you'll have little bits of the old data sticking out from above or below the track. MFM can resolve very localized magnetic fields, far smaller than your disk read head, and can see the misalignment. These misaligned pieces allow peaking back a few generations of overwrite, which allows you to subtract out the newer things (and hence clarify the older). Plus, it doesn't use rapidly spinning the disk, so it can work on beat up platters. Even without misalignment in the generations of overwrite, good MFM can resolve better analog detail as well, so the whole 1 then 0 =
A detailed technical paper about the theory of data destruction/recovery on magnetic media can be found here: http://www.usenix.org/publications/library/procee
If the drive still works, Eraser (http://www.heidi.ie/eraser/) is a great choice. Wipe one file or every block device on the machine. Nothing will be able to get it back after a Gutmann wipe.
If the drive doesn't work, disassemble the drive, remove the magnets (to play with later), and apply a combination of sandpaper and a hammer to the platters. Yeah, magnetic force microscopy might still get it back, but who'd pay that much for your mostly worthless data anyway?
If you're sitting down to read through Knuth all at once, well, gads that is painful to try. They really aren't books to learn the concepts from, but rather serve as a reference for a wide variety of basic topics. Although there are better books on all of the individual topics, Knuth covers a lot in just three volumes. True, volume 1 is of limited practical use, but I've often referenced 2 and 3, for when I needed custom random number generation, custom hash tables, had to deal with unusual sorts, and as a general data structure reference. I feel confident that I'll use it for other things in the future. Knuth probably is of limited use to somebody who finds that they can always use the standard libraries, but if the STL Map classes are just too slow for your particular application, or if you need to sort a 10 GB set of data (whups, guess the system quicksort() isn't going to work... now what's the best way to do an on-disk sort?), Knuth is invaluable. As for being too dense, well, all of the big-O analysis and whatnot explain to you why the various algorithms and data structures work the way they do. If you don't understand how things work and why they behave the way they do, you are likely to make poor choices.
Second on the Knuth and K&R. Especially the Knuth -- the level of detail is beyond belief. If Knuth has anything about it, he has everything you ever wanted to know about it; like the 15 jillion different ways to do a hash table. K&R unfortunately is covering a fairly simple topic, so it doesn't get used as much (actually, use it for lending more than anything).
Besides Knuth, the book I use the most is "Computer Architecture, A Quantitative Approach", by Hennessy & Patterson, although it isn't adorning my shelf so much as much as my desk. If you care about the architecture side of things at all, Hennessy & Patterson is "the" book; plus the later editions cover relatively modern processors, like P4 Netburst. Not to be confused with Patterson & Hennessy, which is a more introductory book, and not nearly as well written.
A few other goodies are Numerical Recipes in C (for the rare cases when Knuth is lacking), and a lot of books by W. Richard Stevens, William Stallings, and Andrew Tanenbaum.
then one can extend it to a variety of areas which might be more useful than movie recomendations
I'd say, the odds are that this is going the other way. They had an existing technique, and then they extended it to movie recomendations. You don't need to offer researchers in data mining a price to get them to advance the state of the art in data mining; that's what they're interested in, and what they're payed for anyway. The prize just got them to apply it to movie recomendations.
The only thing to see here is that Netflix hadn't kept up to date on the latest and greatest in the academic literature. Otherwise, just move along.
It is the other way around. Let me tell a story...
If you die, and somebody is responsible (say, negligence, or recklessness), then your next of kin are entitled to compensation for your death. There's the whole "pain & suffering" bit, which is quite abstract, but there are also economic losses. That is, you, being dead, are no longer being paid, and that is money out of their pockets. Figuring this out is dreary economics and statistics, but it is quite measurable, and well presented economic damages are really hard to argue with, as opposed to those fuzzy pain & suffering numbers.
Of course, there are lots of complicating factors. You could have been fired next year. You aren't doing any housework either (and so someone must either be working harder, or hiring somebody; both are losses). Of course, dead people don't eat, don't wear clothes, need less housing, and don't drink or smoke; this money not spent is subtracted from the damages. Now, actually measuring how much the deceased ate, or that they were in solid with the boss, etc. are difficult things. So, instead, there are tables to look these things up in. Hundreds of tables, computed from census information, surveys, etc. Tables on everything. What the odds are, given age and occupation, of entering or leaving the workforce. What your raise would have been. And how much you eat, drink, and smoke.
The tables are split by family size and income. Richer people, for example, spend more on food; they eat out more, and buy nicer food. They of course spend more on travel, clothes, and housing. Most of these expenses start to level off after a bit, though. Interestingly, smoking is a decreasing expense; the poorest bracket spends less on smoking than the second poorest bracket (guess the poorest really just can't afford it), but spending decreases from there.
The expense that increases most linearly with income is money spent on alcohol. People with twice as much income sure don't spend twice as much on food, but they spend twice as much on booze. Cheers!