Microsoft delivers an operating system to users (and you pay again for development software).
Linux delivers an operating system and development software to every user.
Which approach is going to encourage more developers in the long run?
As the market stands at the moment, Linux is certainly attractive to would-be developers. And as computer use spreads into countries where it is not yet widespread among the general populace (China springs to mind), Linux is going to have further advantages: It requires less powerful hardware, and it is open to home grown localisation.
I wonder what effect of 1,000,000 Chinese hackers will have on Linux? Of course a lot of work will go into some really good Chinese language support, but I expect to see some stuff come out that we can all use.
Unless Microsoft can buy off China. But I think China is old enough skeptical enough enough to see through that.
Of course, while developers are pretty important, developers without users are of questionable value.
Oops! You are absolutely right. Sorry to everyone I criticised.
I guess if they labelled it a 'fully GPL compatible' distribution, then there would have been no controversy, because (I assume) most of us accept RMS' analysis that while the QPL is DFSG-free, it is not GPL compatible.
Since the name is an obvious homage to RMS, the intent to be 'GPL compatible' might be inferred. But by using the word 'free' instead RH perhaps treated KDE a little harshly.
Several people jump in claiming that to call the QPL non-free is disingeuous, stupid, or just plain wrong.
No-one jumps in to claim that the NPL is a free license.
And yet (assuming QT2) the two are very similar: both provide the code originator with a privileged position with respect to the code base.
I am perfectly happy to use Netscape and/or KDE (except that Netscape is bloated beyond the dreams of MS), but from an ideological position it is clear that a free distribution (or at least a GPL||BSD||Artistic distribution) should exclude both.
With respect to the gift economy, I absolutely agree. The gift economy is largely an anarcho-romantic notion popularised by Kim Stanley Robinson. The scientific model (peer-review, building on the work of others) is and always has been a more accurate model.
One perceived difference is that many OSS developers do it for the love of coding. This misses the point that many scientists work for exactly the same reason. I could double my salary if I left science, and have spent more than one year working with no income at all, living off savings, just to stay in the field.
With respect to Microsoft, again the author is spot on. I am reminded of the end first Batman movie, in which the Joker tells Batman 'You made me', to which Batman replies 'You made me first'. Microsoft finds itself responding to a Linux threat, but it may well be Microsoft's contempt for its customers which has put Linux where it is.
The criticism of Linux' security may well be fair, but at the same time can be compared against the record of NT, Office and Internet explorer, which are hardly better. Having said that, I don't think there is any doubt that some of the commercial unices are far more secure than either Linux or NT. Security hasn't been a 'sexy issue' until recently.
On project management and development dictatorships: It does seem to make a difference when the dictator is an individual (Linus), rather than an organisation (TrollTech, Sun), so the 'cult of personality' call is fair. Having said that, I still think Cathedral is an important and interesting piece of work. Most importantly, it came first. Later works have the benefit of referring to it and of consulting a much wider range of OSS projects. Even though some of ESR's ideas may be wide of the mark, his conribution in starting the discussion and laying a framework of ideas must not be undervalued.
The aim of the keyboard piece is clearly to disprove the notion that an inferior solution can remain standard through monopoly inertia. Whether this is true in the keyboard case I don't know, but if you go up to the rest of their writing you see their purpose: to demonstrate that Microsoft's cominant position is not related to inertia and most likely arises through technical superiority.
If you go up to their page about the MS anti-trust case, they put forward some evidence that prices of software products in markets where Microsoft compete have dropped much faster than prices in markets where Microsoft does not compete. At first glance this suggests that Microsoft is not a monopoly, since monopolies usually exert their influence to keep prices inflated.
However, the reasoning is fallacious:
The unit production cost of software is almost zero, so the economies of scale are huge. Software which sells lots of copies should sell at a tiny fraction of the price of specialist software.
Microsoft exercises its monopoly position in one main market: the OS. Its income from this market is so huge that it can afford to loss-lead products in other markets. Thus MS may provide a downward pressure on products in some markets, but only by inflating prices in another market.
The trend for MS monpoly product, its OS, is upward, not downward, despite the increase in the market.
Becuase of the huge economies of scale, it may in the past have proven beneficial in terms of price to have a single provider rather than paying many copies to duplicate effort. But the cost of lack of competition is lack of innovation.
The market is now so huge that software prices are essentially being driven to zero for the most used software. Microsoft's pricing, with the exception of internet explorer, does not reflect this trend.
The only software markets in which Microsoft does not compete are either specialised, or fast turnover (games). In these markets the huge economies of scale are not realised, and so the pricing is not expected to fall in the same way.
There are three mainstream 'Linux for Windows' products in the news at the moment: Armed, Phat, and WinLinux2000. And several other distros that contain the same technology: Mandrake 6.1, Slackware, muLinux, and so on. And DemoLinux.
Technologically, Phat seem to be most mature and advanced, featuring the option of installing to a loopback file as well as UMSDOS, giving potentially better performance. I can't compare the hardware detection in the various products. I presume Armed doesn't do loopback, but their web server won't serve me any pages today, so I can't tell.
But technology will not determine success. My take on the ones which will succeed are as follows:
WinLinux2000 will win. It has the most accessible and descriptive name. It will appeal to Windows users. PhatLinux is an obscure hackerish joke on the name of the file system. Armed Linux sounds like a secure distro, not a newbie intro.
Mandrake 6.1 may grab the market if they chose to push this feature, on the basis of their reputation as a conventional distribution. Corel could do even better if they wanted to do UMS or loopback installs, because they have the Windows brand-name recognition.
DemoLinux will do well, and probably make lots of magazine coverdisks. Again, most importantly, the name says what it is.
Armed are going to take stick for their choice of web hosts, and the fact that their pages are unavailable whenever their product is in the news. Which seems perfectly fair to me.
As a community out technical efforts are occasionally sabotaged by lack of attention when it comes to image.
In the UK we refer to English units as Imperial units. While they have the same names as the US versions, the sizes are slightly different. (e.g. our pint is 20 fl oz, not 16, our ton is 2240 lb).
Imperial measurements were abandoned by schools during the 70's when I was a child. However, they are only now being phased out in retail: it was only a couple of years back when meat started to be sold by the kg rather than the lb.
People still talk about a pint of milk, although it is almost always a 1/2 litre. There are some bizarre mixtures: fabric comes in 45 or 60 inch widths, but is now often priced by the meter lengthways.
Most nuts and bolts, and also spanners, are now metric, although a full toolkit still comes with both - a bit of a waste I guess.
In my scientific work I use the electron as a unit of charge and the angstrom as a unit of distance.
The unix `units' program is always handy if you need to convert forces from hundredweight furlongs per fornight squared into something useful.
I first saw this story in New Scientist a year ago. I think it was reported in Slashdot at the time.
A month later New Scientist published this story, suggesting that the slowing was due to the reaction from heat radiated from the probes RTG power plant.
They still appear to be arguing over whether this effect is big enough. Measurements involving heat are notoriously difficult, as the cold fusion debacle showed.
The free software movement seems to span many political and religious viewpoints, and you must have met more of the movement than most people. Do you have and feeling for what worldviews are more common in the free software movement? Is it every difficult working on a shared cause with people with very different motivations?
Eric, in your papers you've put forward many political, sociological and technical reasons why open source software is a good thing. (For example the gift culture is a political model, peer aclaim is a motivation for some programmers, peer review leads to less buggy software).
Every individual will be differently influenced by these different arguments, depending on their political leanings, emotional makeup, and the problems they are trying to solve. Which justification is the one which is most persuasive to you personally?
Everyone seems to have their own history of Galileo!
If I remember correctly for most of his life Galileo was kept by a rich sponser who guided him in 'politic' expressions of his theories. During this period his main trouble with the church was his claim that the moon when seen through a telesope had features on it, wheras the church insisted the heavenly bodies must be perfect. (The moon's visible features were put down to imperfections in the eye).
It was later on when he was without a more worldly advocate and wrote a cutting satirical dialogue protraying the Pope as an ass that the church started persuing him for Copernican notions. My impression of this was that the church was motivated more by revenge than dogma.
Ironically several of the top astronomers of the period, who went on to verify many of his results, were monks at the Jesuit university in Rome.
But it's a long time since I read any of this, so I may have misremembered.
No, I think you misunderstood my comment. The experimental technique they are using is absolutely sound, and the mathematical tools are sound, although they can easily be abused if not properly understood.
My concern was the appalling reporting, which contained several statements which were factually wrong (and clearly physically impossible), and failed to consider the reliability of the results at all. The real facts were obviously read and then re-written by someone who didn't understand what they were reading.
When I find Sci-Am reports wrong facts in a field I do know, it shakes my faith in what they write about other fields where I am reliant on their layman-explanations.
The work is already published in Nature, where I don't doubt there is an accurate account of the techniques. But I'll wait for the top statisticians in the field to comment, because it is really, really easy to bias this sort of result unintentionally by an inadvertant assumption: There have been several wrong results of this type published before.
The image to which you provide a link appears to have a resolution of around 1.5-2.5 Angstrom (which is reasonable, since that is the size of an atomic tip). Charge density studies and the imaging of atomic orbitals require resolutions of better than 0.7A, I suspect better than 0.5A for the results claimed in Scientific American.
(That is using resolution as defined in crystallographic terminology, it might be used differently in STM).
X-rays (in crystallography experiments) are not scattered by nuclei, they are scattered by the electron density in the crystal, just like the electron beams. If you want to image nuclei, you have to go to neutron scattering.
The X-ray and electron beams are not combined, they are collected separately, and the information is combined.
Moreover, the principle difficulty has been completely ignored. Electron beams can be focused to produce a direct image. X-rays cannot be focused for imaging purposes (although crude focussing to concentrate a beam is just possible). As a result, you only get a diffraction pattern with no phase information. The image must be reconstructed by Fourier transformation, which needs the phases. (There is a strong analogy with optical holography, in which a reference beam must be interfered with the diffracted beam to obtain phase information, but with x-rays the coherence length is too short to get a reference beam).
The trick is to use the phases from the low resolution electron image, and some mathematical relationships to reconstruct the missing phases in the high resolution image, which will show your electron orbitals. The problem is unless the statistics are treated very carefully, all you get is an image which confirms the assumtions of the model you used to get the relationships with which you reconstruct the phases.
The mathematical techniques were just coming on line in X-ray crystallography in 1996 and there was still considerable debate back then over their correct application. So there is a fair possibility that these results are correct, but I would suspend judgement until they have been scrutinised for a year or two.
I have been using Gnome on my home machine: a K6-2/333 with 32Mb of ram, and was also suffering perfomance woes in comparison to KDE on the same system. I think the main issue is memory.
Switching to the default theme helped. I also dumped E and switched to wmx (which is only 82K compiled with Gnome support, no pixmaps, stripped).
The other big gain was setting the default help browser from Netscape to the gnome help browser. (For some reason the control center wouldn't let me do this at first, I eventually editted a text file. It works now). The help browser has some limitations, most noticably no new window on button2, but is OK in general. It would be interesting to try Gnome using Kfm for both help browsing and for desktop icons, since gmc is quite heavy.
Having said that, I never tried themes under KDE, so I don't know if it has the same difficulties. I recently added a cheap 16Mb dimm: having 48Mb makes all the difference.
The way the market has been going recently, it was beginning to look as though US vector supercomputers were dead, and only the Japanese were still advancing them. The T90 was the last major vector machine, but had memory synchronisation problems with more CPUs. The SV1 had some interesting specs, but I don't know of any site which actually installed one - certainly press releases were thin on the ground.
Parallel machines, such as the Cray T3E, IBM SP2+, and to a lesser extent Beowulf clusters just give so many more Gflops/$. But as has been pointed out they are completely unsuited for some problems, for which you simply need all you power concentrated in a small number of CPU's.
My guess is that there is not enough market for a new US vector supercomputer, and the US government are stepping in so not to become dependent on imported hardware. If most SV1 installations are government, it might explain why we've heard so little about them.
BTW, many older supercomputers were single-user machines, which required a front end running a mutli-user operating system to schedule jobs. However all recent machines, including the C90, T90, T3E and I suspect the SV1 and SV2 run their own operating system (they are self hosting). In this case it is UNICOS, a Cray Unix which is gradually being merged with SGI IRIX.
My experience of UMSDOS linuces...
on
WinLinux 2000
·
· Score: 1
I bought an obsolete laptop, and the first thing I tried was putting a tiny Linux, muLinux, on it. muLinux will run from floppy (4 of them if you want X, gcc, and network tools), but can also clone itself to create a UMSDOS installation or a loopback installation (this latter is a Linux filesystem inside a big file on the DOS partition).
The UMSDOS clone worked OK, but the problem is that Linux needs thousands of tiny little files. These are stored rather inefficiently under UMSDOS, and become a nightmare if you need to defrag (for example to partition for a full Linux distro). The performance hit is also noticable. (Notice that WinLinux2000 occupies 500Mb on a FAT32 partition or 1Gb on a FAT16 partition - that is the impact of all those little files).
My guess is that the loopback filesystem, as used by Mandrake's Lnx4win, is a better choice. Hopefully the WinLinux people will add this option in future. Otherwise, a great piece of work - I love the idea of stealing all the settings from Windows.
Also worth looking at is DemoLinux, which runs off the CD.
My laptop now runs Slackware 3.4 in its own (half empty) 120Mb partition, with X, compilers, networking and freeciv. I might try Debian 2.1 if I need libc6.
For improvisation with rules or preparation, you might try a story-telling game.
Once-Upon-a-Time involes making up a fairy-tale including all the elements on cards in you hand, and ending with the line on your ending card. Other players may interrupt if you mention anything on one of their cards, or by using a generic interrupt when you play a card, and then have to continue the tale.
Baron Munchausen is even more free form, with little tokens as the only props. Players take it inturns to tell tall tales of their adventures. Other players may interrupt with challenges to the story (by paying a token), in which case the storyteller must either correct themselves or pay to refute the challenge. At the end, players spend their remaining tokens voting for the best story.
The point-totting games you have described are certainly a significant sub-classification, but I don't think you can legitimately claim that German games fall more often in this genre.
For example, Catan, Manhattan, 1630-something, Civilisation and both Illuminati games are point totting games, but only the first two are German, the third being a rare good UK game, the rest being American of course.
Mississippi Queen, Ricochet Robot, Elfenland, Bausak (where you build a tower of awkward wooden blocks), Carabande (a big wooden motor racing game where you flick the counters) are all German (some republished elsewhere), and don't involve points.
I would agree that point totting games can get frustrating if you play with people who are too deparate to win. In Manhattan in particular the last player each turn can often spend 1/2 hour figuring a play which will turn the whole game upside down (but I've seen Monopoly played that way too). The solution is to play games for fun. Beer helps.
Program your robots a turn in advance to navigate around a maze containg hazards, conveyors, turntables etc. Confused by the fact that you may not have the right program cards, another robot may bump into you throwing your calculations off, and the robots shoot at each other. Long.
Ricochet Robot (Hans Im Gluck/Rio Grande)
Much simpler, and yet far harder. Move robots with no brakes around a board to reach a target. You have to hit things to stop. Usually there is nothing in the right place to bounce off of. So you have to move several robots. Sometimes you have to work out 20 or 30 moves in your head, and then announce before anyone else gets there. (Best call I know of was 63, which involved iteratively bouncing two robots off of eachother).
Die Siedler von Catan/Settlers of Catan (Kosmos)
Probably the best board game ever. A sort of colonisation/town building game, with a random board made up of hex tiles. Superbly balenced, and reasonably quick.
Mississippi Queen (Gold Sieber/Rio Grande)
Actually, this one is easy enough for non-geeks, but has some of the same sort of puzzles as RoboRally - work out how fast you can go without ramming an island in a randomly twisting river.
Euphrat und Tigris (Hans im Gluck) is good but I can't work out how to win. Sixteen-thirty-something (Warfrog) is a very strange twist on the normal board wargame idea.
Needless to say, the best boardgames come from Germany, although there are some good US companies too. Rules translations are sometimes needed from Game Cabinet. In the UK we have the problem that board games are regarded as something you do at Christmas so you don't have to talk to your relatives.
Actually, it is not GA's themselves, which are clearly an effective optimisation scheme for some problems, but the hype and worshipful awe they attract.
Maybe in trying to put down the anti-evolution movement we have gone too far the other way and lost our grip on what evolution as a theory does and does not do. (And the fact that Darwin cribbed it from economics).
My experience is that whenever a young scientist meets GA's for the first time they instantly assume it will solve all the very difficult problems we have been working on for the last 50 years. And it won't. Its just an optimisation algorithm.
GA's won't find any solution you couldn't find by exhaustive search of the parameter space. (But in many cases exhaustive search will take longer than the lifetime of the universe). If your target function won't identify the desired solution, GA's won't fix it.
GA's are a useful tool to be added to the toolkit of optimisation algorithms, along with annealing methods, gradient & curvature methods, quadratic underestimators and such like. For some problems they may be the best of the bunch, but I've yet to work on such a problem.
Still, I don't doubt this is an excellent book, and I'll add it to my reading list.
The module system sounds particularly hopeful, since it sounds as if modules are not loaded until needed, speeding startup and reducing memeory usage. I don't think I've ever used pex, and I don't even know what xie is.
Does anyone know why there are *two* truetype backends?
As Linus says: Linux sucks. It just sucks less than anything else out there.
And so it is entirely just that as soon as there is something better (or even just something with more potential) than Linux, all the enthusiasts will begin to leave.
I'll be among them. But I don't see it happening any time soon, and when it does happen I don't think hurd will be it.
This is partly dealt with by weighting reviews by the authors credibility, and an author has no credibility until they have published something credibile.
There are openings for abuse, for example a small group could submit papers and lend eachother support. A more complex formula for credibility of papers and authors would help. It also won't help the problem of communicating an unpopular idea.
The system could also implement the traditional approach, by simply giving the editorial board and their chosen reviewers credility, and giving none to anyone else.
What I slightly object to in the current system is that my best and worst peer-review papers are given equal weight in any funding review, and all are given equal weight with papers from scientists both better and worse than myself.
Slashdot Mirror
Microsoft delivers an operating system to users (and you pay again for development software).
Linux delivers an operating system and development software to every user.
Which approach is going to encourage more developers in the long run?
As the market stands at the moment, Linux is certainly attractive to would-be developers. And as computer use spreads into countries where it is not yet widespread among the general populace (China springs to mind), Linux is going to have further advantages: It requires less powerful hardware, and it is open to home grown localisation.
I wonder what effect of 1,000,000 Chinese hackers will have on Linux? Of course a lot of work will go into some really good Chinese language support, but I expect to see some stuff come out that we can all use.
Unless Microsoft can buy off China. But I think China is old enough skeptical enough enough to see through that.
Of course, while developers are pretty important, developers without users are of questionable value.
Oops! You are absolutely right. Sorry to everyone I criticised.
I guess if they labelled it a 'fully GPL compatible' distribution, then there would have been no controversy, because (I assume) most of us accept RMS' analysis that while the QPL is DFSG-free, it is not GPL compatible.
Since the name is an obvious homage to RMS, the intent to be 'GPL compatible' might be inferred. But by using the word 'free' instead RH perhaps treated KDE a little harshly.
I mean from slashdot readers, not from RedHat.
Several people jump in claiming that to call the QPL non-free is disingeuous, stupid, or just plain wrong.
No-one jumps in to claim that the NPL is a free license.
And yet (assuming QT2) the two are very similar: both provide the code originator with a privileged position with respect to the code base.
I am perfectly happy to use Netscape and/or KDE (except that Netscape is bloated beyond the dreams of MS), but from an ideological position it is clear that a free distribution (or at least a GPL||BSD||Artistic distribution) should exclude both.
One perceived difference is that many OSS developers do it for the love of coding. This misses the point that many scientists work for exactly the same reason. I could double my salary if I left science, and have spent more than one year working with no income at all, living off savings, just to stay in the field.
If you go up to their page about the MS anti-trust case, they put forward some evidence that prices of software products in markets where Microsoft compete have dropped much faster than prices in markets where Microsoft does not compete. At first glance this suggests that Microsoft is not a monopoly, since monopolies usually exert their influence to keep prices inflated.
However, the reasoning is fallacious:
Technologically, Phat seem to be most mature and advanced, featuring the option of installing to a loopback file as well as UMSDOS, giving potentially better performance. I can't compare the hardware detection in the various products. I presume Armed doesn't do loopback, but their web server won't serve me any pages today, so I can't tell.
But technology will not determine success. My take on the ones which will succeed are as follows:
- WinLinux2000 will win. It has the most accessible and descriptive name. It will appeal to Windows users. PhatLinux is an obscure hackerish joke on the name of the file system. Armed Linux sounds like a secure distro, not a newbie intro.
- Mandrake 6.1 may grab the market if they chose to push this feature, on the basis of their reputation as a conventional distribution. Corel could do even better if they wanted to do UMS or loopback installs, because they have the Windows brand-name recognition.
- DemoLinux will do well, and probably make lots of magazine coverdisks. Again, most importantly, the name says what it is.
Armed are going to take stick for their choice of web hosts, and the fact that their pages are unavailable whenever their product is in the news. Which seems perfectly fair to me.As a community out technical efforts are occasionally sabotaged by lack of attention when it comes to image.
In the UK we refer to English units as Imperial units. While they have the same names as the US versions, the sizes are slightly different.
(e.g. our pint is 20 fl oz, not 16, our ton is 2240 lb).
Imperial measurements were abandoned by schools during the 70's when I was a child. However, they are only now being phased out in retail: it was only a couple of years back when meat started to be sold by the kg rather than the lb.
People still talk about a pint of milk, although it is almost always a 1/2 litre. There are some bizarre mixtures: fabric comes in 45 or 60 inch widths, but is now often priced by the meter lengthways.
Most nuts and bolts, and also spanners, are now metric, although a full toolkit still comes with both - a bit of a waste I guess.
In my scientific work I use the electron as a unit of charge and the angstrom as a unit of distance.
The unix `units' program is always handy if you need to convert forces from hundredweight furlongs per fornight squared into something useful.
A month later New Scientist published this story, suggesting that the slowing was due to the reaction from heat radiated from the probes RTG power plant.
They still appear to be arguing over whether this effect is big enough. Measurements involving heat are notoriously difficult, as the cold fusion debacle showed.
The free software movement seems to span many political and religious viewpoints, and you must have met more of the movement than most people. Do you have and feeling for what worldviews are more common in the free software movement? Is it every difficult working on a shared cause with people with very different motivations?
Eric, in your papers you've put forward many political, sociological and technical reasons why open source software is a good thing. (For example the gift culture is a political model, peer aclaim is a motivation for some programmers, peer review leads to less buggy software).
Every individual will be differently influenced by these different arguments, depending on their political leanings, emotional makeup, and the problems they are trying to solve. Which justification is the one which is most persuasive to you personally?
Everyone seems to have their own history of Galileo!
If I remember correctly for most of his life Galileo was kept by a rich sponser who guided him in 'politic' expressions of his theories. During this period his main trouble with the church was his claim that the moon when seen through a telesope had features on it, wheras the church insisted the heavenly bodies must be perfect. (The moon's visible features were put down to imperfections in the eye).
It was later on when he was without a more worldly advocate and wrote a cutting satirical dialogue protraying the Pope as an ass that the church started persuing him for Copernican notions. My impression of this was that the church was motivated more by revenge than dogma.
Ironically several of the top astronomers of the period, who went on to verify many of his results, were monks at the Jesuit university in Rome.
But it's a long time since I read any of this, so I may have misremembered.
My concern was the appalling reporting, which contained several statements which were factually wrong (and clearly physically impossible), and failed to consider the reliability of the results at all. The real facts were obviously read and then re-written by someone who didn't understand what they were reading.
When I find Sci-Am reports wrong facts in a field I do know, it shakes my faith in what they write about other fields where I am reliant on their layman-explanations.
The work is already published in Nature, where I don't doubt there is an accurate account of the techniques. But I'll wait for the top statisticians in the field to comment, because it is really, really easy to bias this sort of result unintentionally by an inadvertant assumption: There have been several wrong results of this type published before.
The image to which you provide a link appears to have a resolution of around 1.5-2.5 Angstrom (which is reasonable, since that is the size of an atomic tip). Charge density studies and the imaging of atomic orbitals require resolutions of better than 0.7A, I suspect better than 0.5A for the results claimed in Scientific American.
(That is using resolution as defined in crystallographic terminology, it might be used differently in STM).
The X-ray and electron beams are not combined, they are collected separately, and the information is combined.
Moreover, the principle difficulty has been completely ignored. Electron beams can be focused to produce a direct image. X-rays cannot be focused for imaging purposes (although crude focussing to concentrate a beam is just possible). As a result, you only get a diffraction pattern with no phase information. The image must be reconstructed by Fourier transformation, which needs the phases. (There is a strong analogy with optical holography, in which a reference beam must be interfered with the diffracted beam to obtain phase information, but with x-rays the coherence length is too short to get a reference beam).
The trick is to use the phases from the low resolution electron image, and some mathematical relationships to reconstruct the missing phases in the high resolution image, which will show your electron orbitals. The problem is unless the statistics are treated very carefully, all you get is an image which confirms the assumtions of the model you used to get the relationships with which you reconstruct the phases.
The mathematical techniques were just coming on line in X-ray crystallography in 1996 and there was still considerable debate back then over their correct application. So there is a fair possibility that these results are correct, but I would suspend judgement until they have been scrutinised for a year or two.
Switching to the default theme helped. I also dumped E and switched to wmx (which is only 82K compiled with Gnome support, no pixmaps, stripped).
The other big gain was setting the default help browser from Netscape to the gnome help browser. (For some reason the control center wouldn't let me do this at first, I eventually editted a text file. It works now). The help browser has some limitations, most noticably no new window on button2, but is OK in general. It would be interesting to try Gnome using Kfm for both help browsing and for desktop icons, since gmc is quite heavy.
Having said that, I never tried themes under KDE, so I don't know if it has the same difficulties. I recently added a cheap 16Mb dimm: having 48Mb makes all the difference.
The way the market has been going recently, it was beginning to look as though US vector supercomputers were dead, and only the Japanese were still advancing them. The T90 was the last major vector machine, but had memory synchronisation problems with more CPUs. The SV1 had some interesting specs, but I don't know of any site which actually installed one - certainly press releases were thin on the ground.
Parallel machines, such as the Cray T3E, IBM SP2+, and to a lesser extent Beowulf clusters just give so many more Gflops/$. But as has been pointed out they are completely unsuited for some problems, for which you simply need all you power concentrated in a small number of CPU's.
My guess is that there is not enough market for a new US vector supercomputer, and the US government are stepping in so not to become dependent on imported hardware. If most SV1 installations are government, it might explain why we've heard so little about them.
BTW, many older supercomputers were single-user machines, which required a front end running a mutli-user operating system to schedule jobs. However all recent machines, including the C90, T90, T3E and I suspect the SV1 and SV2 run their own operating system (they are self hosting). In this case it is UNICOS, a Cray Unix which is gradually being merged with SGI IRIX.
The UMSDOS clone worked OK, but the problem is that Linux needs thousands of tiny little files. These are stored rather inefficiently under UMSDOS, and become a nightmare if you need to defrag (for example to partition for a full Linux distro). The performance hit is also noticable. (Notice that WinLinux2000 occupies 500Mb on a FAT32 partition or 1Gb on a FAT16 partition - that is the impact of all those little files).
My guess is that the loopback filesystem, as used by Mandrake's Lnx4win, is a better choice. Hopefully the WinLinux people will add this option in future. Otherwise, a great piece of work - I love the idea of stealing all the settings from Windows.
Also worth looking at is DemoLinux, which runs off the CD.
My laptop now runs Slackware 3.4 in its own (half empty) 120Mb partition, with X, compilers, networking and freeciv. I might try Debian 2.1 if I need libc6.
For improvisation with rules or preparation, you might try a story-telling game.
Once-Upon-a-Time involes making up a fairy-tale including all the elements on cards in you hand, and ending with the line on your ending card. Other players may interrupt if you mention anything on one of their cards, or by using a generic interrupt when you play a card, and then have to continue the tale.
Baron Munchausen is even more free form, with little tokens as the only props. Players take it inturns to tell tall tales of their adventures. Other players may interrupt with challenges to the story (by paying a token), in which case the storyteller must either correct themselves or pay to refute the challenge. At the end, players spend their remaining tokens voting for the best story.
- For example, Catan, Manhattan, 1630-something, Civilisation and both Illuminati games are point totting games, but only the first two are German, the third being a rare good UK game, the rest being American of course.
- Mississippi Queen, Ricochet Robot, Elfenland, Bausak (where you build a tower of awkward wooden blocks), Carabande (a big wooden motor racing game where you flick the counters) are all German (some republished elsewhere), and don't involve points.
I would agree that point totting games can get frustrating if you play with people who are too deparate to win. In Manhattan in particular the last player each turn can often spend 1/2 hour figuring a play which will turn the whole game upside down (but I've seen Monopoly played that way too). The solution is to play games for fun. Beer helps.Program your robots a turn in advance to navigate around a maze containg hazards, conveyors, turntables etc. Confused by the fact that you may not have the right program cards, another robot may bump into you throwing your calculations off, and the robots shoot at each other. Long.
Much simpler, and yet far harder. Move robots with no brakes around a board to reach a target. You have to hit things to stop. Usually there is nothing in the right place to bounce off of. So you have to move several robots. Sometimes you have to work out 20 or 30 moves in your head, and then announce before anyone else gets there. (Best call I know of was 63, which involved iteratively bouncing two robots off of eachother).
Probably the best board game ever. A sort of colonisation/town building game, with a random board made up of hex tiles. Superbly balenced, and reasonably quick.
Actually, this one is easy enough for non-geeks, but has some of the same sort of puzzles as RoboRally - work out how fast you can go without ramming an island in a randomly twisting river.
Euphrat und Tigris (Hans im Gluck) is good but I can't work out how to win. Sixteen-thirty-something (Warfrog) is a very strange twist on the normal board wargame idea.
Needless to say, the best boardgames come from Germany, although there are some good US companies too. Rules translations are sometimes needed from Game Cabinet. In the UK we have the problem that board games are regarded as something you do at Christmas so you don't have to talk to your relatives.
And neighbours in crowded streets who play their music too loud. And people on trains who play their portable stereos to loud.
In fact, I can think of no useful uses against computers, but heaps against sound systems.
Maybe in trying to put down the anti-evolution movement we have gone too far the other way and lost our grip on what evolution as a theory does and does not do. (And the fact that Darwin cribbed it from economics).
My experience is that whenever a young scientist meets GA's for the first time they instantly assume it will solve all the very difficult problems we have been working on for the last 50 years. And it won't. Its just an optimisation algorithm.
GA's won't find any solution you couldn't find by exhaustive search of the parameter space. (But in many cases exhaustive search will take longer than the lifetime of the universe). If your target function won't identify the desired solution, GA's won't fix it.
GA's are a useful tool to be added to the toolkit of optimisation algorithms, along with annealing methods, gradient & curvature methods, quadratic underestimators and such like. For some problems they may be the best of the bunch, but I've yet to work on such a problem.
Still, I don't doubt this is an excellent book, and I'll add it to my reading list.
Hopefully not much bigger than 3.3.1?
The module system sounds particularly hopeful, since it sounds as if modules are not loaded until needed, speeding startup and reducing memeory usage. I don't think I've ever used pex, and I don't even know what xie is.
Does anyone know why there are *two* truetype backends?
As Linus says: Linux sucks. It just sucks less than anything else out there.
And so it is entirely just that as soon as there is something better (or even just something with more potential) than Linux, all the enthusiasts will begin to leave.
I'll be among them. But I don't see it happening any time soon, and when it does happen I don't think hurd will be it.
This is partly dealt with by weighting reviews by the authors credibility, and an author has no credibility until they have published something credibile.
There are openings for abuse, for example a small group could submit papers and lend eachother support. A more complex formula for credibility of papers and authors would help. It also won't help the problem of communicating an unpopular idea.
The system could also implement the traditional approach, by simply giving the editorial board and their chosen reviewers credility, and giving none to anyone else.
What I slightly object to in the current system is that my best and worst peer-review papers are given equal weight in any funding review, and all are given equal weight with papers from scientists both better and worse than myself.