Slashdot Mirror

Politics of Division & Hatred. Its far easier to convince someone you are right if you don't bother with reasonable arguements. Ann Coulter makes a living claiming people who disagree with her nutjob point of view are commiting treason, today I read a forwarded email from a very intelligent person that claimed everyone not current or ex-military.police/fire/etc was a sheep with their head in the sand. My grandmother oalmost was ill because I'd rather have an ex-marine and Vietnam vet who protested the war he just returned from (I know we could have one that one if it weren't for all those damned hippies 6,000 miles away) than a party-boy failed businessman who froze for 7 minutes in a classroom when our country was attacked (seriously, he put our counry at risk because he didn't want to scare a classroom full of kids by excusing himself from the room? How do people justify this foolishness?).

Which isn't to say the left doesn't have its share of nutcases and extremists too. Its just that when I eliminate the noise from both sides frothing nuts, the democrats seem to be the party of fiscal responsibility, that seems to have the most control of their extremists, thats most willing to but out and let me lead my life.

What everyone seems to be missing is that saving seeds is not practical with the most widely grown non-GM corn varieties, either. The vast majority of the corn that is commercially grown in the US comes from hybrid varieties that do not breed true, and are far less productive in the second generation (grown from saved seed). That's been true roughly since the Second World War. So if we shouldn't be giving GM corn to people in the third world, we shouldn't be giving them "conventional" corn, either, for the same reasons.

Nearly all the field corn now grown in the United States and most other developed nations is hybrid corn.
http://en.wikipedia.org/wiki/Heterosis

The primary disadvantage of hybrids is the seeds cannot be saved from year to year. Seeds saved from hybrid plants usually will not produce the same plant the following year because most varieties are not self-sustaining. Offspring of hybrids usually show an unpredictable mixture of characteristics from the grandparent plants instead of being similar to the parent.
http://www.urbanext.uiuc.edu/hortihints/0102a.html

You didn't follow the links? I wouldn't call a 65 node cluster just "exploring".

http://arrakis.ncsa.uiuc.edu/ps2/using_vector_unit s.php

There's other projects too:

http://www.sve.man.ac.uk/Research/AtoZ/Playstation 2

The National Center for Supercomputing Applications at the University of Illinois had a PS2 cluster. I think the geology department had a few too.

http://arrakis.ncsa.uiuc.edu/ps2/

Well mostly, the "issues" are more complicated then one might think:

Encryption is classified as a munition under the Internation Traffic in Arms Regulations (ITAR)

source: http://lrs.ed.uiuc.edu/wp/privacy/encrypt.html

A lot of the time the purpose of the military is to "implement policy" (i.e. "kill people and break things"), but some military branches (at least in the U.S.) do good works, and try to save lives.

As written, the "no military use" GPL extension doesn't specifically forbid this type of use, but I'm sure the military isn't going to want to keep track of licenses/usage across its fixed and mobile facilities, so stuff created under this license won't get used at all. Is that a good thing?

I'd say GCC does far more optimizations than FPK or Delphi

Even if you turn off all optimizations GCC is still quite slow, in particular when compiling C++ code. Things got a lot better in 4.0 and 4.1, but it still not exactly fast.

Honestly, how much time do you spend rebuilding your entire application?

Since you ask: actually, since I work on FPC (not FPK, that name is long gone), I spend a lot of time rebuilding the entire compiler and run time library (often 2 to 3 times in a row, compiling the next iteration with the previous result) as a first quick check against newly introduced code generation or parser bugs.

Chances are if you have to rebuild a million lines of code when you fix a typo in some function, then you need to really think about factoring your code properly. I can see if we were talking about 1000 loc/min vs. 1000 loc/sec or whatever. But if Delphi's only feature is that the [from the customers point of view] the SINGLE build is fast but the code generated sucks ... then you're stuck.

As you know (since you mention it below), that isn't Delphi's only feature from the customers point of view. But it's nevertheless extremely convenient when you're used to it. People have been ranting since day one on Apple's Xcode list about the slowness of GCC compared to CodeWarrior (even when comparing a distcc'd build against a single threaded CodeWarrior build), also in debug mode (i.e. no optimizations). Compilation speed is far from a non-issue these days.

Also, Delphi's code generator doesn't suck, it's just average. Which as it happens didn't matter all that much for x86 until now, since the processor itself optimizes a lot behind the scenes. Also, unless you are building numerical applications or 3D engines, most advanced optimizations such as loop tiling, all sorts of prefetching tricks etc have little impact.

I took a course 2 weeks ago at a summerschool from David Padua who works a lot on compiler optimizations. His conclusion was all the time that basically all current compilers suck at optimizing, except in very specific cases and often only if you know how to write your code so the compiler recognises what you want to do.

In my personal experience, a good register allocator and some peephole optimizations can get you a long way with most applications.

GCC performs a lot of passes on the code to look for ways to optimize the code it produces. I think you should take some non-trivial function, pass it to GCC with "-O3" and then look at the assembly code it produces. You'll be amazed at the complexity of the output.

I've already spent plenty of time looking at assembly code generated by various compilers, including GCC :)

Friers set up for french fries are set at about 350 degress Fahrenheit. Ice is presumably 32 degrees Fahrenheit or lower.

That's not the biggest part of the "problem" (or "solution", according to your story!).

If the ice simply floated on the grease, it would erupt in a burst of steam and that would be it. But water is denser than oil, so your ice cubes sink to the bottom. There, they erupt into superheated steam, which is far less dense and comes exploding out of the fryer, carrying that 350-degree grease along for the ride.

I enjoyed "breaking in" the new grease -- while it was still clear enough to see through -- by dropping a small piece of ice into it. A small enough piece will fall to the bottom, but will just melt and stay there a bit. Eventually, it comes floating up to the surface (at which point you'll want to step back). I guess some combination of pressure, thermal inertia, and surface tension kept it from exploding immediately. I didn't think about it much at the time -- I had burgers to flip.

By the way, if someone offers to take you for a swim in the grease tank, DON'T DO IT! (Hint: you are about as dense as an equivalent volume of water)

I'm referring to intrinsic curvature (as opposed to extrinsic curvature) - the idea that it's a property of the surface itself, and not how it is bent in a higher dimension. See http://mathworld.wolfram.com/Curvature.html or http://www.geom.uiuc.edu/docs/doyle/mpls/handouts/ node21.html .

This has to be done because not everyone is a scientist with an experimental, discovery mindset. Everyone is also not a C++ programmer (or Java, Perl, C, C#, etc.) and thus proficient at error checking and dealing with a variety of system interaction. Face it, people hire scientists to develop things no one else is doing, but who learned programming as a necessary evil. People hire professional developers to glue or reform that prototype effort into something a customer wants. I found that my college training taught me to understand what a scientist says, but I never really developed the mindset to be a scientist. Neither could do the other's job to everyone's satisfaction. I serve in roles similar to that in my job where I bridge everyone's needs and goals.

We have faced this decision recently with a Matlab-like program called IDL, by RSI Inc (now called ITT). Our future work is getting so big and monstrous that IDL cannot deal with the datasets we need to process. Fundamentally, our scientists love the language - mostly. The CS people think it sucks as a language. But the language developers wisely provided ways to bridge their work with your code, and vice versa. They even developed an IDL-JAVA bridge a couple of years back. They have rudimentary support for XML (one of our coders rebridged that to Xerces a bit better) and he also bridged in HDF5 better than they did.

Still, our original concept of build it in IDL and optimize the hard parts in C just simple isn't going to cut it now. So now we will build the framework in C++ and let them add plugins that use IDL. C++ provides access to importing/exporting all the nasty data types, user interface, and handles the bulk of memory management. C++ also provides an easy bridge to a multitude of other languages.

It's tough, but the scientists are going to get the flexibility to poke and prod with new ideas, while the C++ folks keep the production users happy.

Since use of the thumb is a big issue, I'd suggest showing her a built-in usability function that comes with windows, MouseKeys. When turned on it converts the number keys into mouse function keys (movement, single click, double click, drag, drop, etc). Now normally this is used to avoid all use of the mouse, but can be a little slow moving the cursor around. In your case thought it would allow moving the mouse with the right hand and using the left hand to do the clicking/dragging. Personally I use it a lot just because using the keyboard keys is a lot less stressful than clicking the mouse keys because you don't have to grip in order to push the buttons and there's less strain, especially when you have a lot of repeated clicking in a short period. There's of course a period of adjustment before it becomes second nature to use Mousekeys but it's a no cost, effective solution. And you can toggle it on and off just using the NUM Lock key so you still have access to the number pad for number entry.

Here is a link to Microsofts description of how to turn them on:
http://www.microsoft.com/enable/training/windowsxp /mousekeys.aspx

and a nice tutorial on what keys do what and tweaks:
http://www.disability.uiuc.edu/infotechaccess/trai ning/windows95/mouse.html

This describes SSI. I believe I've been building sites for over 10 years with this. Does anyone know when it was first introduced. I have a reference back to 1995

Seems to be a rather armchair explanation...

In any case your conclusion is exactly what my conclusion was -_-

Except I don't see how you wouldn't know exactly what the machine did. Of course you can! If you made it, you can analyze very precisely every single computation down to the bitwise level. You will know exactly why the computer made every action that it did. That's what makes computational mimickery so useful. The hardest part about linguistics is that we can't analyze our brain activity precisely or accurately yet. That's why theoretical linguists have jobs. If we can get a computer/robot to do exactly what we do, then we have a computationally and procedurally adequate system, but as I said, not the exact implementation in the natural world, probably.

All linguists from Chomsky onward recognize language as being a computational system. Just read Chomsky's first book publication, Syntactic Structures. That said, computational systems that generate sentences primarily by traditional syntactic analysis are a dead-end, because they don't take into acccount the probabilistic nature of human grammars.

I've gone through the computational approach that would be required for this type of learning mechanism in quite a bit of detail. It's a link-up of various Hidden Markov Models that correspond to the various sensory information. They are combined in a single HMM to classify how the real world corresponds to the language the bots are hearing. This allows for an associative memory in the system to store a semantic representation of the real world.

Check out my professor's webpage, basically the same sort of thing the dudes from Italy are doing:
http://www.ifp.uiuc.edu/speech/acquisition/acquisi tion.html

Humans are not predestined to learn language. If someone is not exposed to language, they do not learn it, and will never, after a certain age. Humans may learn language in a similar way as these robots, at least initially. The languages generated by the robots are much(like, orders of magnitude) simpler than human language, and so take shorter to learn. It actually takes human children a very short time to learn grammar as far as these robots have learned. If we could eventually code a system in which robots learned language exactly as children do, it would take a few years for them to learn as well, methinks.

And ID does not depend on a perfect design, just an intelligent one. And if you want to take the Christian tack to ID, then we would actually expect our language learning system to be imperfect since we're fallen(read:not completely optimized) beings. Considering how difficult it is to make these kinds of systems in robots, and the difficulty of the maths involved to even approach the problem, I'd consider it a pretty intelligent design.

Well it would be nice if the HTML 'p' tag actually worked properly on slashdot! Give him a break.

I've read about the same advances from the University of Illinois so I corroborate it.
here's another (slightly older) article - http://www.news.uiuc.edu/scitips/03/1106feng.html

Talking about bringing out the grinders. Anyway.

I agree that people are going to be buying ever larger and more powerful systems. Yes, this will make bloat less important. Yes, this will give jit type languages a bigger role in larger applications.

I'd rather the future become more charming. http://charm.cs.uiuc.edu/

Ok let's think about it. Magnets react to magnetic fields, right? But what about metallic coils? After all, electromagnets are made with coils and metal.

So imagine a future where nanoscopic coils, or maybe thin-film nanomagnet arrays can be implanted under your skin. Or behind your ears, so you can sense magnetic fields in 3D.

Let's not forget that some animals are sensitive to magnetic fields, and others can sense electricity miles away.

Yes, that sounds good. I need to take a course like this to catch up on some crypto for work!

Sorry, the URLs were stripped for some reason. Possibly because I put them in

Google search for "global warming volcanoes research". Click "scholarly articles".

http://scholar.google.com/scholar?q=global%20warmi ng%20volcanoes%20research&hl=en&hs=KO9&lr=&client= firefox-a&rls=org.mozilla:en-US:official&oi=schola rt

NG Andronova, ME Schlesinger - Geophysical Research Letters, 2000
Causes of Global Temperature Changes during the 19th and 20th Centuries
Climate Research Group, Department of Atmospheric Sciences
University of Illinois at Urbana-Champaign

http://crgd.atmos.uiuc.edu/publications/Causes_of_ dT.pdf

Just don't leave the country again Zimmerman...or you may end up locked inside that customs office where they 'want to leave lawyers out of this' again. :)

PGP Story:
MPG 1.1G
WMV 378M

Just don't leave the country again Zimmerman...or you may end up locked inside that customs office where they 'want to leave lawyers out of this' again. :)

PGP Story:
MPG 1.1G
WMV 378M

Impeachments aren't waiting in the wings, held back by some action from an administration. They are brought to the person in question based on actions, lying to grand juries, etc (ask the last president)

Actually, if two states file for impeachment, the Congress has to start proceedings.

It's this thing called the Constitution: learn it, love it.

We have to remember the last Presidency to fall for this was for just using tape recorders to tap just one phone, which then revealed taped conversations in only one room (the Oval Office) - the information in those tapes was what resulted in the hearings.

Oh, and there was some issue of a quagmire of a war that we didn't need to fight that was bankrupting the nation for no reason. no historical correlation to today, of course ...

Nixon resigned because of the Watergate scandal which climaxed with the the Watergate tapes, uncovered by Mssrs. Woodward and Bernstein.

But he wasn't impeached. He would have been, of course; articles of impeachment were already being planned. However, he wasn't, so the last President to "fall for this" would have to be Andrew Johnson.

More importantly, Nixon's troubles had very little to do with his role in Vietnam. The country had been unhappy about Vietnam, and they were unhappy with the incident at Kent State (which led to the CSN song about "four dead in Ohio"), but the nation didn't blame Nixon for the war per se.

For one thing, our involvement in Vietnam ramped up under Kennedy and Johnson. For another, Nixon was the one who brought the troops home. Here's a timeline for ya: here.

Nixon did make a couple of unpopular Vietnam decisions, such as the Cambodia and Laos actions, but by '74 when he resigned, the nation understood that the troops were coming home.

that's a problem for people that only believe what they see on what they presume to be un-alterable video tape... but why not just ask the people that watched it happen.

As any psychologist will tell you, your memory is pretty easily alterable. Elizabeth Loftus is probably the most prominent name in the area of altering memories, and has done a number of interesting experiments.

My favorite experiment involves counting the number of passes that people make with a pair of basketballs in this video here (someone warn the University of Illinois, they might experience a slashdotting!).



After you've watched the tape, you can read about how perceptive you are (or aren't) in this article.

in PDF

Sure there is, buy an Onyx2 and get some projectors. They can be had from anywhere between $0 and a few thousand dollars depending on config/circumstances. If you're lucky you can find one in a dumpster. You can find some Onyx3x0/3x000 on the market (most are still chugging along with their original owners) but the cost is very high .

A single pipe with a DG5-8 can drive up to 8 monitors, not sure of the res./monitor you'd get though. I only have specs for the InfiniteReality4 (IR4) but 16 pipes gets you 133 million pixels, I think this is the same for IR3 was well. Realistically you'd only need at most one pipe per wall of your CAVE.

There's a port of Quake, Quake II, and QuakeIII for SGI, I have it on my O2 :)

Here's a very cool page, this guy actually works with a CAVE (and an Onyx2) and he likes Quake! (Also, checkout the wall this guy has access too!)

It's a sad story about SGI, I think even if they survive long term as a company that the "G" in their name is already dead.

Sure there is, buy an Onyx2 and get some projectors. They can be had from anywhere between $0 and a few thousand dollars depending on config/circumstances. If you're lucky you can find one in a dumpster. You can find some Onyx3x0/3x000 on the market (most are still chugging along with their original owners) but the cost is very high .

A single pipe with a DG5-8 can drive up to 8 monitors, not sure of the res./monitor you'd get though. I only have specs for the InfiniteReality4 (IR4) but 16 pipes gets you 133 million pixels, I think this is the same for IR3 was well. Realistically you'd only need at most one pipe per wall of your CAVE.

There's a port of Quake, Quake II, and QuakeIII for SGI, I have it on my O2 :)

Here's a very cool page, this guy actually works with a CAVE (and an Onyx2) and he likes Quake! (Also, checkout the wall this guy has access too!)

It's a sad story about SGI, I think even if they survive long term as a company that the "G" in their name is already dead.

"One could readily imagine many uses for immersive 3D environments from remote medical procedures to collaborative architecture to interior design to automobile sales to video games to many other things."

This bibliography is organized to provide a structured introduction to graphical interfaces to information systems.

Geospatial Decision Making Blends 3D World Visualizations With Background Data

IMPROVING HUMAN SPATIAL COGNITION OF BUILT ENVIRONMENTS USING COMPUTER GAMING TECHNOLOGY

You might want to check out RingCycle, a research project at my university. It probably won't end up portable beyond x86, but it provides protection against buggy drivers, with no changes (other than a recompile) necessary to those drivers, and only minimal performance impact.

To add to your comment but not directly a reply to you...

The purpose of thermal paste.
If you could have two perfectly flat and perfectly smooth surfaces and mate them together, you would not need thermal paste. The problem is you can not achieve that level of smoothness. Thermal paste is used to fill the void and flatten out the peaks and valleys between the two surfaces. In theory, you should still have some metal to metal contact throughout the mating surfaces. Here is a link to a PDF of a guy that ran some tests that shows that even a layer of .003 inch thick layer (thickness of 1 sheet of paper) of thermal paste was worse then no paste at all on a power transistor. In fact, they calculated that the transistor running at 30W with the .003in layer of thermal grease, the temperature was 20C higher.

I'm sure different thermal coumpounds and different surface areas would result in different temperatures and the link I provided is one persons single test but the point is, less thermal paste is better then more. We were shown that concept in Electronics 1 in high school.

http://www2.english.uiuc.edu/cybercinema/sounds/fo rbin_sounds/coloss04.wav

You mean something like this?

Just out of curiosity, does anyone not think a national bill involving the sales of videogames to minors is not now inevitable, considering just what kinds of testimony Congress is going to spend their time listening to?

Another link with more info:

http://access.ncsa.uiuc.edu/Stories/TobaccoMosaic/

The main research page may interest some of you. And for those that it doesn't help, perhaps you want to look at our Linux clusters instead?

The main research page may interest some of you. And for those that it doesn't help, perhaps you want to look at our Linux clusters instead?

Dewey can work just fine for large specialty libraries. Witness the many speciality libraries at UIUC--which is the largest public university library in the US, with more than ten million volumes.

During heavy disk read activity, the HD is only uses 15% of all the power. (source) The real key to decreasing laptop power consumption is dimming the screen, which can reduce power consumption percentage from 26% down to 7%.

I shred it, then I set it on fire. I then take the ashes and compress them into a diamond-like form. Then I smash it apart, and put the crystal shards inside the event horizon of a black hole, beyond which no information about the black hole's interior can escape to the outer universe.

its the only way to be completely sure.

Actually, it was run on the NCSA Altix because it's faster than the Xeon cluster and not as crowded.

I think you'd better make sure the NCSA allocations staff doesn't hear you say that "NCSA didn't have anything to do with it". You may have some issues accessing our resources in the future if you're unwilling to give credit for the computing power we provided to perform the simulation.

See:
http://access.ncsa.uiuc.edu/Stories/TobaccoMosaic/

And given that the abstract of their paper states that a 1 million atom simulation was accomplished, they probably had excellent speedup.
Most likely, the simulation was run on NCSA's Tungsten cluster, which has a low latency Myrinet interconnect, just to provide the capability to run NAMD at scale on thousands of cpus.

Here. They're just looking at what holds together the structure of the virus.

Umm... what do you think scales better than NAMD, exactly?

GigE lets us scale to 32-64 nodes reasonably, with the ApoA benchmark. Using Myrinet or Infiniband, you can bring that up to a thousand or so processors.

Also of note: we've done a series of cluster-building workshops specifically focusing on the software and hardware required to run these kinds of simulations. Copies of the presentations and tutorials are linked off of that page.

And if you want to see how we designed our clusters, I've got full specifications up here.

Also of note: we've done a series of cluster-building workshops specifically focusing on the software and hardware required to run these kinds of simulations. Copies of the presentations and tutorials are linked off of that page.

And if you want to see how we designed our clusters, I've got full specifications up here.

1. The full research page for this project is here. This is a lot better than the stuff linked through Nature and such.

2. The image was actually generated by our group, and specifically Anton Arkhipov, using our software package VMD. NCSA didn't have anything to do with it.

1. The full research page for this project is here. This is a lot better than the stuff linked through Nature and such.

2. The image was actually generated by our group, and specifically Anton Arkhipov, using our software package VMD. NCSA didn't have anything to do with it.

1. The full research page for this project is here. This is a lot better than the stuff linked through Nature and such.

2. The image was actually generated by our group, and specifically Anton Arkhipov, using our software package VMD. NCSA didn't have anything to do with it.

1. The full research page for this project is here. This is a lot better than the stuff linked through Nature and such.

2. The image was actually generated by our group, and specifically Anton Arkhipov, using our software package VMD. NCSA didn't have anything to do with it.

Finally, I can say this for real: Imagine a Beowulf cluster (link is to Biowulf) of these!

The modeling software they used is called NAMD, free open source "parallel molecular dynamics code designed for high-performance simulation of large biomolecular systems" that will run on commodity clusters of tens of Linux PCs on gigabit ethernet. In other words, you too can run the virus simulation on your own Beowulf cluster, if you don't mind it taking some years to run. According to NCSA's own press release about the virus simulation, it "only" took 35 processor-years, so if you have a 100 fast Linux PCs on a gigabit network lying around you can do it yourself in not much more than 4 months.

Finally, I can say this for real: Imagine a Beowulf cluster (link is to Biowulf) of these!

The modeling software they used is called NAMD, free open source "parallel molecular dynamics code designed for high-performance simulation of large biomolecular systems" that will run on commodity clusters of tens of Linux PCs on gigabit ethernet. In other words, you too can run the virus simulation on your own Beowulf cluster, if you don't mind it taking some years to run. According to NCSA's own press release about the virus simulation, it "only" took 35 processor-years, so if you have a 100 fast Linux PCs on a gigabit network lying around you can do it yourself in not much more than 4 months.