I'm speculating here (I haven't seen XGrid specs yet).
If (and only if) XGrid is one of the Single System Image clustering technologies like opnMosix, then things like MPI jobs are pretty nice to work with on these kinds of clusters.
In openMosix, the load balancing system simply migrates individual MPI processes to nodes that are the least loaded (and which are projected to provide the maximum instantaneous throughput for the job).
Assuming that XGrid has a similar architecture (a big assumption), having a commercial implementation working across Apple's relatively stable hardware platforms (with AltiVec doing vectorized floating point behind the scenes on each node) sounds like a pretty decent cluster implementation to me. Now, as always, Apple must work on cost.
You are correct that only certain kinds of problems "work" well on LAN clusters. But experience shows that a sufficiently motivated computational scientist can adapt many more "big grunt" problems to various weird-and-wonderful architectures than might be apparant at first glance.
Adapting codes to fly on "true" supercomputers (e.g. Big Iron vector machines) is just as much an art form as adapting them to fly on MPI.
Professor Gold is not only the progenitor of the primordial mantle methane, errrr, hypothesis, he is also the progenitor of the idea that the Lunar landings would not have been feasible because the surface dust was not solid enough to support the weight of a lander.
This idea is rumored to have prompted the entire "Ranger" series of spacecraft (at umpteen million dollars) primarily to demonstrate that the Lunar surface was solid.
Oh, and the supporting statements by Steven Soter in the New Scientist article regarding the thermodynamics of photonic momentum-transfer? Three guesses who was Steven Soter's Ph.D. adviser...
Just repeat after me: "Tenure is *good* for science. Tenure is *good* for science. Tenure is *good* for science."
On the contrary, while the receivers that you spend a hundred bucks on are indeed not accurate enough, GPS based geodesy is a raging success. They use very expensive receivers with multiple frequencies and occupy sites for hours at a time to get the kinds of numbers needed for geodetic measurements. Been going on for years. The major inaccuracies have to do with index-of-refraction effects in the atmosphere (hence the need for multi-frequency instruments).
2.4.20 out of the box? Not in my experience...
on
Linux SMP Round-Up
·
· Score: 1
A quick comment on the toss-away statement in the article that 2.4.20 supports 7505 based systems out of the box.
Be Careful(TM).
The AGP3 stuff requires a patch to stock Marcello/Linus kernels for the 7505 chipset.
I had trouble getting an AGP4x card to work on a Supermicro X5DAL-G board (baby brother to the reviewed X5DA8 board; but at ATX size instead of EATX and able to support unregistered memory) without applying this patch. Once patched, it works fine.
I'm not sure if 7505 support has made it into Marcello's 2.4.21preX series yet, but 7505 support does seem to be in 2.5.6x series (which I'm having trouble getting to boot for unknown reasons).
It has not been covered in ice throughout geologic time.
It's virtually certain there are sedimentary basins for hydrocarbons, and "hard rocks" for metals.
All of this is from geo 1 level knowledge. I have not personally worked on Antarctica, although many geologists have, and their publications should be in the open literature.
If it weren't for the treaty, IMHO, there would already be operating oil fields on or offshore Antarctica.
(And slashdotters are worried about a buldozed ice road doing things to the environment?)
Paul Graham's presentation revolved around a Bayesian algorithm he'd devised which put more weight on features in the headers, as opposed to the bodies, of email; he claimed something like 99.5% effectiveness with only something like 5 false positives in 4000 emails sorted.
Quoting from Tim Peters (the real TimBot, but I digress <wink>) spambayes has a more effective classifier:
We have three
categories: Ham, Unsure, and Spam, and I haven't seen anything to make me believe that a finer distinction than that can be quantitatively justified (but my primary test data makes 2 mistakes out of 34,000 msgs now -- that's what I mean by "can't measure an improvement anymore", and a finer-grained scheme isn't going to touch those 2 mistakes; one of them is formally ham because it was sent by a real person, but consists of a one-line comment followed by a quote of an entire Nigerian scam spam -- nothing useful is ever going to *call* that one ham, and it scores as spam *almost* as solidly as an original Nigerian spam).
While spambayes isn't really Bayesian (anymore; it started out roughly that way, as I understand it), the name stuck
It is a statistical filter, and you do need to train it with your personal collection of spam and ham. However, most of the work in the last month or two on the spambayes list has revolved around building user interfaces, and finding appropriate places to inject the filter into your mail processing, not on improving the classifier.
"Spambayes. Try it. You'll like it!" (But be warned, it's still pre-alpha...)
Lots of things to like about Python, but NumPy is one of the better ones, IMHO. There is a largish and growing community of numerical jocks coalescing around around Python in scientific computation. NumPy makes Python into an "array language" (like Matlab, S-Plus/R, APL, etc. etc.) where the crunching is heavily optimized C code. Links to LAPACK et al., and Fortran wrappers exist in f2py and PyFort, for that old still-running-after-all-these-years code.
(Just another enthusiast, dabbling in the religious-war du jour. Flamage to/dev/null;-)
The H is stored as interstitials in the crystal lattice of the metal. The rate of leakage is determined by the diffusivity of the H through the interstitial "lattice", which means that it is thermally controlled. H discharging from a metal hydride cools the metal (so much so, that it might be useful as a source of "coolth" co-generation). Implication: the discharge rate slows down when leaks happen. Oh, and the energy density can be much higher that that of liquid H.
OK. You're concerned about security. That's what the article in New Scientist, and the lead entry in/. implied. Personally, I think that going for security with this kind of system is a pretty darned big ask (although not beyond the realm of possibility). Good luck to Edelman and his crew, but I'm not holding my breath.
IMHO, as others in this thread have also pointed out, trying to do this on the net, using router delays etc., is just plain silly! (I wouldn't mind being proven wrong, on this point, however!;-)
With regard to your points about turbulence, I'll grant that you do have some valid concerns. However, IMHO, the turbulent/time-varying propagation medium is not the heart of the problem, especially when you are talking about something as big as the ocean basins. I think that the scatterers defined by the bathymetry of the ocean basin will be far more dominant (in determining the quality of the final constructive interference) than the relatively minor effects of a time varying wave-speed water column. (The effects of the bathymetry, incidentally, is what the time-reversed acoustics gets you almost for free!) However, that is mere speculation on my part, and the answer is best determined experimentally. That is what Edelman and his group have done.
The time varying prop. medium plays more of role like slightly varying the transceiver positions before playing back the signal. It changes the phase (and possibly the propagation direction) of the local signal. But the signal at the "sweet spot" of the focus is composed of the sum of all the signal paths, and so, on average, the quality of the constructive interference is more sensitive to the "average" phase roll over all paths (modulo the energy leaked out of the system by the differential scattering, of course, but lacking complete spatial coverage with transceivers, you expect energy leakage problems in the first place). If that "average" Feynman-like-path-integral phase roll is time varying, then yes, your concerns about chaotic media are quite valid. On the other hand, if that "average" phase shift is constant, then accomplishing the constructive interference shouldn't be too big of a problem. Which is right? I don't know, and don't know if anyone else knows either. (I do know which way I'd bet, however.) In fact, I kinda suspect that *both* end-members will crop up in different cases (albeit with different probabilities). That's another reason why the experiment is important.
So, bottom line, we agree that there is likely to be turbulence in the propagation medium. I think that won't be too large an issue to accomplish time-reversed acoustical constructive interference at the focus, and Edelman's experiment seems to support that viewpoint.
As to your accusations of my being a pretty big geek, or having a degree in physics, neither is true, actually. Ummmmmm, "I'm not a seismologist, but I play one on TV", OK?
This system is trying to exploit chaos theory without taking in to account the problems it would impose.
Sorry, but I think you might have a fundamental misunderstanding here. The time-reversed (acoustic/electromagnetic/arbitrary-wave-equation system) phenomena that this technique is seeking to exploit have very little to do with chaos and nonlinear dynamics. They have far, far, more to do with interference in propagating waves.
Think of the wave interference tank in your high school physics class. Randomly place a bunch of rocks into the tank; these are the scatterers. Drop a penny into some arbitrary spot. See the pretty interference patterns (at least in your mind's eye) as the waves spread out? Now consider that arbitrary (perfectly linear) solutions to the wave equation are insensitive to "the arrow of time". (In other words, if you can construct things so that time "runs backwards", then these too are perfectly linear solutions to the wave equation.)
Now, how do you make time "run backwards" you ask? Glad you asked!;-) Well, this class of techniques cheats a little bit. They simply record a signal due to some source at multiple locations (kind of like an array of seismometers). Once the signal has been recorded, all the transceivers (in pricipal at least) play the signal backwards, but synchronously. Voila! You've created a wavefield that is approximately the time-reversal of the original wavefield (at least in a Huygen's principal sense; i.e. you are unlikely to have complete spatial coverage with your transceivers). This wavefield will converge back to the site of the original penny drop. It will not be perfectly focused (remember, the field is an approximation), but will still do a fairly decent job.
Slightly perturbing the positions of your transceivers (i.e. much less than a wavelength) will not affect the quality of the approximation too much. The quality suffers more if the playback trasceivers are not time-synchronous (and hence are out-of-phase with each other), since then the coalescing wavefield does not constructively interfere. A time varying medium of propagation (such as an atmosphere causing stars to twinkle) poses its own challenges.
Within the confines of the validity of a linear wave-equation phenomenon, and a static medium of propagation, there is NO CHAOS in this system (no nonlinearities, no insets and outsets of fixed points in phase space, etc. etc.). The extreme sensitivity in terms of position of the receiver that you allude to in your post is simply a phase-rolling thing. It is perfectly linear, and the bread-and-butter of lots of people who do practical things with waves (e.g. acousticians, seismologists, radio engineers, add-someone-who-studies-your-favorite-kind-of-wave -phenomenon here).
Hope this clarifies things somewhat, but feel free to flame away if you like. This is after all slashdot, and I would kinda fell left out if I didn't get flamed for a posting.
Interesting piece, JLSeagull. I hadn't come across anyone who had actually done experiments in this area.
This is some hope, however, in accomplishing this in time-varying scattering/refractive/wave-prop environments (albeit at different ratios of carrier wavelengths to characteristic-length-of-scatterer). The medical ultrasonics people do essentially the same thing to accoustically destroy kidney-stones in-vivo. Matthias Fink, who is quoted towards the tail end of the NS article, is heavily involved in that work (along with his research group), and has published extensively.
(ObSlamDot: My gosh! An actual rational scientific discussion on SlashDot, where actual factual information is exchanged! We can't have that! Quick! Someone post a flame!;-)
Secure??? Who knows. What it should allow with radio is something I've been calling "Space Division Multiple Access". In effect, using scatterers in the environment (e.g. buildings, mountains, what have you) the "cell size" could be brought down to a few tens of meters using the same number of base-station transceivers as currently exist. Who needs more spectrum when you can focus the same bandwidth on multiple physical locations?
BTW, the New Scientist article is talking about kinda old work. NS had a blurb on this back in '97 or so.
Recipe for avoiding the Broadband Blues...
on
Broadband Crackdown
·
· Score: 1
...or at least being able to control your own destiny.
Find a bandwidth wholesaler, who can get more bits than you'll ever need to your residence, without untenable restrictions.
Run around to your neighbors, and convince them that you can provide better reliability, policies, and general service that sucks less than whomever they are using today. (This is probably the hard part). Charge them a fair price. Don't forget to cost out your own time in deciding your pricing structure.
Light it up, and enjoy the surplus bandwidth that is present at your airhead from your wholesaler.
Don't forget to seek your lawyers and tax accountants advice. You'll probably be utterly surprised at the tax games that are legal when you are the business, trying to make a profit.
Congratulations, you've just become your own ISP, and have built at Network Area Network (NAN).
Book publishers say they need a tough law like the Digital Millennium Copyright Act or
they'll never be able to make money selling electronic books.
Gee. I wonder if I can get the U.S. Congress to pass a law that says I have to be able to make a profit, no matter what stupid business I decide to get into.
Slashdot Mirror
I'll be visiting the U.S. soon for 6 weeks or so, and want a cheap(?) connection with national roaming (no phone needed) along the lines listed above.
/.-ers?
Both Cingular's and ATT's (old) websites are un-navigable when looking for something this specialized, in my book.
Any recomendations of deals from
I'm speculating here (I haven't seen XGrid specs yet).
If (and only if) XGrid is one of the Single System Image clustering technologies like opnMosix, then things like MPI jobs are pretty nice to work with on these kinds of clusters.
In openMosix, the load balancing system simply migrates individual MPI processes to nodes that are the least loaded (and which are projected to provide the maximum instantaneous throughput for the job).
Assuming that XGrid has a similar architecture (a big assumption), having a commercial implementation working across Apple's relatively stable hardware platforms (with AltiVec doing vectorized floating point behind the scenes on each node) sounds like a pretty decent cluster implementation to me. Now, as always, Apple must work on cost.
You are correct that only certain kinds of problems "work" well on LAN clusters. But experience shows that a sufficiently motivated computational scientist can adapt many more "big grunt" problems to various weird-and-wonderful architectures than might be apparant at first glance.
Adapting codes to fly on "true" supercomputers (e.g. Big Iron vector machines) is just as much an art form as adapting them to fly on MPI.
YMMV.
Brings a whole new meaning to mobile computing ;-)
Professor Gold is not only the progenitor of the primordial mantle methane, errrr, hypothesis, he is also the progenitor of the idea that the Lunar landings would not have been feasible because the surface dust was not solid enough to support the weight of a lander.
This idea is rumored to have prompted the entire "Ranger" series of spacecraft (at umpteen million dollars) primarily to demonstrate that the Lunar surface was solid.
Oh, and the supporting statements by Steven Soter in the New Scientist article regarding the thermodynamics of photonic momentum-transfer? Three guesses who was Steven Soter's Ph.D. adviser...
Just repeat after me: "Tenure is *good* for science. Tenure is *good* for science. Tenure is *good* for science."
leaf.sf.net
Not dead. Not even comatose.
Yes, code forks suck.
Yes, trying to make ends meet writing free software is no easier than with many other labors of love.
While I personally feel sorry for Cinege, I use the result of his work 24/7. Not a bad legacy...
Accurate information actually modded up on slashdot. Film at eleven.
On the contrary, while the receivers that you spend a hundred bucks on are indeed not accurate enough, GPS based geodesy is a raging success. They use very expensive receivers with multiple frequencies and occupy sites for hours at a time to get the kinds of numbers needed for geodetic measurements. Been going on for years. The major inaccuracies have to do with index-of-refraction effects in the atmosphere (hence the need for multi-frequency instruments).
A quick comment on the toss-away statement in the article that 2.4.20 supports 7505 based systems out of the box.
Be Careful(TM).
The AGP3 stuff requires a patch to stock Marcello/Linus kernels for the 7505 chipset.
I had trouble getting an AGP4x card to work on a Supermicro X5DAL-G board (baby brother to the reviewed X5DA8 board; but at ATX size instead of EATX and able to support unregistered memory) without applying this patch. Once patched, it works fine.
I'm not sure if 7505 support has made it into Marcello's 2.4.21preX series yet, but 7505 support does seem to be in 2.5.6x series (which I'm having trouble getting to boot for unknown reasons).
YMMV.
It has not been covered in ice throughout geologic time.
It's virtually certain there are sedimentary basins for hydrocarbons, and "hard rocks" for metals.
All of this is from geo 1 level knowledge. I have not personally worked on Antarctica, although many geologists have, and their publications should be in the open literature.
If it weren't for the treaty, IMHO, there would already be operating oil fields on or offshore Antarctica.
(And slashdotters are worried about a buldozed ice road doing things to the environment?)
Is this not true? Are there problems at high latitudes? I've never used GPS above about |55| degrees or so...
Ever dropped one and have the screen crack? For at least one vendor (Handspring) it was cheaper for me to get a new one, than to have one "repaired".
Quoting from Tim Peters (the real TimBot, but I digress <wink>) spambayes has a more effective classifier:
While spambayes isn't really Bayesian (anymore; it started out roughly that way, as I understand it), the name stuck
It is a statistical filter, and you do need to train it with your personal collection of spam and ham. However, most of the work in the last month or two on the spambayes list has revolved around building user interfaces, and finding appropriate places to inject the filter into your mail processing, not on improving the classifier.
"Spambayes. Try it. You'll like it!" (But be warned, it's still pre-alpha...)
Will we get Shakespeare? Will it take a million years? Stay tuned, same slash-time, same slash-channel!
is to chant "It'll never work. It'll never work" while rubbing blue mud solemnly in your navel just before inserting the bounced tape into the drive.
...exists to channel Guido.
Lots of things to like about Python, but NumPy is one of the better ones, IMHO. There is a largish and growing community of numerical jocks coalescing around around Python in scientific computation. NumPy makes Python into an "array language" (like Matlab, S-Plus/R, APL, etc. etc.) where the crunching is heavily optimized C code. Links to LAPACK et al., and Fortran wrappers exist in f2py and PyFort, for that old still-running-after-all-these-years code.
/dev/null ;-)
(Just another enthusiast, dabbling in the religious-war du jour. Flamage to
Disney. NBC. Warner Brothers. Universal Studios (near enough, anyway)...
Enough said?
He's representing his constituents, all right. The ones who contribute to his campaign funds.
Any Questions? (This will be on the test.)
...is no longer the exclusive domain of those with a webserver.
Very cool demo.
IMHO, as others in this thread have also pointed out, trying to do this on the net, using router delays etc., is just plain silly! (I wouldn't mind being proven wrong, on this point, however! ;-)
With regard to your points about turbulence, I'll grant that you do have some valid concerns. However, IMHO, the turbulent/time-varying propagation medium is not the heart of the problem, especially when you are talking about something as big as the ocean basins. I think that the scatterers defined by the bathymetry of the ocean basin will be far more dominant (in determining the quality of the final constructive interference) than the relatively minor effects of a time varying wave-speed water column. (The effects of the bathymetry, incidentally, is what the time-reversed acoustics gets you almost for free!) However, that is mere speculation on my part, and the answer is best determined experimentally. That is what Edelman and his group have done.
The time varying prop. medium plays more of role like slightly varying the transceiver positions before playing back the signal. It changes the phase (and possibly the propagation direction) of the local signal. But the signal at the "sweet spot" of the focus is composed of the sum of all the signal paths, and so, on average, the quality of the constructive interference is more sensitive to the "average" phase roll over all paths (modulo the energy leaked out of the system by the differential scattering, of course, but lacking complete spatial coverage with transceivers, you expect energy leakage problems in the first place). If that "average" Feynman-like-path-integral phase roll is time varying, then yes, your concerns about chaotic media are quite valid. On the other hand, if that "average" phase shift is constant, then accomplishing the constructive interference shouldn't be too big of a problem. Which is right? I don't know, and don't know if anyone else knows either. (I do know which way I'd bet, however.) In fact, I kinda suspect that *both* end-members will crop up in different cases (albeit with different probabilities). That's another reason why the experiment is important.
So, bottom line, we agree that there is likely to be turbulence in the propagation medium. I think that won't be too large an issue to accomplish time-reversed acoustical constructive interference at the focus, and Edelman's experiment seems to support that viewpoint.
As to your accusations of my being a pretty big geek, or having a degree in physics, neither is true, actually. Ummmmmm, "I'm not a seismologist, but I play one on TV", OK?
Sorry, but I think you might have a fundamental misunderstanding here. The time-reversed (acoustic/electromagnetic/arbitrary-wave-equation system) phenomena that this technique is seeking to exploit have very little to do with chaos and nonlinear dynamics. They have far, far, more to do with interference in propagating waves. Think of the wave interference tank in your high school physics class. Randomly place a bunch of rocks into the tank; these are the scatterers. Drop a penny into some arbitrary spot. See the pretty interference patterns (at least in your mind's eye) as the waves spread out? Now consider that arbitrary (perfectly linear) solutions to the wave equation are insensitive to "the arrow of time". (In other words, if you can construct things so that time "runs backwards", then these too are perfectly linear solutions to the wave equation.)
Now, how do you make time "run backwards" you ask? Glad you asked! ;-) Well, this class of techniques cheats a little bit. They simply record a signal due to some source at multiple locations (kind of like an array of seismometers). Once the signal has been recorded, all the transceivers (in pricipal at least) play the signal backwards, but synchronously. Voila! You've created a wavefield that is approximately the time-reversal of the original wavefield (at least in a Huygen's principal sense; i.e. you are unlikely to have complete spatial coverage with your transceivers). This wavefield will converge back to the site of the original penny drop. It will not be perfectly focused (remember, the field is an approximation), but will still do a fairly decent job.
Slightly perturbing the positions of your transceivers (i.e. much less than a wavelength) will not affect the quality of the approximation too much. The quality suffers more if the playback trasceivers are not time-synchronous (and hence are out-of-phase with each other), since then the coalescing wavefield does not constructively interfere. A time varying medium of propagation (such as an atmosphere causing stars to twinkle) poses its own challenges.
Within the confines of the validity of a linear wave-equation phenomenon, and a static medium of propagation, there is NO CHAOS in this system (no nonlinearities, no insets and outsets of fixed points in phase space, etc. etc.). The extreme sensitivity in terms of position of the receiver that you allude to in your post is simply a phase-rolling thing. It is perfectly linear, and the bread-and-butter of lots of people who do practical things with waves (e.g. acousticians, seismologists, radio engineers, add-someone-who-studies-your-favorite-kind-of-wave -phenomenon here).
Hope this clarifies things somewhat, but feel free to flame away if you like. This is after all slashdot, and I would kinda fell left out if I didn't get flamed for a posting.
This is some hope, however, in accomplishing this in time-varying scattering/refractive/wave-prop environments (albeit at different ratios of carrier wavelengths to characteristic-length-of-scatterer). The medical ultrasonics people do essentially the same thing to accoustically destroy kidney-stones in-vivo. Matthias Fink, who is quoted towards the tail end of the NS article, is heavily involved in that work (along with his research group), and has published extensively.
(ObSlamDot: My gosh! An actual rational scientific discussion on SlashDot, where actual factual information is exchanged! We can't have that! Quick! Someone post a flame! ;-)
Secure??? Who knows. What it should allow with radio is something I've been calling "Space Division Multiple Access". In effect, using scatterers in the environment (e.g. buildings, mountains, what have you) the "cell size" could be brought down to a few tens of meters using the same number of base-station transceivers as currently exist. Who needs more spectrum when you can focus the same bandwidth on multiple physical locations?
BTW, the New Scientist article is talking about kinda old work. NS had a blurb on this back in '97 or so.
Find a bandwidth wholesaler, who can get more bits than you'll ever need to your residence, without untenable restrictions.
Stick an "airhead" from rooftop/Nokia on your roof.
Run around to your neighbors, and convince them that you can provide better reliability, policies, and general service that sucks less than whomever they are using today. (This is probably the hard part). Charge them a fair price. Don't forget to cost out your own time in deciding your pricing structure.
Light it up, and enjoy the surplus bandwidth that is present at your airhead from your wholesaler.
Don't forget to seek your lawyers and tax accountants advice. You'll probably be utterly surprised at the tax games that are legal when you are the business, trying to make a profit.
Congratulations, you've just become your own ISP, and have built at Network Area Network (NAN).
There, that was easy, wasn't it? ;-)
Book publishers say they need a tough law like the Digital Millennium Copyright Act or
they'll never be able to make money selling electronic books.
Gee. I wonder if I can get the U.S. Congress to pass a law that says I have to be able to make a profit, no matter what stupid business I decide to get into.