> If you want a single configuration tool 'Windows
> Registry Editor' then why not just go to a binary
> configuration file format. It's much easier and
> faster to parse binary data than text, and since
> you have a universal configuration utility it
> really doesn't have to store it in a human
> readable format.
Well, we'd have to write the binary format. Which negates some advantages of XML (prebuilt parsers, established standard). And the canonical reason not to go binary is simple -- what if I'm rescuing my system from that rescue floppy with only vi on it? (And so on.) XML has all the advantages of an arbitrary text format, fewer of the disadvantages, and a few advantages an arbitrary text format doesn't.
XML gains you i/o more structured than the stream, which is a Good Thing. However, that's not directly related to configuration files. What it/does/ do for configuration files it make it possible for there to be a single 'universal' configuration tool. The semantics of 'what does this mean' can be embedded the same way the kernel build system has its help section; that's not a problem. The configuration tool doesn't actually need to understand the semantics. Why XML? It's an open and well-supported standard. There's no need to invent something.
Regarding a standard format for this -- is there a reason we can't clone Apple's? Didn't they already solve this problem for OS X?
You can create 'communicators' in MPI, which are logical groupings of processors that make it easier to program. However, they don't actually have anything to do with the hardware topology. For clusters, this doesn't really matter as much, such most of the idea is to use off-the-shelf (packet-switched) networking; for supercomputers, it could come in very handy.
I suppose I meant 'no explicit emotion programming.' Since we don't know how human emotion arises from the brain and mind, it's reasonable to suggest -- as you did -- that we'll identitify any machine complex enough to self-direct as emotional. Good point.
Stackless Python is supposed to let you wrap up the program and its state, and send it to a friend who can continue where you left of. Sounds alot like process migration in a system like MOSIX, to me. Could Stackless Python be set up as a daemon and auto-migrate based on load?
You ever used OCR? It can pick up multiple-font (and/or damaged) characters at a very high accuracy without any training or 'thinking' at all.
Aside from that, you have a point -- an outstanding question in AI is how to connect the visual system to a symbolic one that will help interpret the visual data.
Incidentally, you don't need emotions to create a course of action, just a lot of horsepower. (Remember Deep Blue?) I'd also doubt you need emotion to create goals; certainly the three laws* would suffice for goal-creation, and don't necessitate emotion.
But in general, the theory isn't there; we're still waiting for the psychologists to develop a hard science.:)
Uh, no. Because of noise in electrical circuits, 64-bit doubles are more precise/have fewer errors than the best available analog circuitry. Do you think the engineers/like/ discretizing their systems to solve them? In the unlikely event that transmission noise is important to the functioning of the human brain, you'll probably come out ahead going digital anyway; analog (or custom digital) circuitry doesn't generally benefit from Moore's law. (Read about the Lisp Machine, et al.) You can always sample the noise from an analog source and add it in where necessary.
Well, how about the IA64? And is it really AMD's fault that Microsoft refuses to support consumer Windows on anything but x86? And, as others have pointed out, the h/w is whatever Intel and AMD feel will run well -- the point of instruction set to abstract the h/w behind it to allow for improvements. Finally, 3dNow!, MMX, and SSE all allow multiple multiplies per clock-cycle.
Fine. So you can tie yourself to COM and the Windows API in a variety of languages. This is a good thing?.NET's VM may be ported to other platforms, but the APIs certainly will not -- though I'm sure game developers would like it if a supported version of DirectX suddenly cropped on Linux.
You seem to have forgotten that.NET isn't native code... so the question becomes, will it run faster than the equivalent Java? And the answer is, probably not. IBM and Sun (and gcj) have an enourmous head start in developing JVMs, and Microsoft is not noted for its slim and speedy coding.
If you give me a copy of the information developed with that supercollider, you still have that copy -- it's a positive-sum game, not a zero-sum game like physical goods. Copyright and patent laws impose an artificial scarcity of information in order to benefit the public (e.g. by making it (more/more easily) profitable to do research), by trying to encourage creation and publication of works, and the disclosure of trade secrets. Is the current copyright and patent system operating for the greater good? I don't tend to think so, but it's something that Congress needs to be convinced of.
The primary problem is that the only way to hide extra digital data is to put it somewhere where the ear can't hear it anyway. This is precisely where the music compression algorithms throw data away. So if a 'watermark' is robust enough to resist analog->digital re-encoding (in the most common case, to 128kbps mp3), it/must/ affect the music quality more than the encoding process. And that means audiophiles won't buy it, and that effective SDMI tracks probably won't sound much better than FM radio.
Incidentally, what makes you think we'll have SDMI encoders lying around to encode pure sine waves with to perform some of these analyseses? Clearly, nobody not in the RIAA ever records music:)
The abbrev. was "LO", if you're still curious. The 4 groups of 99 barbarians in the castle (Harkin's) to the NW -- just before you had to fight Tarjan to get into Kylerean's tower -- were very susceptible to MIBL. Now, if some one could tell me how to make a BT1 IIGS character disk so I can make use of the disk image I have, I'd be in heaven.
I'd just go for the 100-percent-simulated-daylight-view that the army is working on putting into its tanks (and the airforce its planes, etc), where the computer takes non-visible-light data and uses it to 'fill in the gaps' of what's visible. IR headlights are good example, but their data would be forwarded to your sunglasses rather than displayed on the dash. Could add millimeter-wave-radar-on-a-chip (don't have the reference handy), passive IR, nightvision, etc, as you wanted it. Combine with GPS for directions (e.g. the Crazy Taxi arrow), laserpen point-and-query, etc.
Most of this stuff you wouldn't want/need all of the time and would/should be off-loaded to the thing (building, car, airplane, etc) you happen to be in. Think Provost Zakharov for what normal usage would look like.
Solaris is good for the big iron machines that must not go down, and Sun sells those machines -- Solaris is not where they make money. (It's for consistency that their workstations run Solaris...) The Cobalt Networks machines are not big iron, they aren't sparcs, only some of them are even an architecture to which Solaris has been ported (which isn't really 'proven') -- though Sun may be planning to eliminate the MIPS-based boxes.
The only win I here for Sun is that it may allow them to more closely integrate the Cobalt machines with their own...
If you write $5000 down in the place where the bank checks when your check is cashed, you have 5000 'real' dollars. (Incidentally, as any introductory economics class will tell you, money is a collective illusion we all participate in because it's much more conventient than barter.)
`Cheap' is not what I would call several-thousand-dollars a pound. I could see calling several-thousand-dollars a/person/ (a day) cheap, but that's not even within three orders of magnitude of current launch systems. The Shuttle can meet the needs of the ISS only because NASA is willing to throw so much money at it. I assure you that NASA and Congress would be more than happy to cut the cost of the ISS by a third using SSTO or other 10x-cheaper-to-orbit technologies.
The other issue is turn-around time. The space/shuttle/ was intended to launch roughly once a month, but because of design missteps, required a time-consuming external fuel tank and solid rocket boosters. A launch system without external tiles that just needs refueling would be much cheaper to operate because you'd need many fewer of them.
Incidentally, having water ice is very important for space exploration, because it removes the need for about half of your consumables. (Water can be drunk and used for hydroponics; cracked water can be used for fuel and to breathe; raw ice, if there's enough of it, makes a good dust shield, etc.)
In the long-term future, fabrication plants will probably be moved into space, because a sizeable fraction of the expense is enviromental isolation, which is minimized off-planet. This may not necessarily make them cheaper (:)), but they should upgrade more cleanly...
Wouldn't it be easier just to implement bandwidth-based routing, instead of specifying minimum bandwidth in the search? Right now, I think the search goes to everyone and the low-bandwidth answers are filtered out. If a node were not to forward searches for which nothing can be returned anyway, you can avoid wasting time and bandwidth. All it requires is that the high-bandwdth servers be well-connected, which could be induced by allowing preferences for higher-bandwidth links...
Thank you! If you're going to college and learning to program, you/are/ wasting your time and your money. Learning computer science is a completely different and more intellectually rewarding activity. (For me, anyway. YMMV.)
The contrast between computer science and programming is alot like the contrast between physics and engineering -- and engineer only needs to know enough physics to get his (her) job done, and, as many of the anecdotes here indicate, the better ones can teach themselves the new theory they need. The major difference is that for C/S and programming, the output of the experimentation by the scientists is nearly indistinguishable from the output of the engineers. That is, both 'build bridges.'
Just an aside, "the greater metro washington DC area handles half of the internet's traffic." (Forgot where I heard that, sorry.) Probably this is because of AOL (Herndon, VA), but it's still an interesting statistic...
The Alpha is more powerful than Pentium III, but not by as much as you'd might think; most of its advantage for supercomputing comes from its superior bus. Similarly, the Alpha and the Athlon are suprisingly well matched, as the Athlon uses the Alpha bus; but, of course, the Alpha is designed as supercomputing equipment. The POWER3 and POWER4 chips are less powerful than you might suspect, probably around or below an Alpha -- IBM's machines run so quickly because they so bloody many of them hooked together which such absurdly wide/fast busses. Of course, when the 21364 rolls into town, it should blow the doors off the T-bird; but but the SledgeHammer/Williamette may be ready by then.
The reason (Beowulf) clusters have become common in recent years is that only recently have the 'desktop' chips started run fast enough to compete against the generally more powerful chips from the big iron companies. It's like the G4/Pentium situation (a gHz G4 would spank a gHz PIII, but no such beast exists), except even more so.
Slashdot Mirror
> If you want a single configuration tool 'Windows
> Registry Editor' then why not just go to a binary
> configuration file format. It's much easier and
> faster to parse binary data than text, and since
> you have a universal configuration utility it
> really doesn't have to store it in a human
> readable format.
Well, we'd have to write the binary format. Which negates some advantages of XML (prebuilt parsers, established standard). And the canonical reason not to go binary is simple -- what if I'm rescuing my system from that rescue floppy with only vi on it? (And so on.) XML has all the advantages of an arbitrary text format, fewer of the disadvantages, and a few advantages an arbitrary text format doesn't.
-_Quinn
XML gains you i/o more structured than the stream, which is a Good Thing. However, that's not directly related to configuration files. What it /does/ do for configuration files it make it possible for there to be a single 'universal' configuration tool. The semantics of 'what does this mean' can be embedded the same way the kernel build system has its help section; that's not a problem. The configuration tool doesn't actually need to understand the semantics. Why XML? It's an open and well-supported standard. There's no need to invent something.
Regarding a standard format for this -- is there a reason we can't clone Apple's? Didn't they already solve this problem for OS X?
-_Quinn
You can create 'communicators' in MPI, which are logical groupings of processors that make it easier to program. However, they don't actually have anything to do with the hardware topology. For clusters, this doesn't really matter as much, such most of the idea is to use off-the-shelf (packet-switched) networking; for supercomputers, it could come in very handy.
-_Quinn
I suppose I meant 'no explicit emotion programming.' Since we don't know how human emotion arises from the brain and mind, it's reasonable to suggest -- as you did -- that we'll identitify any machine complex enough to self-direct as emotional. Good point.
-_Quinn
Stackless Python is supposed to let you wrap up the program and its state, and send it to a friend who can continue where you left of. Sounds alot like process migration in a system like MOSIX, to me. Could Stackless Python be set up as a daemon and auto-migrate based on load?
-_Quinn
You ever used OCR? It can pick up multiple-font (and/or damaged) characters at a very high accuracy without any training or 'thinking' at all.
Aside from that, you have a point -- an outstanding question in AI is how to connect the visual system to a symbolic one that will help interpret the visual data.
Incidentally, you don't need emotions to create a course of action, just a lot of horsepower. (Remember Deep Blue?) I'd also doubt you need emotion to create goals; certainly the three laws* would suffice for goal-creation, and don't necessitate emotion.
But in general, the theory isn't there; we're still waiting for the psychologists to develop a hard science.
-_Quinn
Prove to me you're not a chinese room.
-_Quinn
Uh, no. Because of noise in electrical circuits, 64-bit doubles are more precise/have fewer errors than the best available analog circuitry. Do you think the engineers /like/ discretizing their systems to solve them? In the unlikely event that transmission noise is important to the functioning of the human brain, you'll probably come out ahead going digital anyway; analog (or custom digital) circuitry doesn't generally benefit from Moore's law. (Read about the Lisp Machine, et al.) You can always sample the noise from an analog source and add it in where necessary.
-_Quinn
Well, how about the IA64? And is it really AMD's fault that Microsoft refuses to support consumer Windows on anything but x86? And, as others have pointed out, the h/w is whatever Intel and AMD feel will run well -- the point of instruction set to abstract the h/w behind it to allow for improvements. Finally, 3dNow!, MMX, and SSE all allow multiple multiplies per clock-cycle.
-_Quinn
Fine. So you can tie yourself to COM and the Windows API in a variety of languages. This is a good thing? .NET's VM may be ported to other platforms, but the APIs certainly will not -- though I'm sure game developers would like it if a supported version of DirectX suddenly cropped on Linux.
-_Quinn
You seem to have forgotten that .NET isn't native code... so the question becomes, will it run faster than the equivalent Java? And the answer is, probably not. IBM and Sun (and gcj) have an enourmous head start in developing JVMs, and Microsoft is not noted for its slim and speedy coding.
-_Quinn
If you give me a copy of the information developed with that supercollider, you still have that copy -- it's a positive-sum game, not a zero-sum game like physical goods. Copyright and patent laws impose an artificial scarcity of information in order to benefit the public (e.g. by making it (more/more easily) profitable to do research), by trying to encourage creation and publication of works, and the disclosure of trade secrets. Is the current copyright and patent system operating for the greater good? I don't tend to think so, but it's something that Congress needs to be convinced of.
-_Quinn
The primary problem is that the only way to hide extra digital data is to put it somewhere where the ear can't hear it anyway. This is precisely where the music compression algorithms throw data away. So if a 'watermark' is robust enough to resist analog->digital re-encoding (in the most common case, to 128kbps mp3), it /must/ affect the music quality more than the encoding process. And that means audiophiles won't buy it, and that effective SDMI tracks probably won't sound much better than FM radio.
:)
Incidentally, what makes you think we'll have SDMI encoders lying around to encode pure sine waves with to perform some of these analyseses? Clearly, nobody not in the RIAA ever records music
-_Quinn
The abbrev. was "LO", if you're still curious. The 4 groups of 99 barbarians in the castle (Harkin's) to the NW -- just before you had to fight Tarjan to get into Kylerean's tower -- were very susceptible to MIBL. Now, if some one could tell me how to make a BT1 IIGS character disk so I can make use of the disk image I have, I'd be in heaven.
-_Quinn
From the story:
1) Normal mode. This is SBE (scan-before-execute) controlled. Most code is run natively, some instructions are virtualized and thus emulated.
-_Quinn
I'd just go for the 100-percent-simulated-daylight-view that the army is working on putting into its tanks (and the airforce its planes, etc), where the computer takes non-visible-light data and uses it to 'fill in the gaps' of what's visible. IR headlights are good example, but their data would be forwarded to your sunglasses rather than displayed on the dash. Could add millimeter-wave-radar-on-a-chip (don't have the reference handy), passive IR, nightvision, etc, as you wanted it. Combine with GPS for directions (e.g. the Crazy Taxi arrow), laserpen point-and-query, etc.
Most of this stuff you wouldn't want/need all of the time and would/should be off-loaded to the thing (building, car, airplane, etc) you happen to be in. Think Provost Zakharov for what normal usage would look like.
-_Quinn
Solaris is good for the big iron machines that must not go down, and Sun sells those machines -- Solaris is not where they make money. (It's for consistency that their workstations run Solaris...) The Cobalt Networks machines are not big iron, they aren't sparcs, only some of them are even an architecture to which Solaris has been ported (which isn't really 'proven') -- though Sun may be planning to eliminate the MIPS-based boxes.
The only win I here for Sun is that it may allow them to more closely integrate the Cobalt machines with their own...
-_Quinn
If you write $5000 down in the place where the bank checks when your check is cashed, you have 5000 'real' dollars. (Incidentally, as any introductory economics class will tell you, money is a collective illusion we all participate in because it's much more conventient than barter.)
-_Quinn
`Cheap' is not what I would call several-thousand-dollars a pound. I could see calling several-thousand-dollars a /person/ (a day) cheap, but that's not even within three orders of magnitude of current launch systems. The Shuttle can meet the needs of the ISS only because NASA is willing to throw so much money at it. I assure you that NASA and Congress would be more than happy to cut the cost of the ISS by a third using SSTO or other 10x-cheaper-to-orbit technologies.
/shuttle/ was intended to launch roughly once a month, but because of design missteps, required a time-consuming external fuel tank and solid rocket boosters. A launch system without external tiles that just needs refueling would be much cheaper to operate because you'd need many fewer of them.
The other issue is turn-around time. The space
Incidentally, having water ice is very important for space exploration, because it removes the need for about half of your consumables. (Water can be drunk and used for hydroponics; cracked water can be used for fuel and to breathe; raw ice, if there's enough of it, makes a good dust shield, etc.)
-_Quinn
In the long-term future, fabrication plants will probably be moved into space, because a sizeable fraction of the expense is enviromental isolation, which is minimized off-planet. This may not necessarily make them cheaper (:)), but they should upgrade more cleanly...
-_Quinn
Wouldn't it be easier just to implement bandwidth-based routing, instead of specifying minimum bandwidth in the search? Right now, I think the search goes to everyone and the low-bandwidth answers are filtered out. If a node were not to forward searches for which nothing can be returned anyway, you can avoid wasting time and bandwidth. All it requires is that the high-bandwdth servers be well-connected, which could be induced by allowing preferences for higher-bandwidth links...
-_Quinn
Thank you! If you're going to college and learning to program, you /are/ wasting your time and your money. Learning computer science is a completely different and more intellectually rewarding activity. (For me, anyway. YMMV.)
The contrast between computer science and programming is alot like the contrast between physics and engineering -- and engineer only needs to know enough physics to get his (her) job done, and, as many of the anecdotes here indicate, the better ones can teach themselves the new theory they need. The major difference is that for C/S and programming, the output of the experimentation by the scientists is nearly indistinguishable from the output of the engineers. That is, both 'build bridges.'
-_Quinn
Just an aside, "the greater metro washington DC area handles half of the internet's traffic." (Forgot where I heard that, sorry.) Probably this is because of AOL (Herndon, VA), but it's still an interesting statistic...
-_Quinn
The Alpha is more powerful than Pentium III, but not by as much as you'd might think; most of its advantage for supercomputing comes from its superior bus. Similarly, the Alpha and the Athlon are suprisingly well matched, as the Athlon uses the Alpha bus; but, of course, the Alpha is designed as supercomputing equipment. The POWER3 and POWER4 chips are less powerful than you might suspect, probably around or below an Alpha -- IBM's machines run so quickly because they so bloody many of them hooked together which such absurdly wide/fast busses. Of course, when the 21364 rolls into town, it should blow the doors off the T-bird; but but the SledgeHammer/Williamette may be ready by then.
The reason (Beowulf) clusters have become common in recent years is that only recently have the 'desktop' chips started run fast enough to compete against the generally more powerful chips from the big iron companies. It's like the G4/Pentium situation (a gHz G4 would spank a gHz PIII, but no such beast exists), except even more so.
-_Quinn
Maybe they were handling events? Why would you keep a hard reference to an event handler lying around?
-_Quinn