When the canadian postal services discovered e-mail. The offered a free encrypted e-mail account to every canadian....but you can only send e-mail to someone with a Canadian Post e-mail account... kind of useless. It's been at least a year, and I haven't heard of that since then. I think it probably dead and buried.
The principal behind this method of reducing drag in a fluid environment is called "laminar flow".
I know what laminar flow is. However, The flow from such a submarine would be turbulent, not laminar. Let's look at Reynolds number:
R = L*V*rho/mu = 10 m * 300 m/s * 1000 kg/m3 /.01 P = 3,4 x 10^9. IIRC laminar flow ends around 1000, the this flow is turbulent. An object moving at that speed in air will cause a lot of turbulence. Also, the ration between the density and viscosity is higher in water that air, thus at equal speed the Reynods in water will be higher.
As for the "pushing water aside" analogy, you can take Bernouilli's equation with F =.5 * K * rho * v^2 and you will end up with the exact same result for the power (with K=1).
Now, instead of calling names, I'd be very interrested in having a *technical* discussion with you.
A 20,000 HP whale would be cool, but I'd imagine it could also do better than 30 knots. Unfortunately, your calculations need some basis in reality before they make any sort of reasonable approximation (even with a fudge factor of 100x).
Well if you take that "huge" fudge factor, that would mean that your record-braking blue whale has a 200 HP power. This is the same as a sports car. I think a blue whale can be much more powerful than that. Hey, a blue whale is probably thousands of times more powerful than a human being (it can weight 200 tons, ~2000 times more than a human being). It's really really strong!
Also, consider how much power it takes to bring a plane to supersonic speeds and try to imagine that water is about a thousand times heavier than air. It might work for a small torpedo, but not for a submarine.
So I still maintain that my calculations are correct within my 100 fudge factor... So instead of arguing that you don't like the final number, it'd be very pleased if you can point specific errors I made.
2.5 GW is probably achievable with a rocket propulsion over a very short period of time, such as the one required for a torpedo (in ten seconds it can to 3 km). However, this is different for a submarine, which needs to be able to cruise at that speed. The submarine will have at least 100 times the drag and will need to keep that speed for at least several minutes, if not hours to be useful. In terms of total energy, there's about a factor of 1000 between the torpedo and the submarine.
Having done some fluid dynamics before, I know that it's far from being that simple. What I'm trying to do is find an order of magnitude. But the fact remains: the Bernouilli equation tells you that drag is proportionnal to the square of the speed. Also, the power is the drag times the speed, so the power is proportionnal to the third power of the speed. It's *roughly* the same as for airplanes, except that the density of water is more than 1000 times that of air (at high altitude).
As for getting back energy, it won't happen, since the back of the submarine is in low-pressure vapour, this actually adds to the drag.
While a agree that water doesn't have to be pushed at 340 m/s (I'm only roughly approximating), it still has to be pushed aside quite fast in order to produce supercavitation. Remember that the nose is flat, which is not the most hydrodynamic shape. Even if you consider a drag coefficient of.2 (it's probably above that), you get 50 GW power to drive the sub.
Well, if you consider the problem of bringing a load of 50 tons to 500 km of altitude (ie into orbit) in only one second, you get: P = 9.8 m/s2 * 50,000 kg * 500,000 m / 1 s = 250 GW.
This means that the rockets of the submarine would be enough to bring the said submarine into orbit in only 1 second! I doubt such a rocket exists.
Consider a "small" submarine with a radius of 2m, going at a speed of 340 m/s (Mach 1 at sea level). The amount of water this submarine has to push aside is pi*(2m)^2*340m/s = 4300 m3/s
That's 4300 tons of water per seconds. Now this water has to be pushed at around 340 m/s too, which corresponds to a kinetic energy of 58000 kJ/ton of water. Combining the two results gives us 250 GW of power required to move our submarine.
This calculation is very approximate, but it still gives an order of magnitude. Even if I'm 100 times over, it still means thousands of megawatts, the power of a big nuclear plant. This is why I doubt we'll see a supersonic submarine soon.
Having a speech recognition toolbox is only one part of the problem. As many people in the domain (I used to work in speech recognition) will tell you is that sometimes, the key to a good speech recognition engine is not in the code, but in the speech data used to train it. Speech databases are very expensive and speech recognition companies usually have a lot of "proprietary" databases.
One project which addresses the problem is the Open Mind Initiative, and more specifically the Open Mind Speech Recognition project, for which I am the coordinator. Our goal is to collect data from people on the internet and make that data available to people working on speech recognition with a GPL-like license. I think this is the key to having OSS speech recognition engines perform as well as the proprietary ones. The project is not very advanced yet, but any help would be really welcomed.
This is not the first time this happens. I have seen this story a couple hours ago (around 1400 EST?). It disappeared from slashdot for several hours and then reappear with a new time on it. This is not the first time this happens. Does anyone know what this means?
I think it's a different kind of strategy. The strategy is that they simply cannot produce even 1 GHz P3's, but they don't want to look like they're behind AMD in terms of clock speed.
Is this going to be the same quantity as the P3 1 GHz. So far, it still isn't possible to get those. This seems more like "marketware" than anything useful.
The ITU just agreed on the V.94 Standard. This a great improvement on the older V.92, which allow only 56 kbps of download, while V.94 allows 58 kbps. That's a huge improvement of 2 kbps, allowing 1 free Meg of download every hour!
My point is: who needs those improvements? 14.4 to 28.8 gave you a factor of two. V.90 to V.92 gives you almost nothing (add teh fact that the line noise will likely eliminate all this gain). It's like upgrading from a 700 MHz CPU to a 750 MHz. Except for marketing, I really don't see the idea.
I know this opinion isn't popular, but they do make sense. If you call a store and order something by phone, you have to pay the tax. Why shouldn't you pay a tax when you do the same over the web. Do you imagine is, in the 30's they would have said that anyone who takes his car to shop doesn't have to pay a tax. It just doesn't make sense. Either you have a tax or you don't, but if you do, everyone needs to pay it.
Yes, LAME is free from all copyright. However, the real problem with Frauenhoffer is patents. They claim that EVERY mp3 encoder infringes their patent. I don't know for sure the state of the patent issue with LAME, but for sure Frauenhoffer will try to cause them as much trouble as possible.
From what I heard, people are backing away from retina scan. Though it is a very good identification method, it has an evil side effect: Your retina can tell a lot about your health. The problem is thus not reliability, but privary issues. You don't want retina scan as an identification when signing up for a life insurance!
ISP's are very pleased about that. People connect to regular ISP's and the government pays them back. The only people eligigle, of course, are the ones who already recieve "allocations familiales".
Sorry to disapoint you, but it has been happening in a more developed country, just a bit up north, in Canada. There is a program in the province of Quebec that helps families get connected by paying most (if not all) the internet fees for one or two years (don't remember). They also pay a part of the computer. I hope the rest of Canada follows.
results suddenly change about, which happens to be where the array size starts to exceed the Athlon's 64KB level-1 data cache
AFAIK, the double values (and probably the float) aren't cached in the L1 on the Athlon.
Also, I think I've found another problem with your FFT: the use of the "sin" function instead of tables. This function is very slow, and it's performance depends a lot on the math library implementation. It is possible that the JVM implementation of the sine was faster.
There are some constructs in C that are almost impossible to optimize. Pointer arithmetic is an example. When the compiler cannot tell if two pointers point at the same place, it has to be very careful during optimization. Also, here's an example of what I mean by loop unrolling having an effect: consider these two dot product implementations:
/* This one is not optimized and looks like the one for the FFT */ double prod(double *x, double *y, int len) { double sum=0; int i; for (i=0;ilen;i++) sum += x[i]*y[i]; return sum; }
/*This function in an optimized version of the previous */ double prod(double *x, double *y, int len) { double sum1=0, sum2=0, sum3=0, sum4=0; double *end = x + len /*this loop is unrolled by 4*/ while (x end-3) { sum1 += *x++ * *y++; sum2 += *x++ * *y++; sum3 += *x++ * *y++; sum4 += *x++ * *y++; } /*this loop computes the remaining (non multiple of four) */ while (xend) sum1 += *x++ * *y++; return sum1+sum2+sum3+sum4; }
I have recently used this optimization on my code and found a performance increase of about a factor of 3 (on a Athlon 500). The fist version of the function has three problems: 1) The loop overhead is very expensive compared to the 2 float operations inside it. 2) The indexing is ofter more expensive than simple pointer increment (thought not always). 3) This one is the most important. In the first example, each sum (+=) requires the previous result of the sum to compute. Now, the problem is that the FP ADD pipeline is stalled. The time it takes for each addition is no more one cycle, but the length of the pipeline. In the second example, the use of multiple partial sums prevent that.
Also, as I said before, a good FFT coded in C can be as fast as the processor is. The Java code can be as fast if it is good, but not twice faster, as in the benchmarks. But because, the FFT code wasn't optimized, the performance difference likely came from the loop overhead.
I'd be curious about the C++ allocation. I think some STL classes have a special allocator that's supposed to be faster than malloc (is new using the same allocator as malloc?).
I can't comment about the other tests, but I've looked at the FFT code and can say it is very badly optimized. There are some things a C compiler can't optimized because of aliasing, but a Java compiler can. There are ways to code these kinds of things so it can work, for example doing explicit loop unrolling. In the FFT code he had, the FPU pipeline would always be stalled, because lots of loops only have 4 multiplications and 2 additions.
What makes me even more suspicious is that I have a K7-500 too and I have done some tests with a heavily optimized FFT (fftw) and I get a performance around 400 mflops. There's just no way a JVM can be 220% faster than that. So my comclusion is "with poor code and poor optimization, Java can be faster than C".
I don't want to take position of the whole Java vs C speed, but what I'm saying is that at least his FFT test is flawed.
I guess the problem with all this is that a file named Metallica isn't necessary a Metallica song. If the software downloaded the data and actually checked it, I'd feel better about it.
Well, this may be part of the "solution" against it. If people start sharing empty files with Metallica as a name, it makes finding the real Metallica songs harder. All there is to do is the same as e-mail: files like "Metallica - enpty - One.mpg" for the false files and "Metallica - One.mp3" for the real one... just my $.02.
Well it look I'm not the only one with these problems. Except that I actually had at least one real kernel crash (and countless X crashes, some of them requiring reboots) with the NVidia XFree4 driver.
Slashdot Mirror
When the canadian postal services discovered e-mail. The offered a free encrypted e-mail account to every canadian. ...but you can only send e-mail to someone with a Canadian Post e-mail account... kind of useless. It's been at least a year, and I haven't heard of that since then. I think it probably dead and buried.
The principal behind this method of reducing drag in a fluid environment is called "laminar flow".
.01 P = 3,4 x 10^9. IIRC laminar flow ends around 1000, the this flow is turbulent. An object moving at that speed in air will cause a lot of turbulence. Also, the ration between the density and viscosity is higher in water that air, thus at equal speed the Reynods in water will be higher.
.5 * K * rho * v^2 and you will end up with the exact same result for the power (with K=1).
I know what laminar flow is. However, The flow from such a submarine would be turbulent, not laminar. Let's look at Reynolds number:
R = L*V*rho/mu = 10 m * 300 m/s * 1000 kg/m3 /
As for the "pushing water aside" analogy, you can take Bernouilli's equation with F =
Now, instead of calling names, I'd be very interrested in having a *technical* discussion with you.
A 20,000 HP whale would be cool, but I'd imagine it could also do better than 30 knots. Unfortunately, your calculations need some basis in reality before they make any sort of reasonable approximation (even with a fudge factor of 100x).
Well if you take that "huge" fudge factor, that would mean that your record-braking blue whale has a 200 HP power. This is the same as a sports car. I think a blue whale can be much more powerful than that. Hey, a blue whale is probably thousands of times more powerful than a human being (it can weight 200 tons, ~2000 times more than a human being). It's really really strong!
Also, consider how much power it takes to bring a plane to supersonic speeds and try to imagine that water is about a thousand times heavier than air. It might work for a small torpedo, but not for a submarine.
So I still maintain that my calculations are correct within my 100 fudge factor... So instead of arguing that you don't like the final number, it'd be very pleased if you can point specific errors I made.
2.5 GW is probably achievable with a rocket propulsion over a very short period of time, such as the one required for a torpedo (in ten seconds it can to 3 km). However, this is different for a submarine, which needs to be able to cruise at that speed. The submarine will have at least 100 times the drag and will need to keep that speed for at least several minutes, if not hours to be useful. In terms of total energy, there's about a factor of 1000 between the torpedo and the submarine.
Having done some fluid dynamics before, I know that it's far from being that simple. What I'm trying to do is find an order of magnitude. But the fact remains: the Bernouilli equation tells you that drag is proportionnal to the square of the speed. Also, the power is the drag times the speed, so the power is proportionnal to the third power of the speed. It's *roughly* the same as for airplanes, except that the density of water is more than 1000 times that of air (at high altitude).
As for getting back energy, it won't happen, since the back of the submarine is in low-pressure vapour, this actually adds to the drag.
While a agree that water doesn't have to be pushed at 340 m/s (I'm only roughly approximating), it still has to be pushed aside quite fast in order to produce supercavitation. Remember that the nose is flat, which is not the most hydrodynamic shape. Even if you consider a drag coefficient of .2 (it's probably above that), you get 50 GW power to drive the sub.
Well, if you consider the problem of bringing a load of 50 tons to 500 km of altitude (ie into orbit) in only one second, you get:
P = 9.8 m/s2 * 50,000 kg * 500,000 m / 1 s = 250 GW.
This means that the rockets of the submarine would be enough to bring the said submarine into orbit in only 1 second! I doubt such a rocket exists.
Consider a "small" submarine with a radius of 2m, going at a speed of 340 m/s (Mach 1 at sea level). The amount of water this submarine has to push aside is pi*(2m)^2*340m/s = 4300 m3/s
That's 4300 tons of water per seconds. Now this water has to be pushed at around 340 m/s too, which corresponds to a kinetic energy of 58000 kJ/ton of water. Combining the two results gives us 250 GW of power required to move our submarine.
This calculation is very approximate, but it still gives an order of magnitude. Even if I'm 100 times over, it still means thousands of megawatts, the power of a big nuclear plant. This is why I doubt we'll see a supersonic submarine soon.
Having a speech recognition toolbox is only one part of the problem. As many people in the domain (I used to work in speech recognition) will tell you is that sometimes, the key to a good speech recognition engine is not in the code, but in the speech data used to train it. Speech databases are very expensive and speech recognition companies usually have a lot of "proprietary" databases.
One project which addresses the problem is the Open Mind Initiative, and more specifically the Open Mind Speech Recognition project, for which I am the coordinator. Our goal is to collect data from people on the internet and make that data available to people working on speech recognition with a GPL-like license. I think this is the key to having OSS speech recognition engines perform as well as the proprietary ones. The project is not very advanced yet, but any help would be really welcomed.
This is not the first time this happens. I have seen this story a couple hours ago (around 1400 EST?). It disappeared from slashdot for several hours and then reappear with a new time on it. This is not the first time this happens. Does anyone know what this means?
I think it's a different kind of strategy. The strategy is that they simply cannot produce even 1 GHz P3's, but they don't want to look like they're behind AMD in terms of clock speed.
Is this going to be the same quantity as the P3 1 GHz. So far, it still isn't possible to get those. This seems more like "marketware" than anything useful.
The ITU just agreed on the V.94 Standard. This a great improvement on the older V.92, which allow only 56 kbps of download, while V.94 allows 58 kbps. That's a huge improvement of 2 kbps, allowing 1 free Meg of download every hour!
My point is: who needs those improvements? 14.4 to 28.8 gave you a factor of two. V.90 to V.92 gives you almost nothing (add teh fact that the line noise will likely eliminate all this gain). It's like upgrading from a 700 MHz CPU to a 750 MHz. Except for marketing, I really don't see the idea.
I know this opinion isn't popular, but they do make sense. If you call a store and order something by phone, you have to pay the tax. Why shouldn't you pay a tax when you do the same over the web. Do you imagine is, in the 30's they would have said that anyone who takes his car to shop doesn't have to pay a tax. It just doesn't make sense. Either you have a tax or you don't, but if you do, everyone needs to pay it.
Yes, LAME is free from all copyright. However, the real problem with Frauenhoffer is patents. They claim that EVERY mp3 encoder infringes their patent. I don't know for sure the state of the patent issue with LAME, but for sure Frauenhoffer will try to cause them as much trouble as possible.
From what I heard, people are backing away from retina scan. Though it is a very good identification method, it has an evil side effect: Your retina can tell a lot about your health. The problem is thus not reliability, but privary issues. You don't want retina scan as an identification when signing up for a life insurance!
ISP's are very pleased about that. People connect to regular ISP's and the government pays them back. The only people eligigle, of course, are the ones who already recieve "allocations familiales".
Sorry to disapoint you, but it has been happening in a more developed country, just a bit up north, in Canada. There is a program in the province of Quebec that helps families get connected by paying most (if not all) the internet fees for one or two years (don't remember). They also pay a part of the computer. I hope the rest of Canada follows.
results suddenly change about, which happens to be where the array size starts to exceed the Athlon's 64KB level-1 data cache
AFAIK, the double values (and probably the float) aren't cached in the L1 on the Athlon.
Also, I think I've found another problem with your FFT: the use of the "sin" function instead of tables. This function is very slow, and it's performance depends a lot on the math library implementation. It is possible that the JVM implementation of the sine was faster.
...Oops, the post screwed up all the less-than signs in the loops... but it should still be obvious what's missing.
There are some constructs in C that are almost impossible to optimize. Pointer arithmetic is an example. When the compiler cannot tell if two pointers point at the same place, it has to be very careful during optimization. Also, here's an example of what I mean by loop unrolling having an effect: consider these two dot product implementations:
/*this loop is unrolled by 4*/
/*this loop computes the remaining (non multiple of four) */
/* This one is not optimized and looks like the one for the FFT */
double prod(double *x, double *y, int len)
{
double sum=0;
int i;
for (i=0;ilen;i++)
sum += x[i]*y[i];
return sum;
}
/*This function in an optimized version of the previous */
double prod(double *x, double *y, int len)
{
double sum1=0, sum2=0, sum3=0, sum4=0;
double *end = x + len
while (x end-3)
{
sum1 += *x++ * *y++;
sum2 += *x++ * *y++;
sum3 += *x++ * *y++;
sum4 += *x++ * *y++;
}
while (xend)
sum1 += *x++ * *y++;
return sum1+sum2+sum3+sum4;
}
I have recently used this optimization on my code and found a performance increase of about a factor of 3 (on a Athlon 500). The fist version of the function has three problems:
1) The loop overhead is very expensive compared to the 2 float operations inside it.
2) The indexing is ofter more expensive than simple pointer increment (thought not always).
3) This one is the most important. In the first example, each sum (+=) requires the previous result of the sum to compute. Now, the problem is that the FP ADD pipeline is stalled. The time it takes for each addition is no more one cycle, but the length of the pipeline. In the second example, the use of multiple partial sums prevent that.
Also, as I said before, a good FFT coded in C can be as fast as the processor is. The Java code can be as fast if it is good, but not twice faster, as in the benchmarks. But because, the FFT code wasn't optimized, the performance difference likely came from the loop overhead.
I'd be curious about the C++ allocation. I think some STL classes have a special allocator that's supposed to be faster than malloc (is new using the same allocator as malloc?).
I can't comment about the other tests, but I've looked at the FFT code and can say it is very badly optimized. There are some things a C compiler can't optimized because of aliasing, but a Java compiler can. There are ways to code these kinds of things so it can work, for example doing explicit loop unrolling. In the FFT code he had, the FPU pipeline would always be stalled, because lots of loops only have 4 multiplications and 2 additions.
What makes me even more suspicious is that I have a K7-500 too and I have done some tests with a heavily optimized FFT (fftw) and I get a performance around 400 mflops. There's just no way a JVM can be 220% faster than that. So my comclusion is "with poor code and poor optimization, Java can be faster than C".
I don't want to take position of the whole Java vs C speed, but what I'm saying is that at least his FFT test is flawed.
I guess the problem with all this is that a file named Metallica isn't necessary a Metallica song. If the software downloaded the data and actually checked it, I'd feel better about it.
Well, this may be part of the "solution" against it. If people start sharing empty files with Metallica as a name, it makes finding the real Metallica songs harder. All there is to do is the same as e-mail: files like "Metallica - enpty - One.mpg" for the false files and "Metallica - One.mp3" for the real one... just my $.02.
Well it look I'm not the only one with these problems. Except that I actually had at least one real kernel crash (and countless X crashes, some of them requiring reboots) with the NVidia XFree4 driver.