Another good way to pick up HEU is from "broken arrows" -- lost nuclear devices from bomber crashes or downed submarines. But Plutonium is vastly easier to obtain than HEU, since you can produce it from natural Uranium and fast neutrons, and chemically separate it from the load. If one had about a quarter of a million dollars to spend on a small hydro plant, one could produce more than a critical mass of Pu annually, very low tech.
You seem to be restricting the discussion to HEU bombs. Plutonium is relatively quite easy to refine, and easy to produce from natural Uranium, given a source of energetic neutrons. I'm very keen to see the people get weapons which place their power on par with that of their ruling classes. This might enable a democratic society to re-emerge in the future. If I had a patron, or became wealthy, I would definitely produce some devices for test, if an appropriate democratic ruling structure could be relied upon, with checks and balances, to protect against immoral use by madmen.
I beg to differ about inherent scalability limits. Perhaps relative to a given application, such as one which requires random routing superlinear in node count, but for the broad range of applications, for which logarithmic cross-section bandwidth growth is sufficient, there are well-understood architectures that scale without bound.
This stuff was cute and clever in 1992, but these days it's a pretty well-trodden path.
That is consistent with my experience. Minor projects with one or two or three developers will often be a morass, but when an OSS project scales to perhaps 5 or 6 developers, and dozens of patch contributors, it has to have it's ducks in a row, or it will get forked by someone more competent. Documentation is less reliably good than is design. "Architects" who puff up unimplementable dream designs don't exist in the open source world.
> 70%/50% think attacks on the U.S. forces are justified? That smells of high BS.
Really? I'm surprised that it isn't higher. I mean, suppose France invaded the U.S., imposed the Napoleonic code and parliamentary government, outlawed the Republican party, and imprisoned George Bush for war crimes, after killing 2.5% of the population (7.325 million people) and bombing NYC and Chicago into rubble. How many French people would consider U.S. resistance attacks on legionaires to be justified?
I agree with everything you said, but would note that by adding PCI RAID IDE controllers, it scales to the available case cooling and mount points. I'd probably add some mounting brackets from erector set pieces and a few fans (and maybe some acoustic insulation to my closet!!) if I wanted to scale up to, say 12x400. But that really is the wall for my personal taste, 3.6TB of RAID 5. If I ran out of erector set pieces, I'd just pick up an old gutted tower server case for the job from ebay or axe-man.
I'd peg the 3.6TB solution at 550 for the platform, including 3 promise tx2000 cards, and just under $4k for the drives. That's $1.25 per GB, compared to $0.80 per GB for the 4x250 model. Personally, I'd rather keep multiple servers, say one for anime and music, one for data files, one for movies... With network cross-mounting, it all looks like one big directory tree anyhow. Of course the middle road, more drives at the lower $/GB 250GB, $113 sweet spot, still lets you scale to 2.25GB, and keep it close to $0.80/GB. And it is easier to upgrade incrementally, by moving one RAID set at a time to large physical drives, since shuffling data between controllers is a lot faster than pumping it over a LAN connection.
I find it easier, faster, and vastly cheaper, to stick lots of drives and controllers and fans in a big tower than to buy boutique hardware. I also find it more reliable, since my personal budget allows for plenty of spare hardware at the lower pricepoint. If you really want to run multiple RAID controllers in a pc case, though, remember to pick a mondo power supply. you'll probably need a bunch of power Y connectors too, unless you're using all SATA drives.
Take a generic low-end PC (say, sempron 2200, with a pc-chips or ecs motherboard, 128MB DDR -- it's all overkill at this point) in a tall, well-ventilated case with at least 300W power, for about $150, and add 4x 250GB IDE drives, for about $450. Run as raid 5 on a debian sarge install. That gives you 750GB of online storage suitable for streaming media for about $600. When 400GB drives fall to the same $0.40/MB price point, make a new box and sell the old one on ebay, to pay about half the cost of the upgrade.
Consider a 2GHz CPU, with 2 cycle L1 latency loading one-byte from a 16 byte cache line, at stride 16 bytes. Ignoring TLB thrash, etc., to force a factor of 1000 in latency use a bit rate N, 1000ns = 128b * N ns/b. That's 7ns/b. An ECC EDO RAM 9b wide feeding an off chip MMU would do about that, for example.
Now let's start swapping, and see if we can push a factor of 10e6, eh?
People want native applications, like KOffice, Kdevelop, etc. KDE applications require KDE libraries. KDE must be ported in order to build those applications. All of this porting leads to momentum, so that people arent' asking "why?", but "why not?". Then they do it. The benefit is that once you replace all the Microsoft GUI software with KDE, the system is indistinguishable from from a Linux or Solaris or BSD system for the naive user, and thus Windows becomes mooted. Why is that beneficial? Because the more Bill gets, the less you get. Because choice is good. Because sometimes I'm forced to used Windows, and I don't want to learn how to use it.
The difference between running mostly in L1 cache and regularly going to RAM (particularly when load/store patterns are pessimal), multiplied by the parallelism of exploiting SIMD can quite feasibly give a 1000x performance difference.
Yes, that is the point of QT: A large clientele does not want to be locked into a platform controlled by a monopoly, because it means handing the future of your enterprise over to the control of someone else, whose interests do not correspond to those of the enterprise.
At this point, unless developments in other toolkits change the equation in the meantime, it seems safe to say that once a GPL QT is released, in a reliable state, then QT will provide more robust platform abstraction for UI (over X11/Posix, Win32, and OSX) than any other Free C++ toolkit I am aware of, where "robust" is the operative subjective term in my claim. I think SWT for Java is not even as robust, on OSX (and would be usable only with GCC). SDL, GTK, and FLTK are not competitively feature-rich, as of the last time I looked at the issue seriously. wxWindows (to the best of my understanding), like SWT seriously lags in robustness and reliability on OSX.
Following international practice, I will be boycotting Matsushita products. In practice, this means that when I spec systems they will not be systems including Matsushita drives.
That's not true for mini-ITX based systems. You do pay a bit of a premium and your choices are limited, relative to ATX FF MBs, but they are an off-the-shelf item, and there IS a selection, ranging from fanless 1GHz C3s to HT P4s, with a wide variety of IO options.
That is freakin' amazing. Roadrunner caps you at a lower bitrate than you could get from a freakin' analog modem. Forget that crap, and get another phone line.
Slashdot Mirror
Now *that* is a cool idea.
We'll get to work on it right away.
Another good way to pick up HEU is from "broken arrows" -- lost nuclear devices from bomber crashes or downed submarines. But Plutonium is vastly easier to obtain than HEU, since you can produce it from natural Uranium and fast neutrons, and chemically separate it from the load. If one had about a quarter of a million dollars to spend on a small hydro plant, one could produce more than a critical mass of Pu annually, very low tech.
You seem to be restricting the discussion to HEU bombs. Plutonium is relatively quite easy to
refine, and easy to produce from natural Uranium,
given a source of energetic neutrons. I'm very keen
to see the people get weapons which place their
power on par with that of their ruling classes.
This might enable a democratic society to re-emerge
in the future. If I had a patron, or became wealthy,
I would definitely produce some devices for test,
if an appropriate democratic ruling structure could
be relied upon, with checks and balances, to
protect against immoral use by madmen.
In the long run, these will be too cheap to charge.
Ray Kurzweil told me so.
> just thinking that people who make these things love to just pile on hardware is a bit naive
Or a bit brazenly honest. There's no limit to the benefit from scaling your cluster up, since
(1) there are always applications that run better on more cpus, even with a crappy network design.
(2) there are always better network designs that improve the remaining applications.
(3) there is always the option of partitioning it into mutiple sub-clusters if you're too dumb for (1) and too cheap for (2).
I beg to differ about inherent scalability limits. Perhaps relative to a given application, such as one which requires random routing superlinear in node count, but for the broad range of applications,
for which logarithmic cross-section bandwidth growth is sufficient, there are well-understood architectures that scale without bound.
This stuff was cute and clever in 1992, but these days it's a pretty well-trodden path.
That is consistent with my experience. Minor projects with one or two or three developers will often be a morass, but when an OSS project scales to perhaps 5 or 6 developers, and dozens of patch contributors, it has to have it's ducks in a row,
or it will get forked by someone more competent.
Documentation is less reliably good than is design.
"Architects" who puff up unimplementable dream
designs don't exist in the open source world.
Like Jack Kerouac, I just use VERY long pages.
Essentially zero since Iraq was under an international embargo during the Internet era.
> 70%/50% think attacks on the U.S. forces are justified? That smells of high BS.
Really? I'm surprised that it isn't higher. I mean, suppose France invaded the U.S., imposed the
Napoleonic code and parliamentary government, outlawed
the Republican party, and imprisoned George Bush
for war crimes, after killing 2.5% of the population
(7.325 million people) and bombing NYC and Chicago
into rubble. How many French people would consider
U.S. resistance attacks on legionaires to be justified?
I agree with everything you said, but would note
that by adding PCI RAID IDE controllers, it scales
to the available case cooling and mount points.
I'd probably add some mounting brackets from
erector set pieces and a few fans (and maybe some
acoustic insulation to my closet!!) if I wanted to
scale up to, say 12x400. But that really is the
wall for my personal taste, 3.6TB of RAID 5.
If I ran out of erector set pieces, I'd just pick
up an old gutted tower server case for the job
from ebay or axe-man.
I'd peg the 3.6TB solution at 550 for the
platform, including 3 promise tx2000 cards, and
just under $4k for the drives. That's $1.25
per GB, compared to $0.80 per GB for the 4x250
model. Personally, I'd rather keep multiple
servers, say one for anime and music, one for
data files, one for movies... With network
cross-mounting, it all looks like one big
directory tree anyhow. Of course the middle
road, more drives at the lower $/GB 250GB, $113
sweet spot, still lets you scale to 2.25GB, and
keep it close to $0.80/GB. And it is easier to
upgrade incrementally, by moving one RAID set at
a time to large physical drives, since shuffling
data between controllers is a lot faster than
pumping it over a LAN connection.
I find it easier, faster, and vastly cheaper, to
stick lots of drives and controllers and fans in a
big tower than to buy boutique hardware. I also
find it more reliable, since my personal budget
allows for plenty of spare hardware at the lower
pricepoint. If you really want to run multiple
RAID controllers in a pc case, though, remember
to pick a mondo power supply. you'll probably
need a bunch of power Y connectors too, unless
you're using all SATA drives.
Take a generic low-end PC (say, sempron 2200, with a pc-chips or ecs motherboard, 128MB DDR -- it's all overkill at this point) in a tall, well-ventilated case with at least 300W power, for about $150, and add 4x 250GB IDE drives, for about $450. Run as raid 5 on a debian sarge install. That gives you 750GB of online storage suitable for streaming media for about $600. When 400GB drives fall to the same $0.40/MB price point, make a new box and sell the old one on ebay, to pay about half the cost of the upgrade.
Consider a 2GHz CPU, with 2 cycle L1 latency loading one-byte from a 16 byte cache line, at stride 16 bytes. Ignoring TLB thrash, etc., to force a factor of 1000 in latency use a bit rate N, 1000ns = 128b * N ns/b. That's 7ns/b. An ECC EDO RAM 9b wide feeding an off chip MMU would do about that, for example.
Now let's start swapping, and see if we can push a factor of 10e6, eh?
People want native applications, like KOffice, Kdevelop, etc. KDE applications require KDE libraries. KDE must be ported in order to build those applications. All of this porting leads to momentum, so that people arent' asking "why?", but "why not?". Then they do it. The benefit is that once you replace all the Microsoft GUI software with KDE, the system is indistinguishable from from a Linux or Solaris or BSD system for the naive user, and thus Windows becomes mooted. Why is that beneficial? Because the more Bill gets, the less you get. Because choice is good. Because sometimes I'm forced to used Windows, and I don't want to learn how to use it.
The difference between running mostly in L1 cache and regularly going to RAM (particularly when load/store patterns are pessimal), multiplied by the parallelism of exploiting SIMD can quite feasibly give a 1000x performance difference.
No joke. Bottom-end interrupt handlers run with GC locked. You want security, go pure Java.
> - churn up 2GB swap in java (or something similar)
Only if you regard 64MB as being similar to 2GB.
The stack is limited in Java, by default.
I used up the first two boxes, and can't afford the third. Does that mean it's time to crack open number four?
Yes, that is the point of QT: A large clientele does not want to be locked into a platform controlled by a monopoly, because it means handing the future of your enterprise over to the control of someone else, whose interests do not correspond to those of the enterprise.
At this point, unless developments in other toolkits change the equation in the meantime, it seems safe to say that once a GPL QT is released, in a reliable state, then QT will provide more robust platform abstraction for UI (over X11/Posix, Win32, and OSX) than any other Free C++ toolkit I am aware of, where "robust" is the operative subjective term in my claim. I think SWT for Java is not even as robust, on OSX (and would be usable only with GCC). SDL, GTK, and FLTK are not competitively feature-rich, as of the last time I looked at the issue seriously. wxWindows (to the best of my understanding), like SWT seriously lags in robustness and reliability on OSX.
Given that the remaining 10% is pretty much just a reaction formation, I think you can include that as well.
Following international practice, I will be boycotting Matsushita products. In practice, this means that when I spec systems they will not be systems including Matsushita drives.
That's not true for mini-ITX based systems.
You do pay a bit of a premium and your choices
are limited, relative to ATX FF MBs, but they
are an off-the-shelf item, and there IS a
selection, ranging from fanless 1GHz C3s to
HT P4s, with a wide variety of IO options.
10 years from now, CVD diamonds will be worth about as much as a CZ, but the WiFi ring will still work.
That is freakin' amazing. Roadrunner caps you
at a lower bitrate than you could get from a
freakin' analog modem. Forget that crap, and get
another phone line.
And just how, pray tell, does incorporating a transducer into a device violate thermodynamic principles?