Regrettably, the software is only now getting into a sane state, and the hardware is somewhat old. It would not surprise me if they are dumping below production price even.
prototyping in FPGA is sort-of-like 'write your program in visual basic, and test it, and it will work just fine when you recode it in highly optimised assembler'
Yes, it can tell you that there are no fundamental logic bombs in your implementation concept, but it often doesn't really help.
FPGA design is basically a completely seperate field from 'normal' design - many of the issues with normal design don't actually apply to FPGA, and vice versa.
The FPGA chip designers and compiler writers have done much of the hard work, and you get a simple 'compiles and works/won't fit in this chip' binary output most of the time.
That, and you can't really model analog chips at all in it. It's far from useless, but it's not a real solution to the general problem either.
Let's take for example the OpenMoko Freerunner. It's a mobile phone, with open schematics.
How much would it cost to make one?
It sells for $500 or so, so you might guess $250.
But - that's somewhat different to the question of what it would take to make one.
http://www.pcbcart.com/ - as a reasonable priced chinese PCB service I've looked at - though not used - in the past. For one 50*100mm 8 layer PCB (what you need if you're going to put dense chips on both sides) - they charge $40 for 1-5 PCBs. But - with a $200 setup cost.
So - $250 for the first PCB. Parts cost for ten thousand phones may be $150 or so. But - buying ones of everything, all the parts will cost you $400 very optimistically.
Assembling and soldering this together - there are well over a hundred parts - say $100.
So, that's $750 to get your first prototype.
It doesn't boot. After a couple of weeks and several dead-ends, you find you forgot to connect a pin with a slightly ambiguous name on the datasheet that turns out not to be as unimportant as you thought.
So, if you can't work round it - and it turns out that it's a buried high-speed node under several layers of PCB that is completely inacessible, you need a new PCB made.
Another $750 for the whole lot again. Oh - you may try to reuse some of the parts - but all of these parts do not warranty more than one use, and with a 1% failure rate on removed parts, and the fact that a failed part may stall you for weeks - do you want to do that?
So, you get your new PCB, populate it, and it boots and prints 'loading the lin' and crashes.
After another weeks work, you work out that your routing of the RAM tracks has been slightly out of spec, and that unless you clock the system at under 12MHz, it doesn't work at all.
So, you test all you can at 12MHz, and get another board done.
After a week of wondering why this board doesn't work, you find that one component was installed backwards. Fixing that reveals...
For example, the freerunner release candidate boards had over 7 revisions - and there are still issues with it, and this was a professionally made board made by an actual factory that does these sorts of things all the time.
This hasn't even touched on the sourcing of parts. For many parts this isn't an issue. You can get most chips just fine from many sources online.
Some parts and modules however - in the mass produced and phone sector - are simply unavailable unless you are willing to order 100000. You can't even get docs unless the companies think you will order. And any docs you do get will be under NDA.
Some of these have no easy alternative. You simply can't buy a mobile phone radio chipset for example. You can buy modules - which may have a 200% price, 200% volume penalty.
This works for low speed and simple devices. For high speed, low noise, or wide busses, breadboards don't cut it.
And high speed in this instance is something like >20MHz, and wide busses >8 bits or so.
You essentially can't - say - make a device that runs linux at a speed comparable to a newish mobile phone - without manufacturing a PCB, and soldering on surface mount components, often with several hundred terminals spaced at fractions of a milimeter.
That isn't to say that you can'd do useful stuff with breadboards, simply that they are very limited.
It's like mechanical modelling using lego. Sure, you can do interesting stuff. But while you can in theory make a 350BHP engine from it, it's not going to be sane.
It can't. It's not possible to recharge in its current form, and will be comparatively expensive. Even neglecting that - and the poor discharge rate - it can probably only be discharged at a slow rate due to the design - it is a better hearing aid battery that might last - say - 20 days instead of 10, and be a bit less toxic.
You _could_ put it into your car - but it would require a truly massive battery.
How many people here would go on the shuttle today - given that failure rate - under 1%.
NASA is unfortunately not a results driven organisation,they are a welfare organisation.
Consider the last attempt to reduce the cost of launch.
This had three completely untried technologies that all had to work perfectly in the picked vehicle design. (x33/venturestar).
Conformal tanks (non-spherical or cylindrical tanks that are shaped to fit with the structure).
Metallic thermal protection system - replacing the 'tiles' with a metal scale based system.
Linear aerospike - which had never flown.
NASA is in love with complexity.
Everything must work 100%.
It must be the lightest shiniest most perfect thing that it can be.
Cost is not something you reduce after the design, it's a fundamental aspect of the process that NASA gets entirely backwards.
Take for example the shuttle. In round numbers, the cost of the fuel for the shuttle is.1% of a launch cost.
A sizeable fraction is the standing army to service the thing.
A very simple three stage or so rocket with extremely large margins built in shipyards is not actually technically difficult.
Capsules are low tech - however they are extremely simple and reliable way to deorbit crew. Soyuz has a better record of people not dying on the way down than shuttle, and is vastly cheaper.
Server class hardware should never have hardware faults either.
Yes, server class hardware is usually more robust than consumer grade in terms of some bit errors.
However, in the last minutes or seconds before a crash due to hardware failure, something is obviously going way out of spec.
If this is detectable - it's a no-brainer - you simply failover if thresholds are breached, but before the crash occurs. (and you can afford to be a _lot_ more critical if you've got spare hardware)
But a fair proportion of crashes happen with no software-detectable warning, as there is not hardware coverage of the failing part in sensors.
For example, a failing solderjoint on a chipset or CPU, or...
No, it doesn't. This sort of solution protects from a limited subset of faults.
It protects 100% from any fault that causes instant death.
It does not protect from any fault that causes data corruption, where the system continues to run.
Undetected bit-errors cause the states across the machines to differ.
If these bit errors are replicated, you've got a machine in a copied, but corrupt state - the original and the copy may crash at exactly the same point.
If they aren't, then you may get 'lucky', and have it failover, but with corrupted data that leads to corrupted data being saved, or later crashes.
At best in this situation, you can flag a 'parity error'.
With lots of additional complexity, you might be able to hack this into a redundant system. For example - you run 3 machines, and if one gets into a different state, you disable it.
Several years ago I did some numbers. Addressing solely the fundamental physical limits, neglecting politics.
For >>100 billion - you have to vastly reengineer the planet - under 100 billion, there is hope you could do it more or less normally.
For >>500 billion, you need to completely resurface the planet or large fractions of it.
At around 15 trillion, you reach thermal limits, where the waste heat will cook the earth - even cutting off the incoming sunlight. This would be pretty much 'Trantor' - though growing food locally - if you offworlded the food production, you might get a factor of several more.
At about a trillion trillion, the solar system runs out of resource limits. (unless you can mine the sun)
If you refefine what human is, such as for example massive bioengineering, you might multiply some of the numbers by 10 - for example - reducing phosphorus usage in the human body will allow you to make more humans of the solar system, as it's an element in shortage.
Or if you allow digitised humans, vastly more.
650 billion may be just about doable, when you consider that biosphere II, if you put them down all over the world, comes to 200 billion, it's possible a more optimised system could hit that sort of figure.
I'd say that's on the ragged edge of what could be done with a "soft" system though. (normal plants, though engineered for optimal growth, a few animals and reserves)
A "hard" system, could sustain at least 5 trillion, even assuming that you leave 2/3rds of the planet alone. (eating plants/algae grown under lights, stacked 20 stories high, with 200 square meters per person.(you might melt some of the icecaps with the emitted heat though))
'Almost all CFL are made in China, which is particularly poor in terms of energy efficiency in manufacture and has a poor environmental record in manufacturing, especially for CFL's.'
Assume that the total purchase price of 99p (not a cheapest bulb) goes into purchasing energy at 1p/KWh in china.
This is 99KWh of energy.
Assume that the bulb saves 60W when on (a 22W bulb producing about as much light as a 60W bulb).
This means it's a net positive after only 1600 hours of burning.
There is unfortunately an absolutely fundamental reason why this is different from ethernet.
Ethernet collision detection works only if you can detect the other side transmiting at the same time as you.
Even neglecting the 'as the same time as you' part - if the interfering transmitter is beyond the range that you can decode the data from it, you can't tell it from the background noise, and you can't avoid transmitting over it.
As an easy to understand and not too misleading analogy.
Consider a sports stadium.
If everyone is talking quietly, everyone can talk to their neighbour without too much difficulty.
If one person wants to talk to someone ten seats away, they can yell - however this disrupts people next to thems quiet conversations. If everyone wants to talk to someone ten seats away, and tries to do this by yelling, it just results in everyone only being able to hear their neighbour, and sore ears.
Mesh networking doesn't solve this either.
If you replace the yelling by asking your neighbour to pass the message to someone ten seats away along, then there is no immediate overload - but the overall transmission rate drops to 1/10th - as most of the time you're passing messages for neighbours, not youreself.
And there is only a limited amount of spectrum available.
Sure - you can create ten or a hundred times more bandwidth by expanding the 'wifi' spectrum.
And timeslotted operation relies on one of two things.
An operator to allocate the slots.
A user-proof protocol so that the users cannot screw with the timeslot allocation.
If the users can screw with it - you _will_ get people setting thier and their friends nodes to gain a bit more than their share of bandwidth. And it all goes rather downhill from here.
Unfortunately - the physics doesn't let you do 'free wireless broadband for everyone'.
Radio waves travel until they hit something.
To make an analogy.
You're at a sports event in a large stadium.
You can talk to your neighbour just fine, if everyone is also talking to their neighbour, however you can't be heard by someone 5 seats away though they might hear just fine if everyone was quiet for a moment.
But if everyone raises their voice to the level that they can be heard 5 seats away - the background noise level goes way up, and you can only be heard next to you.
Pretty soon everyones screaming, but can still only be heard by those next to you.
All cellphones work well becasue they very carefully schedule who gets to talk at what time.
If you could arrange it so that everyone was silent, you can easily speak to someone 5 seats away without raising your voice.
To back away from the analogy - because there is less background noise as the cellphone networks arrange their connected devices not to interfere with each other, the handset needs much less power to talk to the phone tower, and gets much better battery life and range.
Mesh networks unfortunately don't fix this either.
You can do collision avoidance for the few closest nodes to you, but pretty soon you can't hear the individual signals, and you simply see them as a background murmer - which sharply reduces the range at which you can talk to nearby nodes.
Add to this the problem that if you have a mesh where every connection does 50 hops to talk to someone - each node has only 2% of its bandwidth left for itself.
A) much simply needs recompiled, if it doesn't - with an app with the source - it's usually a bug. B) No - wine is simply a conversion layer between the windows and linux calls - the windows program is never emulated. C) No - again - not without emulation. D) I think you can probably guess this one - but again no.
Emulation may be _lots_ slower than the host processor - slowdowns of ten times or more are not uncommon.
Light output - total light output, not how bright a spot it can make on a wall - is measured in lumens.
The bulb above - look at the box shot - says 450lm for 6.9W.
That's 65lm/W - this is very close to the 60lm/W that most of my CFLs claim.
It's significantly below the 100lm/W that you get from linear flourescants.
The best LEDs at the moment get around 114lm/W - http://www.cree.com/products/xlamp_xpe.asp - and ones sampling now from the same maker get 140. It's not possible to get more than maybe 300 - as then you're getting close to the fundamental amount of power needed to make a watt of wite light.
I did a basic design for a 100lm/W lamp - from AC to replace 100W of lightbulbs. It needed 15 1W LEDs, a high efficiency supply, and would cost perhaps $60 in bulk.
If you drive the LEDs harder, you can use half the number, and get about 2/3 the efficiency, and maybe $40 cost.
But the unfettered market in utilities, without strong requirements for infrastructure margins large enough to cope with multiple failures - is somewhat broken too.
You use public domain resources and a little research to work out which lines are under most stress.
You get maybe 10 guys with vans and big catapults that throw copper wire over a high voltage line.
You drive along the lines - with a plausible excuse - and you work out where a nice little farm road is that nobody uses for each line, and come 1:23:23 AM (or whenever) you kill all 10 lines as your teams are driving away from thier catapults on a 1 minute trigger.
For added fun have a second few teams that do this at 1:33:23
The margins on all power grids are getting smaller. 'Just in time' 'Predicted demand'... In the bad old days of monopoly providers, the monopoly would often reasonably plan the capacity, and built to some way in excess of this.
With the market coming to shine its light on all these restrictive practices, a new way of looking at power grids has arisen. 'What does upgrading the grid so to next years bottom line'.
We (uk) - the EU too - have both database right - copying a substantive portion of a database is an infringement of this right, and the 'sweat of the brow' doctrine of copyright.
If I spend a significant amount of effort collating some data - then my collation of that data has a copyright seperate to that of the original data.
On the flipside - we have a slightly less broken patent system.
You can extract a kml from a my-maps thing you've drawn on top of googles satellite imagery easily.
But what can you do with this?
Google have made vague and unclear statements that 'bulk' use is not allowed - without saying what this is.
Yahoos terms and conditions allow uses like this, and much of OpenStreetMap has been helped by this for example - people able to trace streets, streams, and...
But the license for data derived from maps is still unclear - can I for example take a list of 3000 river crossings from google, crowdsource how easy they are to get across with a 4x4 or a donkey, and then publish this list?
And if I sell the list, or publish a book of maps using this data combined with openstreetmap data?
Slashdot Mirror
mouse + touchscreen-stylus?
Regrettably, the software is only now getting into a sane state, and the hardware is somewhat old.
It would not surprise me if they are dumping below production price even.
prototyping in FPGA is sort-of-like 'write your program in visual basic, and test it, and it will work just fine when you recode it in highly optimised assembler'
Yes, it can tell you that there are no fundamental logic bombs in your implementation concept, but it often doesn't really help.
FPGA design is basically a completely seperate field from 'normal' design - many of the issues with normal design don't actually apply to FPGA, and vice versa.
The FPGA chip designers and compiler writers have done much of the hard work, and you get a simple 'compiles and works/won't fit in this chip' binary output most of the time.
That, and you can't really model analog chips at all in it.
It's far from useless, but it's not a real solution to the general problem either.
Let's take for example the OpenMoko Freerunner.
It's a mobile phone, with open schematics.
How much would it cost to make one?
It sells for $500 or so, so you might guess $250.
But - that's somewhat different to the question of what it would take to make one.
http://www.pcbcart.com/ - as a reasonable priced chinese PCB service I've looked at - though not used - in the past.
For one 50*100mm 8 layer PCB (what you need if you're going to put dense chips on both sides) - they charge $40 for 1-5 PCBs.
But - with a $200 setup cost.
So - $250 for the first PCB.
Parts cost for ten thousand phones may be $150 or so.
But - buying ones of everything, all the parts will cost you $400 very optimistically.
Assembling and soldering this together - there are well over a hundred parts - say $100.
So, that's $750 to get your first prototype.
It doesn't boot.
After a couple of weeks and several dead-ends, you find you forgot to connect a pin with a slightly ambiguous name on the datasheet that turns out not to be as unimportant as you thought.
So, if you can't work round it - and it turns out that it's a buried high-speed node under several layers of PCB that is completely inacessible, you need a new PCB made.
Another $750 for the whole lot again. Oh - you may try to reuse some of the parts - but all of these parts do not warranty more than one use, and with a 1% failure rate on removed parts, and the fact that a failed part may stall you for weeks - do you want to do that?
So, you get your new PCB, populate it, and it boots and prints 'loading the lin' and crashes.
After another weeks work, you work out that your routing of the RAM tracks has been slightly out of spec, and that unless you clock the system at under 12MHz, it doesn't work at all.
So, you test all you can at 12MHz, and get another board done.
After a week of wondering why this board doesn't work, you find that one component was installed backwards. Fixing that reveals...
For example, the freerunner release candidate boards had over 7 revisions - and there are still issues with it, and this was a professionally made board made by an actual factory that does these sorts of things all the time.
This hasn't even touched on the sourcing of parts.
For many parts this isn't an issue.
You can get most chips just fine from many sources online.
Some parts and modules however - in the mass produced and phone sector - are simply unavailable unless you are willing to order 100000. You can't even get docs unless the companies think you will order. And any docs you do get will be under NDA.
Some of these have no easy alternative. You simply can't buy a mobile phone radio chipset for example. You can buy modules - which may have a 200% price, 200% volume penalty.
This works for low speed and simple devices.
For high speed, low noise, or wide busses, breadboards don't cut it.
And high speed in this instance is something like >20MHz, and wide busses >8 bits or so.
You essentially can't - say - make a device that runs linux at a speed comparable to a newish mobile phone - without manufacturing a PCB, and soldering on surface mount components, often with several hundred terminals spaced at fractions of a milimeter.
That isn't to say that you can'd do useful stuff with breadboards, simply that they are very limited.
It's like mechanical modelling using lego.
Sure, you can do interesting stuff.
But while you can in theory make a 350BHP engine from it, it's not going to be sane.
Why?
If they can sell the widget for $100, then there isn't a problem for them paying $20 in tooling costs.
It's just a regrettable factor that opening the design may kill any profit.
It can't.
It's not possible to recharge in its current form, and will be comparatively expensive.
Even neglecting that - and the poor discharge rate - it can probably only be discharged at a slow rate due to the design - it is a better hearing aid battery that might last - say - 20 days instead of 10, and be a bit less toxic.
You _could_ put it into your car - but it would require a truly massive battery.
How many people here would go on the shuttle today - given that failure rate - under 1%.
NASA is unfortunately not a results driven organisation,they are a welfare organisation.
Consider the last attempt to reduce the cost of launch.
This had three completely untried technologies that all had to work perfectly in the picked vehicle design. (x33/venturestar).
Conformal tanks (non-spherical or cylindrical tanks that are shaped to fit with the structure).
Metallic thermal protection system - replacing the 'tiles' with a metal scale based system.
Linear aerospike - which had never flown.
NASA is in love with complexity.
Everything must work 100%.
It must be the lightest shiniest most perfect thing that it can be.
Cost is not something you reduce after the design, it's a fundamental aspect of the process that NASA gets entirely backwards.
Take for example the shuttle. .1% of a launch cost.
In round numbers, the cost of the fuel for the shuttle is
A sizeable fraction is the standing army to service the thing.
A very simple three stage or so rocket with extremely large margins built in shipyards is not actually technically difficult.
Capsules are low tech - however they are extremely simple and reliable way to deorbit crew.
Soyuz has a better record of people not dying on the way down than shuttle, and is vastly cheaper.
Server class hardware should never have hardware faults either.
Yes, server class hardware is usually more robust than consumer grade in terms of some bit errors.
However, in the last minutes or seconds before a crash due to hardware failure, something is obviously going way out of spec.
If this is detectable - it's a no-brainer - you simply failover if thresholds are breached, but before the crash occurs. (and you can afford to be a _lot_ more critical if you've got spare hardware)
But a fair proportion of crashes happen with no software-detectable warning, as there is not hardware coverage of the failing part in sensors.
For example, a failing solderjoint on a chipset or CPU, or ...
No, it doesn't.
This sort of solution protects from a limited subset of faults.
It protects 100% from any fault that causes instant death.
It does not protect from any fault that causes data corruption, where the system continues to run.
Undetected bit-errors cause the states across the machines to differ.
If these bit errors are replicated, you've got a machine in a copied, but corrupt state - the original and the copy may crash at exactly the same point.
If they aren't, then you may get 'lucky', and have it failover, but with corrupted data that leads to corrupted data being saved, or later crashes.
At best in this situation, you can flag a 'parity error'.
With lots of additional complexity, you might be able to hack this into a redundant system.
For example - you run 3 machines, and if one gets into a different state, you disable it.
This has problems of its own though.
Several years ago I did some numbers. Addressing solely the fundamental physical limits, neglecting politics.
For >>100 billion - you have to vastly reengineer the planet - under 100 billion, there is hope you could do it more or less normally.
For >>500 billion, you need to completely resurface the planet or large fractions of it.
At around 15 trillion, you reach thermal limits, where the waste heat will cook the earth - even cutting off the incoming sunlight. This would be pretty much 'Trantor' - though growing food locally - if you offworlded the food production, you might get a factor of several more.
At about a trillion trillion, the solar system runs out of resource limits. (unless you can mine the sun)
If you refefine what human is, such as for example massive bioengineering, you might multiply some of the numbers by 10 - for example - reducing phosphorus usage in the human body will allow you to make more humans of the solar system, as it's an element in shortage.
Or if you allow digitised humans, vastly more.
650 billion may be just about doable, when you consider
that biosphere II, if you put them down all over the world, comes
to 200 billion, it's possible a more optimised system could hit
that sort of figure.
I'd say that's on the ragged edge of what could be done with a "soft"
system though.
(normal plants, though engineered for optimal growth, a few animals and reserves)
A "hard" system, could sustain at least 5 trillion, even assuming that
you leave 2/3rds of the planet alone. (eating plants/algae grown under
lights, stacked 20 stories high, with 200 square meters per person.(you
might melt some of the icecaps with the emitted heat though))
'Almost all CFL are made in China, which is particularly poor in terms of energy efficiency in manufacture and has a poor environmental record in manufacturing, especially for CFL's.'
Assume that the total purchase price of 99p (not a cheapest bulb) goes into purchasing energy at 1p/KWh in china.
This is 99KWh of energy.
Assume that the bulb saves 60W when on (a 22W bulb producing about as much light as a 60W bulb).
This means it's a net positive after only 1600 hours of burning.
Most of my CFLs last a _lot_ longer than that.
No, it doesn't.
Dialup works just fine for accessing bank site, or ... if they're properly coded.
There is unfortunately an absolutely fundamental reason why this is different from ethernet.
Ethernet collision detection works only if you can detect the other side transmiting at the same time as you.
Even neglecting the 'as the same time as you' part - if the interfering transmitter is beyond the range that you can decode the data from it, you can't tell it from the background noise, and you can't avoid transmitting over it.
As an easy to understand and not too misleading analogy.
Consider a sports stadium.
If everyone is talking quietly, everyone can talk to their neighbour without too much difficulty.
If one person wants to talk to someone ten seats away, they can yell - however this disrupts people next to thems quiet conversations.
If everyone wants to talk to someone ten seats away, and tries to do this by yelling, it just results in everyone only being able to hear their neighbour, and sore ears.
Mesh networking doesn't solve this either.
If you replace the yelling by asking your neighbour to pass the message to someone ten seats away along, then there is no immediate overload - but the overall transmission rate drops to 1/10th - as most of the time you're passing messages for neighbours, not youreself.
It's actually quite trivial - and getting more so to move your data out of google apps.
See the recent 'data liberation' things they've been doing.
And there is only a limited amount of spectrum available.
Sure - you can create ten or a hundred times more bandwidth by expanding the 'wifi' spectrum.
And timeslotted operation relies on one of two things.
An operator to allocate the slots.
A user-proof protocol so that the users cannot screw with the timeslot allocation.
If the users can screw with it - you _will_ get people setting thier and their friends nodes to gain a bit more than their share of bandwidth.
And it all goes rather downhill from here.
Unfortunately - the physics doesn't let you do 'free wireless broadband for everyone'.
Radio waves travel until they hit something.
To make an analogy.
You're at a sports event in a large stadium.
You can talk to your neighbour just fine, if everyone is also talking to their neighbour, however you can't be heard by someone 5 seats away though they might hear just fine if everyone was quiet for a moment.
But if everyone raises their voice to the level that they can be heard 5 seats away - the background noise level goes way up, and you can only be heard next to you.
Pretty soon everyones screaming, but can still only be heard by those next to you.
All cellphones work well becasue they very carefully schedule who gets to talk at what time.
If you could arrange it so that everyone was silent, you can easily speak to someone 5 seats away without raising your voice.
To back away from the analogy - because there is less background noise as the cellphone networks arrange their connected devices not to interfere with each other, the handset needs much less power to talk to the phone tower, and gets much better battery life and range.
Mesh networks unfortunately don't fix this either.
You can do collision avoidance for the few closest nodes to you, but pretty soon you can't hear the individual signals, and you simply see them as a background murmer - which sharply reduces the range at which you can talk to nearby nodes.
Add to this the problem that if you have a mesh where every connection does 50 hops to talk to someone - each node has only 2% of its bandwidth left for itself.
Indeed - it's trivial to generate the huffman dictionary given the input data and a copy of the TV listings.
That's not really the point.
The point is that anyone distributing such a dictionary is infringing the BBCs Database Right - which is akin to copyright for databases.
Teh only way you can get a legitimate copy of this database/key is to agree to the BBC's terms and conditions - which require DRM.
It's - largely - to stop TV recorders saving to DVD or something.
It has the incidental effect of making anyone using linux and a DTTV card to recieve the broadcasts act illegally - but...
A) much simply needs recompiled, if it doesn't - with an app with the source - it's usually a bug.
B) No - wine is simply a conversion layer between the windows and linux calls - the windows program is never emulated.
C) No - again - not without emulation.
D) I think you can probably guess this one - but again no.
Emulation may be _lots_ slower than the host processor - slowdowns of ten times or more are not uncommon.
Light output - total light output, not how bright a spot it can make on a wall - is measured in lumens.
The bulb above - look at the box shot - says 450lm for 6.9W.
That's 65lm/W - this is very close to the 60lm/W that most of my CFLs claim.
It's significantly below the 100lm/W that you get from linear flourescants.
The best LEDs at the moment get around 114lm/W - http://www.cree.com/products/xlamp_xpe.asp - and ones sampling now from the same maker get 140. It's not possible to get more than maybe 300 - as then you're getting close to the fundamental amount of power needed to make a watt of wite light.
I did a basic design for a 100lm/W lamp - from AC to replace 100W of lightbulbs.
It needed 15 1W LEDs, a high efficiency supply, and would cost perhaps $60 in bulk.
If you drive the LEDs harder, you can use half the number, and get about 2/3 the efficiency, and maybe $40 cost.
This was for relatively small numbers.
Yes - monopolies don't do everything right.
But the unfettered market in utilities, without strong requirements for infrastructure margins large enough to cope with multiple failures - is somewhat broken too.
Or you cut a lot of lines.
You use public domain resources and a little research to work out which lines are under most stress.
You get maybe 10 guys with vans and big catapults that throw copper wire over a high voltage line.
You drive along the lines - with a plausible excuse - and you work out where a nice little farm road is that nobody uses for each line, and come 1:23:23 AM (or whenever) you kill all 10 lines as your teams are driving away from thier catapults on a 1 minute trigger.
For added fun have a second few teams that do this at 1:33:23
The margins on all power grids are getting smaller. ...
'Just in time'
'Predicted demand'
In the bad old days of monopoly providers, the monopoly would often reasonably plan the capacity, and built to some way in excess of this.
With the market coming to shine its light on all these restrictive practices, a new way of looking at power grids has arisen.
'What does upgrading the grid so to next years bottom line'.
Clearly this is the better solution.
And the US is not the whole world.
We (uk) - the EU too - have both database right - copying a substantive portion of a database is an infringement of this right, and the 'sweat of the brow' doctrine of copyright.
If I spend a significant amount of effort collating some data - then my collation of that data has a copyright seperate to that of the original data.
On the flipside - we have a slightly less broken patent system.
The highest rated suggestion - over a thousand votes - on the data liberation site is about Google Maps.
Specifically - the rather loose definition of what we can and can't do with the data.
http://moderator.appspot.com/#15/e=43649&t=4364a
You can extract a kml from a my-maps thing you've drawn on top of googles satellite imagery easily.
But what can you do with this?
Google have made vague and unclear statements that 'bulk' use is not allowed - without saying what this is.
Yahoos terms and conditions allow uses like this, and much of OpenStreetMap has been helped by this for example - people able to trace streets, streams, and ...
But the license for data derived from maps is still unclear - can I for example take a list of 3000 river crossings from google, crowdsource how easy they are to get across with a 4x4 or a donkey, and then publish this list?
And if I sell the list, or publish a book of maps using this data combined with openstreetmap data?