Hiring H1-B is pretty easy actually. It's no harder than filling out your tax return every year. You get your candidate lined up, you tailor the advertisement to his resume, you file the paperwork, post an advertisement, hire the dude (order doesn't really matter).
H1-B's are pretty cheap, you can get a PhD for $30k/year. The paperwork may cost you ~$5k if you don't have your own legal staff but given that comparable PhD's go for about $120k and a bunch of entitlements state-side, you save quite a bit of money in the long run. The PhD lives cheap locally, sends money back home and doesn't want to leave because he makes like double or triple what he made back home.
Well, the Pi can give you a 1GHz CPU with dual-core GPU, Ethernet, USB and 512MB RAM for $25 at 3.5W. There is just not a budget (yet) for what you're asking and probably by the time it is, you'd be saying "quad core and 4GB RAM? I want the power. So a 16 core with 32GB RAM using just 1W of power...".
The biggest cost of hardware these days sits in manufacturing and transportation costs and sometimes licensing, not component costs (the component costs of the Pi are probably ~$5)
If he hadn't run away and made it a publicity stunt, how far do you think he would've got with either his story or his life? It may have been a blip on someone's newsfeed but he would've made as much noise as Bradley Manning - 3 days of cover story, 3 years of court-battles and torture just to convert his sentence from the death penalty to life. Most people you ask today still don't know who Manning was or why he's being convicted to the death penalty or why he's being tortured in US prisons.
As much as you may snort at the guy for running away, he's still free.
The US is one of the most free countries as long as you stay within the guidelines of it's government. As soon as you blow the whistle on any of it's corruptions, you're going to be just as free as you would be in China or Thailand. Will you die? The chance is high in either country. The fact that the US isn't more overt than say China or Russia about what happens to their constituents that don't remain in-line doesn't mean that it's more free.
The writer doesn't seem to understand the difference between big data in the banking systems and big data in content analytics. Big banking systems you pretty much know what's happening at any point, the data pointers are relatively small even if the chunks are big so you can use mmap because your index most likely fits in a faster version of your range of memory.
However when the number of data pointers grows beyond the size of your actual data or your index is larger than what any of your fast memory supports, your mmap won't work. A quick example would be ZFS which has this problem when your de-duplication index grows beyond memory (which is a very small, very efficient piece of code using a very efficient form of memory mappings for it's scale) and things revert to using the hard drive.
It seems to me this is an advertisement for this company that sells hand-crafted code at $100k/CPU. Sure, if you hand-craft your code, it will be very efficient, very fast and a lot of other things. However it won't be cheap, it won't be easy to develop, it definitely won't be easy to read, debug, port, extend, fork or anything else the higher level languages has allowed us to do.
As a neck-bearded coder that REMEMBERS actually using some of these systems - yeah, we did a LOT in 64kb of memory. However we also spent HOURS to fit that stuff in 64kb and when we were done, well Perl coders would cry if they saw some of the constructions. Also, a lot of this was done towards a very specific architecture and even timings. Point-in-case would be pre-Intel 386 code - we used game assets to generate audio data. Increasing the speed of the CPU from a 80286 would make those games run faster and things that were previously running from a floppy would crash because the floppy routine couldn't supply data fast enough.
Actually mesh routing works quite well, it is basically how the Internet operates. TCP/IP works quite well on mesh because it will automatically figure out the lowest cost and the broken paths. The problem is typically with UDP or higher-layer (usually closed-source) protocols that pretend to re-invent TCP on another layer as well as too many technologies are bolted onto the wrong protocols. There is no reason VoIP-protocols couldn't work on TCP other than bandwidth and a bit more processing on either end of the link, but at least you can see the rationale. But when you then bolt on video streaming services on top of UDP (UDP doesn't care whether or not it gets delivered) you end up streaming lots and lots of unnecessary content to dead endpoints. So let's bolt on another Transport Protocol and a Messaging Protocol using the same UDP-stream (looking at you Adobe).
In-house mesh networking (wireless) is physically impractical because you need new wiring and in most cases a single router can handle the load and isn't going to be the problem. It's a solution looking for a problem.
There is no reason we don't have an Internet-of-things other than cost. It costs $5-25 to add "Teh Interwebz" to a device, it costs probably $5/device in development costs for a UI and $5 in support costs. That's $15-45 on a microwave that costs $35 at Wal-Mart. So you're at Wal-Mart looking for another microwave that will die somewhere in the next 2 years - $35 or $80...
It seems more and more to me that Bin Laden was just a smoke screen. The whole thing being organized by the US seems more and more likely. How many people do you need to shut up to pull something like the widespread wiretapping - 100's if not 1000's of people up and down the chain of command down to the techs splicing the fiber into the NSA box, plenty of people that frequent dive bars and get drunk on a frequent basis. How many people do you need to hijack 2 or 3 planes? Maybe about a dozen people within the CoC and half a dozen mercenaries. I wouldn't be surprised that within 30-50 years documents get declassified that show the entire thing to be a NSA-funded job (we already know the CIA trained and funded these 'terrorists' in the 80's)
Especially that last one, the infrastructure is cheap (3 or 4 servers and a single sysadmin will give you management for 400,000 iPads). When each e-book costs on average $60/student, that's where most of your money goes.
Experience, I was talking about serious computations, not something you use gaming GPU's for and not a single optimized task either. You could probably take any random task and optimize the shit out of it on either platform, doesn't mean said platform is good for a "general purpose" (the GP in GPGPU).
Also, even for Bitcoin mining, a lot of rigs require Windows, not Linux, a lot of cards in the Bitcoin guides have disclaimers such as "don't use x with this AMD card" where x is some type of OpenCL instruction or computation type.
I have written code for computational biology - CUDA is a lot easier to pick up if you're just converting from C. They have great examples and documentation, great plugins but you're stuck on a single hardware platform. OpenCL on the other hand is a lot less 'nice' to begin with (pouring over 250 page PDFs with minimal explanation) but allows you to leverage both CPU and GPU efficiently and a lot less hardware independent although these days it's just nVidia for serious GPU computing and maybe Intel is starting to get into the game (don't know, haven't come across their hardware yet), AMD is a joke, not even all their GPUs (or drivers) have support for GPGPU yet and their drivers just suck.
Actually, Tesla is the only vehicle that does not require government subsidies and was being made before the government even stepped in. Fisker and the rest is what happens when you only use government subsidies - the project was porked, overburdened with regulation and doomed before it even started.
"There is no reason anyone would want a computer in their home." -- Ken Olson, founder, chairman & president of DEC, 1977
“This ‘telephone’ has too many shortcomings to be seriously considered as a means of communication. The device is inherently of no value to us.” — A memo at Western Union, 1878 (or 1876).
“Television won’t last. It’s a flash in the pan.” — Mary Somerville, pioneer of radio educational broadcasts, 1948.
Who is to say that Exchange 201x won't do the same thing or doesn't already? Or any number of proprietary systems? You don't know because you can't see what's really happening with closed protocols, software and devices.
It's relatively easy to get those people though. But you have to actually offer them at least higher than what they currently make.
If there was truly a shortage of workers, you could easily poach droves of developers from other companies, you would get heaps of requests from people trying to jump ship. "getting candidates who can think critically and actually problem solve is damn near impossible" - True, the problem is that there are way too many developers unemployed and not enough companies that want to invest both time and money to hire and train them properly and sift the chaff.
If you are really what you say you are, I suggest actively pursuing developers, give them a salary and train them. There is no such thing as a developer-in-a-can and offering them $100,000 while the market suggests the average junior developer makes $120,000 (still 6 figures)? I know a guy who is shit for developer (he couldn't think his way out of a cardboard box) and got offered $120k for a position.
Quad Core CPU, Single Core GPU vs Single Core CPU, Dual Core GPU. The CPU is not really the problem on most of these platforms, it's the GPU. With RPi now getting Wayland builds that finally use the GPU (mind you, everything (compositing etc) was done on that paltry 700MHz ARM CPU which is really only meant to be an interface to the GPU)
I think you underestimate how useless the Android sticks are and how underpowered it's GPU. The RPi GPU clocks at whatever the CPU clocks at (700MHz), the Mali GPU at 500MHz, the MK802 clocks not at 1GHz but is underclocked to somewhere around ~780MHz to keep it relatively stable. Mind you that you can also clock the RPi to 1GHz if you don't mind a relatively unstable board (800MHz is typically okay).
I developed an embedded media platform on the Pi and I've tested some other designs as well, the Android stick while being $15-25 more expensive was slower doing ffmpeg tasks and had problems handling more than one video stream, at 100% CPU (ffmpeg conversion tasks) the Pi would chug along for hours while the Android would regularly reboot and heat up tremendously in the process, it also demands a lot more power (roughly double that of the Pi). The Pi overlays one video stream on the other without much of a hiccup. If the Android did any good it was detecting issues with my programs when they were unexpectedly interrupted.
However it's really good at doing things that use the GPU, it's a lot more stable and faster than the alternatives at that price point. A single core 700MHz CPU with 256/512MB RAM and 128MB of VRAM was all we had a couple of years ago and we did really well with it. Sure you won't drive the most modern accelerated GUIs with it but a static, usable GUI works pretty well.
Disclaimer: I have developed a professional embedded advertisement system on the RPi with currently about 50 Pi's deployed.
Not necessarily. What if you simply get your media carrier stolen. You can't derive passwords from it, you can derive a history and cookies however which if you simply copy them to a similar system gives you straight back access to this.
Also, if someone intercepts the cookies or does a HTTPS MITM redirect, the malicious site could simply ask for the cookies for said domain and then still forward you to the legitimate HTTPS domain.
Traffic court doesn't have to follow due process of law, heck some cities don't even have traffic courts anymore, they simply use a traffic violations bureau in which there is an administrator basically rubber-stamping whatever the ticket/officer says, there is no due process, no judge, no jury. Those offices have been mandated to maintain a certain conviction rate (65% or so).
Those things have been considered constitutional for all non-criminal offenses, I don't know the legal argument but I think it boils down to "driving is a privilege not a right, these tickets aren't criminal, therefore the constitution doesn't apply".
Off course the state can't compel me to give it to them but if I don't, they'll just keep you running in the administrative treadmill forever. If you don't mind not driving for the next 20 years before the Supreme Court takes a look at it, then you're free to do so.
I recommend Apple Airport's as well. They are somewhat limited in feature but they are really stable, I have some that are over 10 years old that work perfectly (b/g only). I also second Buffalo routers. A little more expensive than your el-cheapo home router but stable as hell and comes with DD-WRT.
Asus also makes some good ones but make sure to replace the wall-wart it comes (mine was a 2A) with with a slightly higher wattage/current (I used a 3A but measured usage spikes of 2.2A). The 2A that came in the box blew out about 9 months in and I suspect that's what happens with a lot of other routers too - they simply don't or only barely have the juice to power the radio under certain conditions and any line power issue reduces the voltage it supplies to the point the processor gets stuck, reboots or simply burns out. They also come under spec for quality routing this day and age. I still see routers shipping with 16 or 32MB of RAM and 200MHz processors and they're supposed to do gigabit routing on 4-8 ports and 600Mbps wireless, firewall, dhcpd, NAT,... A simple routing and NAT table can fill that amount of RAM up in no time when a couple of people have torrents on.
In the city I'm at they don't even consider it a fine. They consider it a surcharge that you have to pay to the company that owns the red light camera's. The city probably gets a cut from it but you can't fight it in court because it's not a recognized fine under local or state laws. However if you don't pay the surcharge, they can still suspend your license because not paying the surcharge is recognized in the law. So you can go to court when they suspend your license during which proceedings (6 months for the next court date) you don't have a license (which in the US is practically a death sentence). You can then get your license back on a constitutional technicality (right to a speedy trial) but since the surcharge remains pending, every 12 months they'll re-suspend it.
What I meant is with a typical keyboard, the typical poll rate is ~250Hz on most OS HID drivers. You can increase that to ~1000Hz (either direct hardware access or alter the driver).
Yes, the keyboard has an inherent delay as well but there is no reason it should be as high as 40ms, even very slow microprocessors can scan the matrix much faster than that. In fact that would be fairly noticeable (40ms + OS latency). The typical modern keyboard I have tested to be about 10ms which includes it's own delay and the OS delay.
Slashdot Mirror
Hiring H1-B is pretty easy actually. It's no harder than filling out your tax return every year. You get your candidate lined up, you tailor the advertisement to his resume, you file the paperwork, post an advertisement, hire the dude (order doesn't really matter).
H1-B's are pretty cheap, you can get a PhD for $30k/year. The paperwork may cost you ~$5k if you don't have your own legal staff but given that comparable PhD's go for about $120k and a bunch of entitlements state-side, you save quite a bit of money in the long run. The PhD lives cheap locally, sends money back home and doesn't want to leave because he makes like double or triple what he made back home.
Well, the Pi can give you a 1GHz CPU with dual-core GPU, Ethernet, USB and 512MB RAM for $25 at 3.5W. There is just not a budget (yet) for what you're asking and probably by the time it is, you'd be saying "quad core and 4GB RAM? I want the power. So a 16 core with 32GB RAM using just 1W of power...".
The biggest cost of hardware these days sits in manufacturing and transportation costs and sometimes licensing, not component costs (the component costs of the Pi are probably ~$5)
Russia's a nice place, go visit sometimes.
If he hadn't run away and made it a publicity stunt, how far do you think he would've got with either his story or his life? It may have been a blip on someone's newsfeed but he would've made as much noise as Bradley Manning - 3 days of cover story, 3 years of court-battles and torture just to convert his sentence from the death penalty to life. Most people you ask today still don't know who Manning was or why he's being convicted to the death penalty or why he's being tortured in US prisons.
As much as you may snort at the guy for running away, he's still free.
The US is one of the most free countries as long as you stay within the guidelines of it's government. As soon as you blow the whistle on any of it's corruptions, you're going to be just as free as you would be in China or Thailand. Will you die? The chance is high in either country. The fact that the US isn't more overt than say China or Russia about what happens to their constituents that don't remain in-line doesn't mean that it's more free.
The US is far from the most free country in the world, in fact on several scales it barely even enters the top 10:
http://en.wikipedia.org/wiki/Press_Freedom_Index
http://en.wikipedia.org/wiki/Index_of_Economic_Freedom
The writer doesn't seem to understand the difference between big data in the banking systems and big data in content analytics. Big banking systems you pretty much know what's happening at any point, the data pointers are relatively small even if the chunks are big so you can use mmap because your index most likely fits in a faster version of your range of memory.
However when the number of data pointers grows beyond the size of your actual data or your index is larger than what any of your fast memory supports, your mmap won't work. A quick example would be ZFS which has this problem when your de-duplication index grows beyond memory (which is a very small, very efficient piece of code using a very efficient form of memory mappings for it's scale) and things revert to using the hard drive.
It seems to me this is an advertisement for this company that sells hand-crafted code at $100k/CPU. Sure, if you hand-craft your code, it will be very efficient, very fast and a lot of other things. However it won't be cheap, it won't be easy to develop, it definitely won't be easy to read, debug, port, extend, fork or anything else the higher level languages has allowed us to do.
As a neck-bearded coder that REMEMBERS actually using some of these systems - yeah, we did a LOT in 64kb of memory. However we also spent HOURS to fit that stuff in 64kb and when we were done, well Perl coders would cry if they saw some of the constructions. Also, a lot of this was done towards a very specific architecture and even timings. Point-in-case would be pre-Intel 386 code - we used game assets to generate audio data. Increasing the speed of the CPU from a 80286 would make those games run faster and things that were previously running from a floppy would crash because the floppy routine couldn't supply data fast enough.
Actually mesh routing works quite well, it is basically how the Internet operates. TCP/IP works quite well on mesh because it will automatically figure out the lowest cost and the broken paths. The problem is typically with UDP or higher-layer (usually closed-source) protocols that pretend to re-invent TCP on another layer as well as too many technologies are bolted onto the wrong protocols. There is no reason VoIP-protocols couldn't work on TCP other than bandwidth and a bit more processing on either end of the link, but at least you can see the rationale. But when you then bolt on video streaming services on top of UDP (UDP doesn't care whether or not it gets delivered) you end up streaming lots and lots of unnecessary content to dead endpoints. So let's bolt on another Transport Protocol and a Messaging Protocol using the same UDP-stream (looking at you Adobe).
In-house mesh networking (wireless) is physically impractical because you need new wiring and in most cases a single router can handle the load and isn't going to be the problem. It's a solution looking for a problem.
There is no reason we don't have an Internet-of-things other than cost. It costs $5-25 to add "Teh Interwebz" to a device, it costs probably $5/device in development costs for a UI and $5 in support costs. That's $15-45 on a microwave that costs $35 at Wal-Mart. So you're at Wal-Mart looking for another microwave that will die somewhere in the next 2 years - $35 or $80...
It seems more and more to me that Bin Laden was just a smoke screen. The whole thing being organized by the US seems more and more likely. How many people do you need to shut up to pull something like the widespread wiretapping - 100's if not 1000's of people up and down the chain of command down to the techs splicing the fiber into the NSA box, plenty of people that frequent dive bars and get drunk on a frequent basis. How many people do you need to hijack 2 or 3 planes? Maybe about a dozen people within the CoC and half a dozen mercenaries. I wouldn't be surprised that within 30-50 years documents get declassified that show the entire thing to be a NSA-funded job (we already know the CIA trained and funded these 'terrorists' in the 80's)
Especially that last one, the infrastructure is cheap (3 or 4 servers and a single sysadmin will give you management for 400,000 iPads). When each e-book costs on average $60/student, that's where most of your money goes.
Experience, I was talking about serious computations, not something you use gaming GPU's for and not a single optimized task either. You could probably take any random task and optimize the shit out of it on either platform, doesn't mean said platform is good for a "general purpose" (the GP in GPGPU).
Also, even for Bitcoin mining, a lot of rigs require Windows, not Linux, a lot of cards in the Bitcoin guides have disclaimers such as "don't use x with this AMD card" where x is some type of OpenCL instruction or computation type.
I have written code for computational biology - CUDA is a lot easier to pick up if you're just converting from C. They have great examples and documentation, great plugins but you're stuck on a single hardware platform. OpenCL on the other hand is a lot less 'nice' to begin with (pouring over 250 page PDFs with minimal explanation) but allows you to leverage both CPU and GPU efficiently and a lot less hardware independent although these days it's just nVidia for serious GPU computing and maybe Intel is starting to get into the game (don't know, haven't come across their hardware yet), AMD is a joke, not even all their GPUs (or drivers) have support for GPGPU yet and their drivers just suck.
Imagine the licensing cost to run 1M servers on MS Windows Server ($1k/CPU or something like that). They would save a lot of money switching to Linux!
Actually, Tesla is the only vehicle that does not require government subsidies and was being made before the government even stepped in. Fisker and the rest is what happens when you only use government subsidies - the project was porked, overburdened with regulation and doomed before it even started.
"There is no reason anyone would want a computer in their home."
-- Ken Olson, founder, chairman & president of DEC, 1977
“This ‘telephone’ has too many shortcomings to be seriously considered as a means of communication. The device is inherently of no value to us.” — A memo at Western Union, 1878 (or 1876).
“Television won’t last. It’s a flash in the pan.” — Mary Somerville, pioneer of radio educational broadcasts, 1948.
Who is to say that Exchange 201x won't do the same thing or doesn't already? Or any number of proprietary systems? You don't know because you can't see what's really happening with closed protocols, software and devices.
It's relatively easy to get those people though. But you have to actually offer them at least higher than what they currently make.
If there was truly a shortage of workers, you could easily poach droves of developers from other companies, you would get heaps of requests from people trying to jump ship. "getting candidates who can think critically and actually problem solve is damn near impossible" - True, the problem is that there are way too many developers unemployed and not enough companies that want to invest both time and money to hire and train them properly and sift the chaff.
If you are really what you say you are, I suggest actively pursuing developers, give them a salary and train them. There is no such thing as a developer-in-a-can and offering them $100,000 while the market suggests the average junior developer makes $120,000 (still 6 figures)? I know a guy who is shit for developer (he couldn't think his way out of a cardboard box) and got offered $120k for a position.
Quad Core CPU, Single Core GPU vs Single Core CPU, Dual Core GPU. The CPU is not really the problem on most of these platforms, it's the GPU. With RPi now getting Wayland builds that finally use the GPU (mind you, everything (compositing etc) was done on that paltry 700MHz ARM CPU which is really only meant to be an interface to the GPU)
I think you underestimate how useless the Android sticks are and how underpowered it's GPU. The RPi GPU clocks at whatever the CPU clocks at (700MHz), the Mali GPU at 500MHz, the MK802 clocks not at 1GHz but is underclocked to somewhere around ~780MHz to keep it relatively stable. Mind you that you can also clock the RPi to 1GHz if you don't mind a relatively unstable board (800MHz is typically okay).
I developed an embedded media platform on the Pi and I've tested some other designs as well, the Android stick while being $15-25 more expensive was slower doing ffmpeg tasks and had problems handling more than one video stream, at 100% CPU (ffmpeg conversion tasks) the Pi would chug along for hours while the Android would regularly reboot and heat up tremendously in the process, it also demands a lot more power (roughly double that of the Pi). The Pi overlays one video stream on the other without much of a hiccup. If the Android did any good it was detecting issues with my programs when they were unexpectedly interrupted.
However it's really good at doing things that use the GPU, it's a lot more stable and faster than the alternatives at that price point. A single core 700MHz CPU with 256/512MB RAM and 128MB of VRAM was all we had a couple of years ago and we did really well with it. Sure you won't drive the most modern accelerated GUIs with it but a static, usable GUI works pretty well.
Disclaimer: I have developed a professional embedded advertisement system on the RPi with currently about 50 Pi's deployed.
Not necessarily. What if you simply get your media carrier stolen. You can't derive passwords from it, you can derive a history and cookies however which if you simply copy them to a similar system gives you straight back access to this.
Also, if someone intercepts the cookies or does a HTTPS MITM redirect, the malicious site could simply ask for the cookies for said domain and then still forward you to the legitimate HTTPS domain.
Traffic court doesn't have to follow due process of law, heck some cities don't even have traffic courts anymore, they simply use a traffic violations bureau in which there is an administrator basically rubber-stamping whatever the ticket/officer says, there is no due process, no judge, no jury. Those offices have been mandated to maintain a certain conviction rate (65% or so).
Those things have been considered constitutional for all non-criminal offenses, I don't know the legal argument but I think it boils down to "driving is a privilege not a right, these tickets aren't criminal, therefore the constitution doesn't apply".
Off course the state can't compel me to give it to them but if I don't, they'll just keep you running in the administrative treadmill forever. If you don't mind not driving for the next 20 years before the Supreme Court takes a look at it, then you're free to do so.
I have deployed Airport Express devices on an enterprise network. 2 base stations easily power about 50 devices with RADIUS.
I recommend Apple Airport's as well. They are somewhat limited in feature but they are really stable, I have some that are over 10 years old that work perfectly (b/g only).
I also second Buffalo routers. A little more expensive than your el-cheapo home router but stable as hell and comes with DD-WRT.
Asus also makes some good ones but make sure to replace the wall-wart it comes (mine was a 2A) with with a slightly higher wattage/current (I used a 3A but measured usage spikes of 2.2A). The 2A that came in the box blew out about 9 months in and I suspect that's what happens with a lot of other routers too - they simply don't or only barely have the juice to power the radio under certain conditions and any line power issue reduces the voltage it supplies to the point the processor gets stuck, reboots or simply burns out. They also come under spec for quality routing this day and age. I still see routers shipping with 16 or 32MB of RAM and 200MHz processors and they're supposed to do gigabit routing on 4-8 ports and 600Mbps wireless, firewall, dhcpd, NAT, ... A simple routing and NAT table can fill that amount of RAM up in no time when a couple of people have torrents on.
In the city I'm at they don't even consider it a fine. They consider it a surcharge that you have to pay to the company that owns the red light camera's. The city probably gets a cut from it but you can't fight it in court because it's not a recognized fine under local or state laws. However if you don't pay the surcharge, they can still suspend your license because not paying the surcharge is recognized in the law. So you can go to court when they suspend your license during which proceedings (6 months for the next court date) you don't have a license (which in the US is practically a death sentence). You can then get your license back on a constitutional technicality (right to a speedy trial) but since the surcharge remains pending, every 12 months they'll re-suspend it.
What I meant is with a typical keyboard, the typical poll rate is ~250Hz on most OS HID drivers. You can increase that to ~1000Hz (either direct hardware access or alter the driver).
Yes, the keyboard has an inherent delay as well but there is no reason it should be as high as 40ms, even very slow microprocessors can scan the matrix much faster than that. In fact that would be fairly noticeable (40ms + OS latency). The typical modern keyboard I have tested to be about 10ms which includes it's own delay and the OS delay.