100% code coverage is considered a code smell. Almost always means you've forced it and probably aren't doing proper testing. The masters tell the laymen to strive for 100%, but they also face-palm when they see it actually "accomplished."
People with strong abstract reasoning make use of nearly all parts of their brain at the same time and the different parts are more highly connected to each other that people with lower abstract reasoning. It was difficult to see this with older brain scans because high levels of abstract reasoning cause larger neural-activity bursts but of shorter duration. One of the first parts of the brain to connect to the frontal lobe is the visual cortex. It is not far fetched to think that people visualizing logical problems is probably a side effect of the visual cortex helping with the processing of the logic. Below a certain threshold, the different brain regions are quite isolated. Once this inflection point is reached, it snowballs very quickly, reinforcing itself.
Most of this is from research in the past few years from advancements with brain scans. There has been a large surge in research in relation to abstract reasoning and meta-cognition, which is starting to look like a power curve with a long tail. To make matters worse, there seems to be no way to teach someone else abstract reasoning. Learning abstract reasoning above the norm seems to be a self taught. Most people haphazardly acquire some abstract reasoning skills, but above a certain meta-cognition level, people can purposefully exercise abstract reasoning entirely by reflecting on their own thoughts. Some propose that reflecting on thoughts increases the interconnections among the brain regions, allowing for better pattern matching from the different regions working in unison instead of individually.
There have been some experiments that showed certain mental exercises could increase one's ability to score higher on a particular abstract reasoning test, but the benefit did not transfer to other abstract reasoning tests, meaning the mental exercise made the person more efficient that particular test, but not actually better at abstract reasoning. Like I said, no known way to increase abstract reasoning except introspection.
They'll still be programmed. Programming is the breakdown of a problem into it's logical ornithological pieces. Coding is what will go away, and I can't wait. Even if you get rid of coding, you will still have the problem that most people can't program. Training is statistical in nature. Great when you want statistical results. You still have the issue of being able to determine if the data you have is actually representative of the results you want.
I've worked enough with customers, and a lot of other programmers, and people typically can't clearly describe what they want done. They will have all kinds of contradictory or logically impossible requirements.
Internet traffic is mostly broken up into two categories of traffic, Transit, which is expensive because you need to move the data all around the world, and non-transit peering, which is almost always free because it benefits both networks and it gets rid of the transit costs for both of them. Verizon, for example, was trying to get Netflix to pay 10x the going rate of transit, for non-transit peering.
This put Netflix in quite the pickle. Level 3 wanted to use local peering with Verizon because Level 3 and Verizon had a transit peering agreement to keep both directions fairly even. Dumping Netflix on these links would skew the ratio. Level 3 wanted to do cold-potato routing and drop the data on Verizon's network near where the data was going to be consumed, but Level 3 expected the peering to be free or possible just to pay for the equipment, as is the norm.
To use Verizon's own logic, they should be paying their customers. If Verizon wants access to their customer's networks, Verizon should pay. Of course it's stupid because Verizon is providing the service and the customers should be the ones paying for the service. Same thing. Level 3 is providing the service and Verizon should be jumping up and down about reducing their operating costs.
The problem is Verizon is actually broken up into more than one company. One part is the residential ISP and the other is the business transit provider. Normally ISPs want to reduce their transit costs by peering. In this case, the Verizon residential asked its sister company to flex its tier 1 status. This is a conflict of interest. Normally the ISP side of things pays the transit, but because they're really the same company, they get to have their cake and eat it to, the best of both worlds with none of the downsides.
Said very simplistically. Unless you plan to cache their entire 1PiB+ catalog, you are going to need to make sure your caching strategy has enough storage to be useful. Let say you need 100Gb/s of Netflix bandwidth, and you want to reduce that by 80%. You will need 80Gb/s of bandwidth, plus 100TiB of storage per server. One of the newest Netflix server can handle nearly 80Gb/s, but they're SSD and don't have 100TiB of storage. Lets say you have their 100TiB server but it can only handle 20Gb/s. Now you need 4 of these, which means 400TiB of storage.
But you mentioned a VM an ISP could get. Now each server can only handle 10Gb/s, so now you need 8 servers, each with 100TiB of storage. Are you starting to see the issue? Netflix is pushing the boundary of IO. Of course there are other ways this problem could be solved, and it could be standardized, but that requires the entire video stream industry to accept a standard and architecture everything to work with that standard. Talking many years of work just to break even with what we have today. but now everyone is tied to a standard and it makes it difficult to innovate how to architect the caching systems, which are in constant flux as new designs are constantly being tested and tweaked.
You have two choices. 1) Throw money at the problem by using generic services 2) Use highly customized designs that are 10x+ more efficient, but no sharing
A third pseudo-option is to make the problem mature more quickly by throwing R&D money at it with the intention of creating a standard that will scale and be useful for the next N years where N is a useful amount of time to invest into a standard.
They primarily use Level 3, which is for the most part, one of the least evil Tier 1 providers. They did use Cogent, but under duress. They had many situations where the ISP refused to peer with Level 3 for Netflix traffic, but would allow Cogent. Cogent is one of the most evil. They abused this relationship by having Netflix being higher priority than many of their other services, so a large degradation of non-Netflix services would cause customers to call and complain. Level 3, on the other hand, kept Netflix traffic mostly separate.
My biggest complaint about Netflix on Level 3 is receiving 40Gb/s microbrusts from non-paced TCP streams that average 8Mb/s. There is virtually zero congestion between my home connection and Netflix via Level 3.
Use them unlimited cellphone minutes to call and complain every time you have issues. The average rate of an ISP's callcenter is $3/minute. 20 minutes a month will cost them $60/m. Incentive them to fix their network.
You can't use generic caching for the request patterns Netflix sees. They have spent a lot of time customizing their caching, which is not even dynamic in nature, but static, to reduce cache thrashing. Netflix' infrastructure is an optimized single vision of how everything should work. The software is architected to work well with the hardware and networks, the hardware is architected to work well with the software and network, and the networks are architected to work well with the hardware and software.
There is almost nothing reusable about their design, it is highly optimized exactly for their use case.
If you're using a generic dynamic cache, a 100TiB cache would see less than a 50% hit rate with Netflix. While a 100TiB Netflix optimized cache will see in the 80-90% range, and a 10TiB Netflix optimized cache would see about 50%.
Since there is no way to dynamically determine what data to cache, all you can do is statically partition the data by video streaming provider, and give those provided access to how they want to cache the data. Defeats the entire purpose.
Your primary argument is sound, but what is this about Argon being "bad"? It's a noble gas, it is inert. It doesn't really do anything except exist, unless you'e trying to turn it into a plasma.
My employer recently switched to a system where you can only rollover 5 days from the last year at most. They used to have an unlimited roll-over at an accumulation of 50% per year, but they said most employees were only using the bear minimum amount of vacation. Something about "all work and no play" as their reasoning for limiting how much you can roll-over. Heck, we even get our birthdays as a yearly floating personal paid holiday.
All in all, 12 paid holidays, 24 personal vacation days, and 5 sick-days. And I get comped for overtime with more vacation time.
I regularly hit 1TiB/week from streaming and it's just my wife and I at home, and we don't watch that much. 20Mbit/s streams tear up bandwidth. We also like to stream while we're away from home so the cats have some noise. They have separation anxiety and the noise helps. For $99.99/m, I could get a dedicated 500/500 fiber connection. Guaranteed to get that rate to my ISP's transit provider 24x7. I prefer to save money and only have 150/150 for $50, $52.75 after taxes and fees, NOT promo and out of contract.
My point is you have to program to do that, which requires a much deeper understanding of the problem that is almost always ignored. A human that is incapable of adjusting for their own mistakes will not be able to create software that can adjust for its mistakes.
I appreciate what you're saying, but programming is not complex. You use simple patterns to create complex design fractals. If you're creating complex designs, you're doing it wrong. As one person described it to me, "You run a Fourier transform over an idea and break it down into simple fundamental parts". In my experience, the technical side of a programming domain can be mastered in 2-12 weeks.
I was hand-rolling lock-less data structures in production with only a few weeks experience in programming. I have independently discovered several concurrency bugs in.Net, although I was not the first to have noticed, most have been fixed. When I write my code, I write it with the intention of being bug free. When me or someone else discovers a bug, I don't fire up the debugger, I think about the problem, then I read my code, then I fix the bug. The debugger is a crutch that I prefer to never use. I design my code to no require using the debugger. My code works exactly how I want it to work, and when it doesn't there's not many reasons why it doesn't. This is how I discovered the.Net bugs, not because I was able to reproduce them, but because I could not explain them with my own code, so the problem had to be outside my code and work a very specific way.
I have turned writing code and git branching into a game. How clean, understandable, and informative of intentions can I make the code and commits. This is an art that only experience can improve.
Yes, but as a human, you can reprogram yourself and adjust for mistakes you or others made. Computers cannot do this. Not to mention that with computers, all cases can be accounted for, which cannot be done in the real world. A computer is a fast idiot that does exactly what you say. You cannot treat a computer like a human.
Classic economics would say that if there's a business in which there are 35 percent net margins, that would attract a huge amount of new capital to capture some of that, and none of that has happened.
Sounds like someone found out what talented individuals can bring to the table. The competition cannot compete by reducing costs, they must become technologically on par or superior to compete. Google is not perfect, but they focus more on quality than nearly anyone else. Never give up quality for speed. When you focus on quality, you get speed for free.
Like I said, they're not perfect, they're just much better than the alternatives, and in many cases, quite good.
Contracting something out doesn't negate any responsibility you take for the deliverable. "I didn't do the work, therefore I am not responsible" even if you're the one responsible to make sure it gets done?
The budget does not reflect actual military spending. There are many ways for the government to spend several times the budgeted amount. One such way is "war time" expenditures. Spending is almost not even questioned as they have to happen.
I'm paid to do a job, not take orders. They tell me what job they want done and I do it my way. They hired a a professional, not some "yes man". Being professional sometimes means standing up for what's right. If you want to be another disposable cog, be my guest. The nail that stands out gets hammered the hardest, but it also gets recognized.
I remember reading that girls are showing biases the day they're born. The bias is highly correlated with the amount of maternal testosterone during fetal development. Of course boys tend to have higher testosterone levels. If you look at just testosterone levels, men and women have the same interests. Maybe we need to remove the testosterone bias to level the field?
You can't solve a problem without first understanding the problem, and the AI can't understand the problem better than the programmer by inferring what the programmer may be doing. I could see an AI making suggestions or advanced forms of refactoring, but not out-right making useful stand-alone code. I could be biased because I mostly deal with hyper-optimizing high level language systems.
Slashdot Mirror
Yay, crashing at run time for something that could have been trivially tested for at compile time.
100% code coverage is considered a code smell. Almost always means you've forced it and probably aren't doing proper testing. The masters tell the laymen to strive for 100%, but they also face-palm when they see it actually "accomplished."
People with strong abstract reasoning make use of nearly all parts of their brain at the same time and the different parts are more highly connected to each other that people with lower abstract reasoning. It was difficult to see this with older brain scans because high levels of abstract reasoning cause larger neural-activity bursts but of shorter duration. One of the first parts of the brain to connect to the frontal lobe is the visual cortex. It is not far fetched to think that people visualizing logical problems is probably a side effect of the visual cortex helping with the processing of the logic. Below a certain threshold, the different brain regions are quite isolated. Once this inflection point is reached, it snowballs very quickly, reinforcing itself.
Most of this is from research in the past few years from advancements with brain scans. There has been a large surge in research in relation to abstract reasoning and meta-cognition, which is starting to look like a power curve with a long tail. To make matters worse, there seems to be no way to teach someone else abstract reasoning. Learning abstract reasoning above the norm seems to be a self taught. Most people haphazardly acquire some abstract reasoning skills, but above a certain meta-cognition level, people can purposefully exercise abstract reasoning entirely by reflecting on their own thoughts. Some propose that reflecting on thoughts increases the interconnections among the brain regions, allowing for better pattern matching from the different regions working in unison instead of individually.
There have been some experiments that showed certain mental exercises could increase one's ability to score higher on a particular abstract reasoning test, but the benefit did not transfer to other abstract reasoning tests, meaning the mental exercise made the person more efficient that particular test, but not actually better at abstract reasoning. Like I said, no known way to increase abstract reasoning except introspection.
They'll still be programmed. Programming is the breakdown of a problem into it's logical ornithological pieces. Coding is what will go away, and I can't wait. Even if you get rid of coding, you will still have the problem that most people can't program. Training is statistical in nature. Great when you want statistical results. You still have the issue of being able to determine if the data you have is actually representative of the results you want.
I've worked enough with customers, and a lot of other programmers, and people typically can't clearly describe what they want done. They will have all kinds of contradictory or logically impossible requirements.
Parent is Goatse.cx. Please down vote.
Internet traffic is mostly broken up into two categories of traffic, Transit, which is expensive because you need to move the data all around the world, and non-transit peering, which is almost always free because it benefits both networks and it gets rid of the transit costs for both of them. Verizon, for example, was trying to get Netflix to pay 10x the going rate of transit, for non-transit peering.
This put Netflix in quite the pickle. Level 3 wanted to use local peering with Verizon because Level 3 and Verizon had a transit peering agreement to keep both directions fairly even. Dumping Netflix on these links would skew the ratio. Level 3 wanted to do cold-potato routing and drop the data on Verizon's network near where the data was going to be consumed, but Level 3 expected the peering to be free or possible just to pay for the equipment, as is the norm.
To use Verizon's own logic, they should be paying their customers. If Verizon wants access to their customer's networks, Verizon should pay. Of course it's stupid because Verizon is providing the service and the customers should be the ones paying for the service. Same thing. Level 3 is providing the service and Verizon should be jumping up and down about reducing their operating costs.
The problem is Verizon is actually broken up into more than one company. One part is the residential ISP and the other is the business transit provider. Normally ISPs want to reduce their transit costs by peering. In this case, the Verizon residential asked its sister company to flex its tier 1 status. This is a conflict of interest. Normally the ISP side of things pays the transit, but because they're really the same company, they get to have their cake and eat it to, the best of both worlds with none of the downsides.
Caching video is caching video
Said very simplistically. Unless you plan to cache their entire 1PiB+ catalog, you are going to need to make sure your caching strategy has enough storage to be useful. Let say you need 100Gb/s of Netflix bandwidth, and you want to reduce that by 80%. You will need 80Gb/s of bandwidth, plus 100TiB of storage per server. One of the newest Netflix server can handle nearly 80Gb/s, but they're SSD and don't have 100TiB of storage. Lets say you have their 100TiB server but it can only handle 20Gb/s. Now you need 4 of these, which means 400TiB of storage.
But you mentioned a VM an ISP could get. Now each server can only handle 10Gb/s, so now you need 8 servers, each with 100TiB of storage. Are you starting to see the issue? Netflix is pushing the boundary of IO. Of course there are other ways this problem could be solved, and it could be standardized, but that requires the entire video stream industry to accept a standard and architecture everything to work with that standard. Talking many years of work just to break even with what we have today. but now everyone is tied to a standard and it makes it difficult to innovate how to architect the caching systems, which are in constant flux as new designs are constantly being tested and tweaked.
You have two choices. 1) Throw money at the problem by using generic services 2) Use highly customized designs that are 10x+ more efficient, but no sharing A third pseudo-option is to make the problem mature more quickly by throwing R&D money at it with the intention of creating a standard that will scale and be useful for the next N years where N is a useful amount of time to invest into a standard.
They chose shitty Tier1 ISPs for a reason
They primarily use Level 3, which is for the most part, one of the least evil Tier 1 providers. They did use Cogent, but under duress. They had many situations where the ISP refused to peer with Level 3 for Netflix traffic, but would allow Cogent. Cogent is one of the most evil. They abused this relationship by having Netflix being higher priority than many of their other services, so a large degradation of non-Netflix services would cause customers to call and complain. Level 3, on the other hand, kept Netflix traffic mostly separate.
My biggest complaint about Netflix on Level 3 is receiving 40Gb/s microbrusts from non-paced TCP streams that average 8Mb/s. There is virtually zero congestion between my home connection and Netflix via Level 3.
Netflix consumes on average 8Mb/s/stream for devices at my home. It's downgrading the bit-rate for your connection.
Use them unlimited cellphone minutes to call and complain every time you have issues. The average rate of an ISP's callcenter is $3/minute. 20 minutes a month will cost them $60/m. Incentive them to fix their network.
You can't use generic caching for the request patterns Netflix sees. They have spent a lot of time customizing their caching, which is not even dynamic in nature, but static, to reduce cache thrashing. Netflix' infrastructure is an optimized single vision of how everything should work. The software is architected to work well with the hardware and networks, the hardware is architected to work well with the software and network, and the networks are architected to work well with the hardware and software.
There is almost nothing reusable about their design, it is highly optimized exactly for their use case.
If you're using a generic dynamic cache, a 100TiB cache would see less than a 50% hit rate with Netflix. While a 100TiB Netflix optimized cache will see in the 80-90% range, and a 10TiB Netflix optimized cache would see about 50%.
Since there is no way to dynamically determine what data to cache, all you can do is statically partition the data by video streaming provider, and give those provided access to how they want to cache the data. Defeats the entire purpose.
Your primary argument is sound, but what is this about Argon being "bad"? It's a noble gas, it is inert. It doesn't really do anything except exist, unless you'e trying to turn it into a plasma.
My employer recently switched to a system where you can only rollover 5 days from the last year at most. They used to have an unlimited roll-over at an accumulation of 50% per year, but they said most employees were only using the bear minimum amount of vacation. Something about "all work and no play" as their reasoning for limiting how much you can roll-over. Heck, we even get our birthdays as a yearly floating personal paid holiday.
All in all, 12 paid holidays, 24 personal vacation days, and 5 sick-days. And I get comped for overtime with more vacation time.
I regularly hit 1TiB/week from streaming and it's just my wife and I at home, and we don't watch that much. 20Mbit/s streams tear up bandwidth. We also like to stream while we're away from home so the cats have some noise. They have separation anxiety and the noise helps. For $99.99/m, I could get a dedicated 500/500 fiber connection. Guaranteed to get that rate to my ISP's transit provider 24x7. I prefer to save money and only have 150/150 for $50, $52.75 after taxes and fees, NOT promo and out of contract.
My point is you have to program to do that, which requires a much deeper understanding of the problem that is almost always ignored. A human that is incapable of adjusting for their own mistakes will not be able to create software that can adjust for its mistakes.
I appreciate what you're saying, but programming is not complex. You use simple patterns to create complex design fractals. If you're creating complex designs, you're doing it wrong. As one person described it to me, "You run a Fourier transform over an idea and break it down into simple fundamental parts". In my experience, the technical side of a programming domain can be mastered in 2-12 weeks.
.Net, although I was not the first to have noticed, most have been fixed. When I write my code, I write it with the intention of being bug free. When me or someone else discovers a bug, I don't fire up the debugger, I think about the problem, then I read my code, then I fix the bug. The debugger is a crutch that I prefer to never use. I design my code to no require using the debugger. My code works exactly how I want it to work, and when it doesn't there's not many reasons why it doesn't. This is how I discovered the .Net bugs, not because I was able to reproduce them, but because I could not explain them with my own code, so the problem had to be outside my code and work a very specific way.
I was hand-rolling lock-less data structures in production with only a few weeks experience in programming. I have independently discovered several concurrency bugs in
I have turned writing code and git branching into a game. How clean, understandable, and informative of intentions can I make the code and commits. This is an art that only experience can improve.
Yes, but as a human, you can reprogram yourself and adjust for mistakes you or others made. Computers cannot do this. Not to mention that with computers, all cases can be accounted for, which cannot be done in the real world. A computer is a fast idiot that does exactly what you say. You cannot treat a computer like a human.
That's how the Earth got destroyed to make way for a space highway.
Classic economics would say that if there's a business in which there are 35 percent net margins, that would attract a huge amount of new capital to capture some of that, and none of that has happened.
Sounds like someone found out what talented individuals can bring to the table. The competition cannot compete by reducing costs, they must become technologically on par or superior to compete. Google is not perfect, but they focus more on quality than nearly anyone else. Never give up quality for speed. When you focus on quality, you get speed for free.
Like I said, they're not perfect, they're just much better than the alternatives, and in many cases, quite good.
Contracting something out doesn't negate any responsibility you take for the deliverable. "I didn't do the work, therefore I am not responsible" even if you're the one responsible to make sure it gets done?
The budget does not reflect actual military spending. There are many ways for the government to spend several times the budgeted amount. One such way is "war time" expenditures. Spending is almost not even questioned as they have to happen.
I'm paid to do a job, not take orders. They tell me what job they want done and I do it my way. They hired a a professional, not some "yes man". Being professional sometimes means standing up for what's right. If you want to be another disposable cog, be my guest. The nail that stands out gets hammered the hardest, but it also gets recognized.
I remember reading that girls are showing biases the day they're born. The bias is highly correlated with the amount of maternal testosterone during fetal development. Of course boys tend to have higher testosterone levels. If you look at just testosterone levels, men and women have the same interests. Maybe we need to remove the testosterone bias to level the field?
You can't solve a problem without first understanding the problem, and the AI can't understand the problem better than the programmer by inferring what the programmer may be doing. I could see an AI making suggestions or advanced forms of refactoring, but not out-right making useful stand-alone code. I could be biased because I mostly deal with hyper-optimizing high level language systems.