"Com-" is "together", and "prise" (from Latin prendere) is "take". "Comprise" therefore originally meant something along the lines of "bring together" or "unite". But once it entered English, it changed, naturally. Giraffedata looks a bit like the mythical version of King Canute, shouting at the waves to cease their advance.
Oui, bien sûr, mais ça c'est pour fr.wikipedia.org, pas pour en.wikipedia.org. If we were to take French usage as key, we'd have to root out the abuse of "vintage" as "old" rather than "abstract noun: relating to vinting (wine-making)", among a great many other things.
There are many common idioms that are used incorrectly in conversation or casual writing. But that doesn't mean they should be used in formal writing, such as an encyclopedia.
Well met, friend, for thou speakst great sooth! Many people have I encountered who are such dullards as to employ incorrectly the English tongue. 'Tis tragedy of the vtmost that the youth of our times know not how the language should properly speak itself. A gay fellow would I be were my fellow man to renew his acquaintance with the King's English.
Alas! but I must forsake thy gentle companie, for mine friends await me in a local hostelrie, and so must I away! Parting is such sweet sorry. Anon, good sir, anon!
I can imagine it now. "Right, so we can't claim sovereignty over the moon. What now?" "We're a republic." "So?" "We have no sovereign, and we sold off the national reserve, so there's no gold sovereigns either." (Open champagne, toast to sell...regulating the moon.
But pointing those out or contributing to correcting them would not be in your interest.
It's not about whether it's in my interest to correct them or not, it is about whether it's in my interest to watch the videos in the first place or not. I have no reason to watch the ones that I have the knowledge to correct, and I have no desire to watch ones that I can't determine the veracity of. Khan Academy's quality may have improved in the last couple of years, but the initial quality was poor enough to lose my interest completely.
It's a bunch of videos -- nothing more.
And that demonstrates irrefutably that you are just acting defensively and in doing so trying to spread FUD.
No it doesn't. It demonstrates that I am oversimplifying. Yes, there are questions and online code editors/checkers. Yes it assesses your knowledge and skips the odd video to personalise the learning track a little, but that's a poor approximation of adaptive learning at best. Yes it has a handful of teacher tools to allow you to integrate it into your own curriculum, but again, this is kind of lip-service as it is not really any more flexible than most offline materials -- and perhaps less so, because it's much easier to alter photocopiable spreadsheets than a web video.
Khan Academy got big not because of good pedagogy, but because of its price. It was created for free by a guy who didn't know a lot about teaching. He made the videos in an ad hoc manner that was quick and easy for him. Cheap production values for free materials. Sadly, the whole online learning sector has gone this way, and ploughed lots of money into making materials with cheap production values (see Coursera, edX etc). It's maddening. Why are all these screencasts and slidecasts touted as "the future of technology" when they're really just a professor-and-chalkboard lecture in a different medium? When I was a child, the UK's Open University had materials like that -- you'd see them on TV in the early morning writing equations on the board -- but they abandoned that and started using the medium of video in its own terms, borrowing more and more techniques from TV to convey information efficiently and effectively. Yes, that takes time, and it takes money. No, Sal couldn't have done that on his own. But the world has fetishised the lo-fi aesthetic, and no-one's willing to step away from it when the money's there to do something better. It's practically ludditism.
They were cheap, underpowered with low resolution screens and poor quality trackpads making them a race to the bottom competing only on price. They were only useful for the sorts of things that are easier on a tablet, everything else is better on an ultrabook which is why the market has expelled them.
I liked writing blog posts and student worksheets on the train. I wanted to code lightweight programs in Python and Javascript. I didn't need a heavy, full-sized laptop for that, but I certainly needed a keyboard. It happily played back the audio and video files I needed in class, and connected perfectly happily to any standard projector. That the market for netbooks is smaller than the market for tablets, I understand; however, the niche I was in was well-served by the eeePC, and in trying to embrace and extend customer appeal, they extinguished the netbook. That said, a reflashed Chromebook is an acceptable alternative.
Citation? Windows has been reducing its hunger for system resources
Windows has been reducing its hunger for system resources recently. Five years ago, when netbooks were the "next big thing", Windows XP's life was extended because they couldn't get Windows 7 squeezed into the specs of the netbooks at the time. Microsoft is shrinking the Windows footprint now because of the convergence of mobile and desktop, but that wasn't on the cards back then.
They were not useless -- I'd been waiting almost a decade for netbooks. The problem was bloatware -- MS apps had expanded to fill the vacuum of a much bigger computer... but why? I would love to see computing becoming more efficient, rather than algorithms abhoring a vacuum.
Yeuch. KA is riddled with errors and omissions. The idea of a "mastery curriculum" was not dreamt up by Sal Khan, and KA barely attempts to measure mastery anyway. It's a bunch of videos -- nothing more.
That's irrelevant. The limitarion is the teacher, not the whiteboard. I had a degree in Computer Science long before I moved into teaching, and I barely scratched the surface of what's possible with interactive whiteboards when I was teaching English....
Still, if the front of the queue doesn't have priority, you can't guarantee that a given task will ever be performed - you can play roulette 1000 times and still never see the ball land on number 1....
Yes, but that would mean using a much more powerful processor. The researchers have basically loaded their dice to roll snake-eyes often enough to win big, but not so often as to get chucked out of the casino.
The whole point is that they're claiming they can get throughput that is approaching best-case while keeping latency acceptable. It's just a matter of setting the probabilities such that one of the processors is likely to hit the head of the list in an acceptably short timeframe, which getting the probability of two trying to hit the head simultaneously as near to zero as possible.
As I've pointed out a few times -- this is a stochastic system, not a truly random one.
That's silly. SprayList only optimizes the case where the workload can be divided in independent jobs that can be executed out of order. Most real-life problems don't work that way.
Playing MP3, drawing the desktop, listening for new mail on POP, loading Slashdot, polling the keyboard... there's lots of things that my computer does at one time that are genuinely independent of each other. This is even more true of a multi-user cloud system. But more to the point, it's generally considered bad programming style to parallelise code that is inherently serial. If B relies on A, do A, then do B. Simple. If you need to thread them, then the threads should be self-syncronising, and if B needs data from A, it will give up the processor.
Then why did it still beat SprayList at 32+ cores? The only time SprayList beats the competition is at 16 cores.
The graph shows total throughput, but that isn't a perfect measure, because it ignores the delay to high priority tasks. If you used random scheduling on a games console, you would risk rockets freezing in midair, erratic screen refresh rates, sound-effects that were out of sync with the action, server lag, etc etc etc.
The researchers claim that SprayList is almost as good as non-prioritised scheduling, but that it respects priority enough that the worst side-effects of truly random scheduling are mitigated.
Rule number one about computer performance . . . know your workload!
I think of the algorithm in TFA as "Socratic computing performance" -- the only true knowledge about your workload is knowing that you know nothing about your workload. The more the computer has to introspect about individual tasks on the fly, the higher the computational load. Eventually you end up with the cost of optimisation being higher than the cost of the inefficiencies in the scheduling.
But assuming you have a known number of worker threads you could predistribute tasks in a round-robin fashion so that when thread #3 asks for a tasks it gets nextTask[3], off you go and we'll repopulate nextTask lazily once we've got lock on the main queue again, I assume if it's acceptable to pick from a range any one task will finish within an acceptable delay so that next task still gets done in time too. I wouldn't want to put a call to rand() anywhere in there.
It's not about assigning tasks to threads, though, it's about assigning threads to cores. There is no authority above the core, and so the cores have to manage their own scheduling. If you assign each core an index, then you no longer have a true queue, and if core 1 hangs, the head of the list will just sit there indefinitely as core 2 continues to take task 2, etc. And what about dynamic systems that power down cores. Do you reassign core indexes ever time a core shuts down? Now suddenly you're introducing more intercore coordination.
The thing to remember is that "random" isn't always "truly random". Think instead of it as "stochastic" in this case. The results are not exactly predictable, but give a predictable distribution of results. That those results include fewer resource clashes is a Very Good Thing.
If open was their goal, why did they go with components which require closed source drivers and firmware?
Not only this, but components that are not available in quantities of less than tens of thousands of units. The last attempt to make a genuine Raspberry Pi clone collapsed when they ran out of their "sample" batch of Broadcom SoCs and Broadcom refused to sell to them in the sort of quantities that they could have hoped to sell.
Here in the philippines, water is often kept cool by storing it in porus pots, the water slowly seeps through the walls of the clay pot, and evaporates from the outside, there are no channels as its using the micro pore structure of the earthen ceramic pot. The evaporation lowers the temperature of the whole pot.
Yes, but I'd bet good money your granny taught you to let the pot dry before refilling, or at least every two or three refills, because if it stays damp all the time, things will grow in it. Clay pots also have a tendency to get broken and replaced every now and then.
To do the same with this "cool brick" (dry your walls out completely), you're going to need to go without cooling for an extended period at the hottest time of the day, which is the time you most want cooling. Something of an intractable dilemma.
Slashdot Mirror
"Com-" is "together", and "prise" (from Latin prendere) is "take". "Comprise" therefore originally meant something along the lines of "bring together" or "unite". But once it entered English, it changed, naturally. Giraffedata looks a bit like the mythical version of King Canute, shouting at the waves to cease their advance.
Oui, bien sûr, mais ça c'est pour fr.wikipedia.org, pas pour en.wikipedia.org. If we were to take French usage as key, we'd have to root out the abuse of "vintage" as "old" rather than "abstract noun: relating to vinting (wine-making)", among a great many other things.
It's a common enough idiom.
There are many common idioms that are used incorrectly in conversation or casual writing. But that doesn't mean they should be used in formal writing, such as an encyclopedia.
Well met, friend, for thou speakst great sooth! Many people have I encountered who are such dullards as to employ incorrectly the English tongue. 'Tis tragedy of the vtmost that the youth of our times know not how the language should properly speak itself. A gay fellow would I be were my fellow man to renew his acquaintance with the King's English.
Alas! but I must forsake thy gentle companie, for mine friends await me in a local hostelrie, and so must I away! Parting is such sweet sorry. Anon, good sir, anon!
I can imagine it now. "Right, so we can't claim sovereignty over the moon. What now?" "We're a republic." "So?" "We have no sovereign, and we sold off the national reserve, so there's no gold sovereigns either." (Open champagne, toast to sell...regulating the moon.
But pointing those out or contributing to correcting them would not be in your interest.
It's not about whether it's in my interest to correct them or not, it is about whether it's in my interest to watch the videos in the first place or not. I have no reason to watch the ones that I have the knowledge to correct, and I have no desire to watch ones that I can't determine the veracity of. Khan Academy's quality may have improved in the last couple of years, but the initial quality was poor enough to lose my interest completely.
It's a bunch of videos -- nothing more.
And that demonstrates irrefutably that you are just acting defensively and in doing so trying to spread FUD.
No it doesn't. It demonstrates that I am oversimplifying. Yes, there are questions and online code editors/checkers. Yes it assesses your knowledge and skips the odd video to personalise the learning track a little, but that's a poor approximation of adaptive learning at best. Yes it has a handful of teacher tools to allow you to integrate it into your own curriculum, but again, this is kind of lip-service as it is not really any more flexible than most offline materials -- and perhaps less so, because it's much easier to alter photocopiable spreadsheets than a web video.
Khan Academy got big not because of good pedagogy, but because of its price. It was created for free by a guy who didn't know a lot about teaching. He made the videos in an ad hoc manner that was quick and easy for him. Cheap production values for free materials. Sadly, the whole online learning sector has gone this way, and ploughed lots of money into making materials with cheap production values (see Coursera, edX etc). It's maddening. Why are all these screencasts and slidecasts touted as "the future of technology" when they're really just a professor-and-chalkboard lecture in a different medium? When I was a child, the UK's Open University had materials like that -- you'd see them on TV in the early morning writing equations on the board -- but they abandoned that and started using the medium of video in its own terms, borrowing more and more techniques from TV to convey information efficiently and effectively. Yes, that takes time, and it takes money. No, Sal couldn't have done that on his own. But the world has fetishised the lo-fi aesthetic, and no-one's willing to step away from it when the money's there to do something better. It's practically ludditism.
They were cheap, underpowered with low resolution screens and poor quality trackpads making them a race to the bottom competing only on price. They were only useful for the sorts of things that are easier on a tablet, everything else is better on an ultrabook which is why the market has expelled them.
I liked writing blog posts and student worksheets on the train. I wanted to code lightweight programs in Python and Javascript. I didn't need a heavy, full-sized laptop for that, but I certainly needed a keyboard. It happily played back the audio and video files I needed in class, and connected perfectly happily to any standard projector. That the market for netbooks is smaller than the market for tablets, I understand; however, the niche I was in was well-served by the eeePC, and in trying to embrace and extend customer appeal, they extinguished the netbook. That said, a reflashed Chromebook is an acceptable alternative.
Citation? Windows has been reducing its hunger for system resources
Windows has been reducing its hunger for system resources recently. Five years ago, when netbooks were the "next big thing", Windows XP's life was extended because they couldn't get Windows 7 squeezed into the specs of the netbooks at the time. Microsoft is shrinking the Windows footprint now because of the convergence of mobile and desktop, but that wasn't on the cards back then.
I said "if", not "though". Microsoft's goal will be to mark the territory, one way or the other.
Unless you want people to believe "computer=Windows", in which case NOT having Windows is stupid.
You're just a filthy commie... no, wait, that's 16k too much... filthy speccy...! ;-)
They were not useless -- I'd been waiting almost a decade for netbooks. The problem was bloatware -- MS apps had expanded to fill the vacuum of a much bigger computer... but why? I would love to see computing becoming more efficient, rather than algorithms abhoring a vacuum.
Yeuch. KA is riddled with errors and omissions. The idea of a "mastery curriculum" was not dreamt up by Sal Khan, and KA barely attempts to measure mastery anyway. It's a bunch of videos -- nothing more.
Excellence != Khan Academy. Khan Academy == Acceptable mediocrity.
That's irrelevant. The limitarion is the teacher, not the whiteboard. I had a degree in Computer Science long before I moved into teaching, and I barely scratched the surface of what's possible with interactive whiteboards when I was teaching English....
Still, if the front of the queue doesn't have priority, you can't guarantee that a given task will ever be performed - you can play roulette 1000 times and still never see the ball land on number 1....
Yes, but that would mean using a much more powerful processor. The researchers have basically loaded their dice to roll snake-eyes often enough to win big, but not so often as to get chucked out of the casino.
The whole point is that they're claiming they can get throughput that is approaching best-case while keeping latency acceptable. It's just a matter of setting the probabilities such that one of the processors is likely to hit the head of the list in an acceptably short timeframe, which getting the probability of two trying to hit the head simultaneously as near to zero as possible.
As I've pointed out a few times -- this is a stochastic system, not a truly random one.
That's silly. SprayList only optimizes the case where the workload can be divided in independent jobs that can be executed out of order. Most real-life problems don't work that way.
Playing MP3, drawing the desktop, listening for new mail on POP, loading Slashdot, polling the keyboard... there's lots of things that my computer does at one time that are genuinely independent of each other. This is even more true of a multi-user cloud system. But more to the point, it's generally considered bad programming style to parallelise code that is inherently serial. If B relies on A, do A, then do B. Simple. If you need to thread them, then the threads should be self-syncronising, and if B needs data from A, it will give up the processor.
Then why did it still beat SprayList at 32+ cores? The only time SprayList beats the competition is at 16 cores.
The graph shows total throughput, but that isn't a perfect measure, because it ignores the delay to high priority tasks. If you used random scheduling on a games console, you would risk rockets freezing in midair, erratic screen refresh rates, sound-effects that were out of sync with the action, server lag, etc etc etc.
The researchers claim that SprayList is almost as good as non-prioritised scheduling, but that it respects priority enough that the worst side-effects of truly random scheduling are mitigated.
Rule number one about computer performance . . . know your workload!
I think of the algorithm in TFA as "Socratic computing performance" -- the only true knowledge about your workload is knowing that you know nothing about your workload. The more the computer has to introspect about individual tasks on the fly, the higher the computational load. Eventually you end up with the cost of optimisation being higher than the cost of the inefficiencies in the scheduling.
But assuming you have a known number of worker threads you could predistribute tasks in a round-robin fashion so that when thread #3 asks for a tasks it gets nextTask[3], off you go and we'll repopulate nextTask lazily once we've got lock on the main queue again, I assume if it's acceptable to pick from a range any one task will finish within an acceptable delay so that next task still gets done in time too. I wouldn't want to put a call to rand() anywhere in there.
It's not about assigning tasks to threads, though, it's about assigning threads to cores. There is no authority above the core, and so the cores have to manage their own scheduling. If you assign each core an index, then you no longer have a true queue, and if core 1 hangs, the head of the list will just sit there indefinitely as core 2 continues to take task 2, etc. And what about dynamic systems that power down cores. Do you reassign core indexes ever time a core shuts down? Now suddenly you're introducing more intercore coordination.
The thing to remember is that "random" isn't always "truly random". Think instead of it as "stochastic" in this case. The results are not exactly predictable, but give a predictable distribution of results. That those results include fewer resource clashes is a Very Good Thing.
But that's neither a rational or logical objection. It's a biased objection born of ignorance.
I agree, but we have to function in the world we live in, and selling to the ignorant is easier than educating the whole world.
If open was their goal, why did they go with components which require closed source drivers and firmware?
Not only this, but components that are not available in quantities of less than tens of thousands of units. The last attempt to make a genuine Raspberry Pi clone collapsed when they ran out of their "sample" batch of Broadcom SoCs and Broadcom refused to sell to them in the sort of quantities that they could have hoped to sell.
Understanding sarcasm is clearly your strong point.
Logically, water can be considered abundant wherever rain is imminent, and under new US definitions of imminent, water is abundant everywhere.
Here in the philippines, water is often kept cool by storing it in porus pots, the water slowly seeps through the walls of the clay pot, and evaporates from the outside, there are no channels as its using the micro pore structure of the earthen ceramic pot. The evaporation lowers the temperature of the whole pot.
Yes, but I'd bet good money your granny taught you to let the pot dry before refilling, or at least every two or three refills, because if it stays damp all the time, things will grow in it. Clay pots also have a tendency to get broken and replaced every now and then.
To do the same with this "cool brick" (dry your walls out completely), you're going to need to go without cooling for an extended period at the hottest time of the day, which is the time you most want cooling. Something of an intractable dilemma.