No. You're wrong. Passing by reference is slower when you cannot analyse aliasing, but pass by copy allows CSE.
And again, no. Programmer efficiency is directly related to the expressiveness of the language. There tends to be a trade-off between expressiveness and efficiency.
So you know two imperative languages. Do you really think that you are qualified to talk about the range of possible languages?
So language semantics have nothing to do with performance eh? Oh dear. Your experience is clearly much more limited than you estimate it to be. There's a big old world of languages and implementation techniques out there that it would benefit you to learn about.
From a comment like that you clearly have no idea of the impact on using pass-by-copy or pass-by-reference semantics in a language. Furthermore you are unable to establish what sort of features would make the pass-by-copy faster as it is more amenable to analysis and optimisation. Do you know the difference between static and dynamic despatch in a language, and the performance costs of each? Do you where closures and co-routines can improve performance if the language provides decent support for the programmer to use them?
Lastly there is the issue of programmer efficiency, as opposed to machine efficiency. If I can code up a problem solution in 1/10 of the time using a higher level language, then the relative efficiency of my high level language and a low level language become very important. People who have learnt more expressive languages than Pascal and C can write code in them much more quickly, and concisely than a low level implementation. For those who are more experienced that you are, the trade-off between verbosity and performance is very important.
Well.... yes and no. Chess is not that intractable. It is just beyond the current horizon of what we can compute. Before I started writing this comment I had thought the size of the game-tree was conjectured to be about 2^120 which would put it in the sun burning out category. But then I stumbled across this reference on google:
Jurg Nievergelt, of ETH Zurich, quoted the number 2^70 (or about 10^21) in e-mail, and referred to his paper "Information content of chess positions", ACM SIGART Newsletter 62, 13-14, April 1977, to be reprinted in "Machine Intelligence" (ed Michie), to appear 1990.
Now 2^70 is not that intractable at all. Let assume that you're going to build a *large* machine. 2^20 processors is quite doable. Each of these processors can perform 2^32 operations per second without getting to far ahead of what we have now. That's 2^52 operations per second. Or about 2^18 seconds, roughly 2 days +/- some fudge factor. That fudge factor is how many atomic operations it takes to evaluate a position. On a commodity machine it will be a large number, on a custom architecture (like Deep Blue) it will be very low.
So brute-forcing chess is currently on the horizon of what can be done now. Memory for the game tree would be more of an issue than processing speed. Because we can loop through the same position the game tree is really a dag, so it is not feasible to compute local solutions to sub-trees. As a (very) loose estimate of the lower bound on the memory required: assume an average branching factor of 32, each position can be stored in 70 bits, so 70*(32+1) bits per tree node is about 2^11 bits, so the tree would need 2^81 bits. That is a hell of a lot, but memory densities are increasing exponentially as well. I don't know how big the Googleplex is, but researchers are starting to consider exabyte-sized systems. It will probably be done by somebody in 10-20 years.
As far as the living breathing human part goes, I think that was the OP's point. We can solve the game with a "dumb" approach, so it is not amongst the hardest of problems (those that we can't evaluate solutions to effectively). Whether or not that "dumb" approach is effective depends entirely where you are on the current exponential curve of processing power.
PS I mean MaskedSlacker's point, not the OP. I can see that the AC he was answering was actually using trivial in completely the wrong sense.
As a correction, can you really have missed all of media noise around the equation that brought down Wall Street? That's a pretty big example of a broken model that directly led to a lot of the current chaos, rather than people feeding garbage into a good model.
I was actually supposed to meta-mod this comment but decided to reply instead. You don't understand what trivial means. It doesn't mean that it's "easy" as in a low number of computational steps. It means that there are no complicated decisions in the algorithm. For brute-forcing a game this is true, you simply evaluate every possible response to every possible until you reach the leafs of the game tree. So it is a trivial algorithm even though it is computationally intractable.
But how would ethernet between the display and the device accessing the network help in that? If the device is a computer, or an XBox or whatever, then it can view youtube and then display the video over HDMI. Having IP between the display and the device using the display doesn't seem to serve any purpose that I can think of but one*.
* And I don't like that purpose because it is a back-channel for DRM authentication.
Why Ethernet? Can you think of any devices that connect to your home theater (Game console, DVR, etc.), that have video, audio, and ethernet? Here is a hint.. I just named two. Can you look forward and see how more devices in the future will have network connectivity? I sure hope so.
And what is the point of having a network channel between these devices and a *display* ? As the GP asked, why do I need ethernet on my display device?
I didn't claim that it would be ideal. The description of noop suggests that it leaves most of the logic up to the drive controller - which is an improvement for SSD over scheduler logic tailored to a mechanical drive.
So given the massive erase blocks what is the best scheduler for an SSD?
You've quoted the marketing fluff from the article about what Microsoft says TRIM support in Windows 7 will achieve. Do you think that this is a demonstration that you understand TRIM?
I'd refer you to the link higher up the thread. Now it's a hell of a long article, but at least it explains what TRIM is. It allows blocks to be invalidated on the drive directly. Without waiting for them to be overwritten. Note that this explanation is two short sentences and explains *exactly* what TRIM is. Your quote is a marketing attempt to explain what TRIM will achieve.
So the noop scheduler would be the correct choice for a drive that supports TRIM, as the GP claimed. Although the scheduler itself will still need direct support for sending TRIM commands to the storage.
It's strange that you think that - real (actual physical) whiteboards are great for engineers and scientists. There is still a division between people who are happy thrashing ideas out on a board, and those who feel that they need to nurture something in private. I have to ask whether your comments about the environment being poor for research and insight are limited to virtual whiteboards or if they extend to their real-life counterparts?
So where do you think virtual whiteboards break down? Is it the lack of associated physical cues, or just the terrible interfaces that they ship with?
Sure, a lot of references amongst programmers point to these books - but that doesn't mean that these books provide pointers towards programming. I'm sure that the difference between bijection and implication is something that needs to be learnt during a programming course...:)
A lot of the books that you recommend are good, but have nothing to do with programming. I think you should change the word "essential" to stop people getting the wrong idea about how relevant most of the fiction is.
I don't think everyone can do it, but it probably isn't that uncommon amongst the young. I could hear TVs and other electronics until my early 20s and then it faded away along with the rest of my hearing:) It does sound like your hearing is exceptionally acute though.
Errr.... so close and yet so far? Why not change that so that programs are recipes - closer to the actual truth and then they gain some understanding that the recipe controls what gets done with the oven.
Or, if you want to stick to the programs as tools metaphor then make the chef the CPU. The oven is just another tool, and you've made a weird distinction between one tool, and all the other tools in the kitchen just to shoehorn it in.
You're not comparing like with like. When you have 5-6 tabs open it is as you describe. But these proposals are aimed at users with 50-60 tabs. If you fire open that many tabs then how wide is each one? 20 pixels? So a fairer comparison would be which dimension do you want to lose 20 pixels from.
If you have a 16:9 screen then this is an easy choice. For 4:3 it is somewhat harder, but most websites are mostly text, and text flows better in narrow columns than in wide ones.
This Add-on is brilliant. I found the link yesterday when this story came up on another site - after just 24-hours I don't think I could go back to a browser that doesn't work like this.
Some of proposed ideas are nice, the one that I would really like to see would integrate the tab tree with the browser bookmarks to make something like the Dock on OS-X. Pages could be pinned into the tree, if they are not open then they act like a bookmark and launch when you click them. If they are open then clicking switches to that tab.
Come on Mozilla, please clone the Dock...
PS If you change your Google search prefs to open results in a new window, your browser prefs to open new windows in a tab, and use this extension - it is nirvana.
I've wondered how much the side effects of sleep deprivation change between different people. I would guess that I'm more sensitive to these side-effects than most - I start getting minor auditory hallucinations after being awake for 16+ hours (not much more than a normal day), and visual effects about 20+ hours in. But then I don't think I could survive more than a couple of days, I hit a really hard brick wall at about 40 hours and can't stay awake. By that stage I've already gone through the temperature changes that they describe in the article.
Although it is a hellish way to work there is something to be said for not having to pick up context repeatedly in a problem. A 24-hr stretch in the office seems to produce about as much work as a standard 40-hr week.
Then you don't understand how life as an academic actually works. There are duties that consume a fixed amount amount of time: lecturing, marking, contact time, admin etc. Research then consumes whatever is left. By its nature it is hard to measure the productiveness of research - but the salary being paid to an academic covers this activity.
Now, how many academics do you know that have written books? In every case that I can think of this "in addition to" was carried out during office hours as well as at home, and was a direct replacement for the research activities that would normally have occupied that time.
It is all very well having specific job requirements - but they don't mean much when one of the requirements is for an amorphous ill-defined activity that naturally saturates all time that could be allocated to it.
Actually, as an academic I know exactly what I'm talking about. Nobody writes textbooks for money - it does not pay. And we have enough flexibility in our work to choose to write books in the first place. Try spending a large fraction of a year working on a textbook instead of normal duties in any other career than academia and come back and tell me that it is not something you get paid for.
You've missed a third case - the author was paid a salary while he wrote the book. Does he then deserve any further payment? Peter Wayner (the author and submitter) is (I'm guessing) an academic. So he would have been paid a salary during the years that he wrote the book. He didn't need a fee on completion because it had already been paid.
Like many other readers I plugged "data compression wayner" into Google to see which book it is. In my case I was wondering if it is the same text sitting on my boss' desk that I've been planning to buy.
I can't actually find your book anymore. There are hundreds of results with your question, and thousands more spam farm sites that have sprung up because this is suddenly a popular search term.
As a result, all information on getting your book (either legitimately or otherwise) has been drowned under a sea of crap. I would have liked to know if you have an electronic copy available for a reasonable price. The MacKay text linked to above is now sitting on my harddrive because it is easy to find and a legitimate copy is freely available.
Slashdot Mirror
No. You're wrong. Passing by reference is slower when you cannot analyse aliasing, but pass by copy allows CSE.
And again, no. Programmer efficiency is directly related to the expressiveness of the language. There tends to be a trade-off between expressiveness and efficiency.
So you know two imperative languages. Do you really think that you are qualified to talk about the range of possible languages?
So language semantics have nothing to do with performance eh? Oh dear. Your experience is clearly much more limited than you estimate it to be. There's a big old world of languages and implementation techniques out there that it would benefit you to learn about.
From a comment like that you clearly have no idea of the impact on using pass-by-copy or pass-by-reference semantics in a language. Furthermore you are unable to establish what sort of features would make the pass-by-copy faster as it is more amenable to analysis and optimisation. Do you know the difference between static and dynamic despatch in a language, and the performance costs of each? Do you where closures and co-routines can improve performance if the language provides decent support for the programmer to use them?
Lastly there is the issue of programmer efficiency, as opposed to machine efficiency. If I can code up a problem solution in 1/10 of the time using a higher level language, then the relative efficiency of my high level language and a low level language become very important. People who have learnt more expressive languages than Pascal and C can write code in them much more quickly, and concisely than a low level implementation. For those who are more experienced that you are, the trade-off between verbosity and performance is very important.
Well.... yes and no. Chess is not that intractable. It is just beyond the current horizon of what we can compute. Before I started writing this comment I had thought the size of the game-tree was conjectured to be about 2^120 which would put it in the sun burning out category. But then I stumbled across this reference on google:
Jurg Nievergelt, of ETH Zurich, quoted the number 2^70 (or about 10^21) in
e-mail, and referred to his paper "Information content of chess positions",
ACM SIGART Newsletter 62, 13-14, April 1977, to be reprinted in "Machine
Intelligence" (ed Michie), to appear 1990.
Now 2^70 is not that intractable at all. Let assume that you're going to build a *large* machine. 2^20 processors is quite doable. Each of these processors can perform 2^32 operations per second without getting to far ahead of what we have now. That's 2^52 operations per second. Or about 2^18 seconds, roughly 2 days +/- some fudge factor. That fudge factor is how many atomic operations it takes to evaluate a position. On a commodity machine it will be a large number, on a custom architecture (like Deep Blue) it will be very low.
So brute-forcing chess is currently on the horizon of what can be done now. Memory for the game tree would be more of an issue than processing speed. Because we can loop through the same position the game tree is really a dag, so it is not feasible to compute local solutions to sub-trees. As a (very) loose estimate of the lower bound on the memory required: assume an average branching factor of 32, each position can be stored in 70 bits, so 70*(32+1) bits per tree node is about 2^11 bits, so the tree would need 2^81 bits. That is a hell of a lot, but memory densities are increasing exponentially as well. I don't know how big the Googleplex is, but researchers are starting to consider exabyte-sized systems. It will probably be done by somebody in 10-20 years.
As far as the living breathing human part goes, I think that was the OP's point. We can solve the game with a "dumb" approach, so it is not amongst the hardest of problems (those that we can't evaluate solutions to effectively). Whether or not that "dumb" approach is effective depends entirely where you are on the current exponential curve of processing power.
PS I mean MaskedSlacker's point, not the OP. I can see that the AC he was answering was actually using trivial in completely the wrong sense.
As a correction, can you really have missed all of media noise around the equation that brought down Wall Street? That's a pretty big example of a broken model that directly led to a lot of the current chaos, rather than people feeding garbage into a good model.
I was actually supposed to meta-mod this comment but decided to reply instead. You don't understand what trivial means. It doesn't mean that it's "easy" as in a low number of computational steps. It means that there are no complicated decisions in the algorithm. For brute-forcing a game this is true, you simply evaluate every possible response to every possible until you reach the leafs of the game tree. So it is a trivial algorithm even though it is computationally intractable.
But how would ethernet between the display and the device accessing the network help in that? If the device is a computer, or an XBox or whatever, then it can view youtube and then display the video over HDMI. Having IP between the display and the device using the display doesn't seem to serve any purpose that I can think of but one*.
* And I don't like that purpose because it is a back-channel for DRM authentication.
Why Ethernet? Can you think of any devices that connect to your home theater (Game console, DVR, etc.), that have video, audio, and ethernet? Here is a hint.. I just named two. Can you look forward and see how more devices in the future will have network connectivity? I sure hope so.
And what is the point of having a network channel between these devices and a *display* ? As the GP asked, why do I need ethernet on my display device?
I didn't claim that it would be ideal. The description of noop suggests that it leaves most of the logic up to the drive controller - which is an improvement for SSD over scheduler logic tailored to a mechanical drive.
So given the massive erase blocks what is the best scheduler for an SSD?
You've quoted the marketing fluff from the article about what Microsoft says TRIM support in Windows 7 will achieve. Do you think that this is a demonstration that you understand TRIM?
I'd refer you to the link higher up the thread. Now it's a hell of a long article, but at least it explains what TRIM is. It allows blocks to be invalidated on the drive directly. Without waiting for them to be overwritten. Note that this explanation is two short sentences and explains *exactly* what TRIM is. Your quote is a marketing attempt to explain what TRIM will achieve.
So the noop scheduler would be the correct choice for a drive that supports TRIM, as the GP claimed. Although the scheduler itself will still need direct support for sending TRIM commands to the storage.
It's strange that you think that - real (actual physical) whiteboards are great for engineers and scientists. There is still a division between people who are happy thrashing ideas out on a board, and those who feel that they need to nurture something in private. I have to ask whether your comments about the environment being poor for research and insight are limited to virtual whiteboards or if they extend to their real-life counterparts?
So where do you think virtual whiteboards break down? Is it the lack of associated physical cues, or just the terrible interfaces that they ship with?
Wouldn't Sweden have to start using the Euro before they could switch away from it?
Sure, a lot of references amongst programmers point to these books - but that doesn't mean that these books provide pointers towards programming. I'm sure that the difference between bijection and implication is something that needs to be learnt during a programming course... :)
A lot of the books that you recommend are good, but have nothing to do with programming. I think you should change the word "essential" to stop people getting the wrong idea about how relevant most of the fiction is.
I don't think everyone can do it, but it probably isn't that uncommon amongst the young. I could hear TVs and other electronics until my early 20s and then it faded away along with the rest of my hearing :) It does sound like your hearing is exceptionally acute though.
Errr.... so close and yet so far? Why not change that so that programs are recipes - closer to the actual truth and then they gain some understanding that the recipe controls what gets done with the oven.
Or, if you want to stick to the programs as tools metaphor then make the chef the CPU. The oven is just another tool, and you've made a weird distinction between one tool, and all the other tools in the kitchen just to shoehorn it in.
You're not comparing like with like. When you have 5-6 tabs open it is as you describe. But these proposals are aimed at users with 50-60 tabs. If you fire open that many tabs then how wide is each one? 20 pixels? So a fairer comparison would be which dimension do you want to lose 20 pixels from.
If you have a 16:9 screen then this is an easy choice. For 4:3 it is somewhat harder, but most websites are mostly text, and text flows better in narrow columns than in wide ones.
This Add-on is brilliant. I found the link yesterday when this story came up on another site - after just 24-hours I don't think I could go back to a browser that doesn't work like this.
Some of proposed ideas are nice, the one that I would really like to see would integrate the tab tree with the browser bookmarks to make something like the Dock on OS-X. Pages could be pinned into the tree, if they are not open then they act like a bookmark and launch when you click them. If they are open then clicking switches to that tab.
Come on Mozilla, please clone the Dock...
PS If you change your Google search prefs to open results in a new window, your browser prefs to open new windows in a tab, and use this extension - it is nirvana.
I've wondered how much the side effects of sleep deprivation change between different people. I would guess that I'm more sensitive to these side-effects than most - I start getting minor auditory hallucinations after being awake for 16+ hours (not much more than a normal day), and visual effects about 20+ hours in. But then I don't think I could survive more than a couple of days, I hit a really hard brick wall at about 40 hours and can't stay awake. By that stage I've already gone through the temperature changes that they describe in the article.
Although it is a hellish way to work there is something to be said for not having to pick up context repeatedly in a problem. A 24-hr stretch in the office seems to produce about as much work as a standard 40-hr week.
Then you don't understand how life as an academic actually works. There are duties that consume a fixed amount amount of time: lecturing, marking, contact time, admin etc. Research then consumes whatever is left. By its nature it is hard to measure the productiveness of research - but the salary being paid to an academic covers this activity.
Now, how many academics do you know that have written books? In every case that I can think of this "in addition to" was carried out during office hours as well as at home, and was a direct replacement for the research activities that would normally have occupied that time.
It is all very well having specific job requirements - but they don't mean much when one of the requirements is for an amorphous ill-defined activity that naturally saturates all time that could be allocated to it.
Actually, as an academic I know exactly what I'm talking about. Nobody writes textbooks for money - it does not pay. And we have enough flexibility in our work to choose to write books in the first place. Try spending a large fraction of a year working on a textbook instead of normal duties in any other career than academia and come back and tell me that it is not something you get paid for.
Your post is officially full of awesome.
Or possibly just a small program that decompresses the deltas and applies them. Kolmogorov was a fan of updating his work this way...
You've missed a third case - the author was paid a salary while he wrote the book. Does he then deserve any further payment? Peter Wayner (the author and submitter) is (I'm guessing) an academic. So he would have been paid a salary during the years that he wrote the book. He didn't need a fee on completion because it had already been paid.
Perhaps this will be a lesson in irony?
Like many other readers I plugged "data compression wayner" into Google to see which book it is. In my case I was wondering if it is the same text sitting on my boss' desk that I've been planning to buy.
I can't actually find your book anymore. There are hundreds of results with your question, and thousands more spam farm sites that have sprung up because this is suddenly a popular search term.
As a result, all information on getting your book (either legitimately or otherwise) has been drowned under a sea of crap. I would have liked to know if you have an electronic copy available for a reasonable price. The MacKay text linked to above is now sitting on my harddrive because it is easy to find and a legitimate copy is freely available.
But they will read secondary sources about twitter, facebook, myspace .... and feel happy that it is gone.