I once shared a house with a guy whose amiga (500?) had been run over by a passenger jet (a 737 or something like that). It had fallen off the baggage trolley and then the plane had reversed over it.
By the time I got to see it, the thing had a new case (made of perspex) and he'd soldered links over the cracked tracks in the circuit board. Ran just like a bought one!
ENIAC... was designed to be Turing Complete! (I don't think so)
Have you *seen* ENIAC (or what remains of it - go to UPenn and ask). ENIAC was not only *not* Turing Complete - it wasn't even designed to support branches. They were only discovered by accident by an operator who discovered that by kludging a cable together, the numerical output of a computation could be routed into the control input for the next computation (thus making it conditional - if the output was 0, the next computation was not initiated).
ENIAC did not even have a reusable ALU - the program had to be laid out around the room with the output of each stage of the computation wired into the input to the next stage (each ALU stage was about 3' wide and about 6' high) and they needed a large number of them to do anything useful.
or indeed Konrad Zuse who finished building his third computer in 1941 complete with reusable floating point arithmetic unit (way *way* ahead of ENIAC)
um... any chance I can get a copy of this from you (assuming that the terrasoft license permits it). I have been waiting a while to run linux on my twin g5 and would really like to try it out...
Even if you can't do this I would appreciate an email just to hear your comments on this version of ydl - email to my id at eng.cam.ac.uk
well, my hand is on the mouse already (I was going for the menu right..?) and yes - it's only 46" if the window is tiny and absolutely in the bottom right hand corner and I was going for the first (file?) menu - but you get the idea... - I never found this a problem before I got the huge screen (which is a *wonderful* piece of hardware btw).
To make things worse, some idiot at apple decided that the active point of the mouse pointer isn't the tip of the arrow - it's somewhere randomly under the arrowhead - so you can think you're clicking on the (bottom of the) menu and still hit the background...
I bought a twin processor G5 with the aim of running linux for several reasons:
1) The hardware is comparatively cheap for the performance - it's about the same price as a twin processor 64 bit athlon machine and for my applications the altivec processor appears to have superior performance.
2) This is particularly true when the price of the 23" flat panel display is taken into account.
3) I do real time video processing and unbelievably it is easier to get video out of firewire cameras under linux than under OS X (I do this on my ibook). It took days of my time to figure out how to use quicktime to do this (you have to say you want to record a movie to disk and then make a call saying "don't actually record this to disk" and then another call to say "here is a callback to use for each video frame" - and even then you only get the data after it's been reformatted by quicktime at a cost of 30% of a cpu!)
4) Under linux you get a decent virtual window manager and everthing runs much more smoothly (currently I run OS X almost entirely in X windows mode and I use fink (which makes OS X half usable)).
5) On a large screen Aqua (or whatever apple call their window manager) sucks:
i) Sooner or later you end up using a window in the bottom right hand corner of the screen. When you do this the menus are 23" away. If (in your haste) you miss the menu and click on the background the finder gets focus and you have another 46" round trip to get your menu back.
ii) Unfortunately focus implies raise and you have to click to focus - personally I don't like either of these policies (although I accept that others may).
I would be really interested to hear anyone's experience with YDL on the g5 - unfortunately I'm not prepared to pay terrasoft just to become a beta tester for them (although in this instance I would be *very* happy to pay them if it works well).
Evangelism is WAR
And war it is â" but a war of words, not bullets
Which is odd because the first chapter is basically a set of bullet points; The guy can't string more than two sentences together so I don't think he's going to win the "war of words".
%! why do I get a "you can't post to this page" when I post as me?
The best solution so far is the one where only one prisoner can turn the light on and everybody else turns it off just once. That should take a few over 100 visits from the special prisoner = ~10000 days.
The solutions that require a prisoner to arrive on a day that equals their number (mod 100) are going to take a lot longer (since each prisoner will have to wait ~10000 days for this event to occur).
The best solution I can think of goes as follows:
Each prisoner has a value which starts at 1 with the exception that one prisoner is given value 29 to start with.
Each day has a value which is always 2^n for some n>=0 (some suggestions for the rule giving a day's value are at the end).
The algorithm has three steps when a prisoner arrives in the room:
1) If the light bulb is on when a prisoner arrives, they add the value for the previous day to their own, otherwise their value is unchanged.
2) If they now have value 128, they announce.
3) If their new worth expressed in binary anded with that day's worth != 0 they make sure the light is on when they leave and they subtract that day's worth from their own.
(If you think of this as the lightbulb having a value when it's on, there's a conservation of total value).
Why the algorithm works:
Units of 2^n in value are never broken up, but pairs of 2^n values can be combined to give 2^(n+1). Given a good choice for determining the day's value from it's number, this combining can proceed quite rapidly, in particular, if the first hundred days have day value = 1, then all prisoners who make at least one visit to the room during those days end up with either value = 0 or value = 2.
Suggestions on the day value formula:
suggestion 1:
days 1-100 have value 1
days 101 - 200 have value 2
days 201 - 300 have value 4
...
days 601 - 700 have value 64
repeat.
suggestion 2:
Compute the the evolution of the expected density of powers of 2 present in the population of prisoners. Draw from this expected distribution
using a pseudo-random number generator (although mabye this might require a bit too much brain power on the part of the convicted felons...).
Slashdot Mirror
I once shared a house with a guy whose amiga (500?) had been run over by a passenger jet (a 737 or something like that). It had fallen off the baggage trolley and then the plane had reversed over it.
By the time I got to see it, the thing had a new case (made of perspex) and he'd soldered links over the cracked tracks in the circuit board. Ran just like a bought one!
The airline paid for a new one too IIRC.
ENIAC ... was designed to be Turing Complete! (I don't think so)
Have you *seen* ENIAC (or what remains of it - go to UPenn and ask). ENIAC was not only *not* Turing Complete - it wasn't even designed to support branches. They were only discovered by accident by an operator who discovered that by kludging a cable together, the numerical output of a computation could be routed into the control input for the next computation (thus making it conditional - if the output was 0, the next computation was not initiated).
ENIAC did not even have a reusable ALU - the program had to be laid out around the room with the output of each stage of the computation wired into the input to the next stage (each ALU stage was about 3' wide and about 6' high) and they needed a large number of them to do anything useful.
or indeed Konrad Zuse who finished building his third computer in 1941 complete with reusable floating point arithmetic unit (way *way* ahead of ENIAC)
... It was the breakthrough machine and had a Megabyte, operated at one Megaherz and had a Megapixel display.
Well things have moved on and now I'm waiting for my Gigapixel display. It sounds like I'm going to need one anytime now...
(or not)
soon the linux kernel will be so big, it will become possible to have a picokernel...
Surely these guys have seen the film and know how declations of principle usually turn out...
Beetjebrak,
um... any chance I can get a copy of this from you (assuming that the terrasoft license permits it). I have been waiting a while to run linux on my twin g5 and would really like to try it out...
Even if you can't do this I would appreciate an email just to hear your comments on this version of ydl - email to my id at eng.cam.ac.uk
twd20
well, my hand is on the mouse already (I was going for the menu right..?) and yes - it's only 46" if the window is tiny and absolutely in the bottom right hand corner and I was going for the first (file?) menu - but you get the idea... - I never found this a problem before I got the huge screen (which is a *wonderful* piece of hardware btw).
To make things worse, some idiot at apple decided that the active point of the mouse pointer isn't the tip of the arrow - it's somewhere randomly under the arrowhead - so you can think you're clicking on the (bottom of the) menu and still hit the background...
twd20
I bought a twin processor G5 with the aim of running linux for several reasons:
1) The hardware is comparatively cheap for the performance - it's about the same price as a twin processor 64 bit athlon machine and for my applications the altivec processor appears to have superior performance.
2) This is particularly true when the price of the 23" flat panel display is taken into account.
3) I do real time video processing and unbelievably it is easier to get video out of firewire cameras under linux than under OS X (I do this on my ibook). It took days of my time to figure out how to use quicktime to do this (you have to say you want to record a movie to disk and then make a call saying "don't actually record this to disk" and then another call to say "here is a callback to use for each video frame" - and even then you only get the data after it's been reformatted by quicktime at a cost of 30% of a cpu!)
4) Under linux you get a decent virtual window manager and everthing runs much more smoothly (currently I run OS X almost entirely in X windows mode and I use fink (which makes OS X half usable)).
5) On a large screen Aqua (or whatever apple call their window manager) sucks:
i) Sooner or later you end up using a window in the bottom right hand corner of the screen. When you do this the menus are 23" away. If (in your haste) you miss the menu and click on the background the finder gets focus and you have another 46" round trip to get your menu back.
ii) Unfortunately focus implies raise and you have to click to focus - personally I don't like either of these policies (although I accept that others may).
I would be really interested to hear anyone's experience with YDL on the g5 - unfortunately I'm not prepared to pay terrasoft just to become a beta tester for them (although in this instance I would be *very* happy to pay them if it works well).
twd20
Evangelism is WAR
And war it is â" but a war of words, not bullets
Which is odd because the first chapter is basically a set of bullet points; The guy can't string more than two sentences together so I don't think he's going to win the "war of words".
%! why do I get a "you can't post to this page" when I post as me?
The fastest algorithm to factor composites is the number field sieve for which the complexity is:
$e^{c(\log n)^{1/3} (\log \log n)^{2/3}}$
IIRC the smallest value of c is around 1.9
I coded up my method and with my original suggestion for assigning the day value, I got 10752 days on average (no improvement :-( ).
When I extended the run length to 500 days(*), I brought the average time down to 5886 days (only a factor of 2 improvement).
I wonder if there's a better formula for assigning the day value...?
(*) ie days 1-500 had value 1, days 501-1000 had value 2 etc.
The solutions that require a prisoner to arrive on a day that equals their number (mod 100) are going to take a lot longer (since each prisoner will have to wait ~10000 days for this event to occur).
The best solution I can think of goes as follows:
Each prisoner has a value which starts at 1 with the exception that one prisoner is given value 29 to start with.
Each day has a value which is always 2^n for some n>=0 (some suggestions for the rule giving a day's value are at the end).
The algorithm has three steps when a prisoner arrives in the room:
1) If the light bulb is on when a prisoner arrives, they add the value for the previous day to their own, otherwise their value is unchanged.
2) If they now have value 128, they announce.
3) If their new worth expressed in binary anded with that day's worth != 0 they make sure the light is on when they leave and they subtract that day's worth from their own.
(If you think of this as the lightbulb having a value when it's on, there's a conservation of total value).
Why the algorithm works:
Units of 2^n in value are never broken up, but pairs of 2^n values can be combined to give 2^(n+1). Given a good choice for determining the day's value from it's number, this combining can proceed quite rapidly, in particular, if the first hundred days have day value = 1, then all prisoners who make at least one visit to the room during those days end up with either value = 0 or value = 2.
Suggestions on the day value formula:
suggestion 1:
days 1-100 have value 1
days 101 - 200 have value 2
days 201 - 300 have value 4
days 601 - 700 have value 64
repeat.
suggestion 2:
Compute the the evolution of the expected density of powers of 2 present in the population of prisoners. Draw from this expected distribution using a pseudo-random number generator (although mabye this might require a bit too much brain power on the part of the convicted felons...).