Greg Egan has a great story (I believe it is called "learning to be me") about this small computer (jewel) that you get implanted in your brain as a small child. The premise is that all other parts of the body can be readily replaced, appart from the brain. Thus, the only obstacle to eternal life is copying the brain.
The jewel sits in your head, monitoring your inputs (sight, sound, tactile...) and your outputs. Eventually, it is consistently able to predict your actions. It has learned how to be you.
Later in life, it is time for your transference, where the jewel is given control over the outputs, and your brain takes the back seat. Of course, being a good fiction short, the jewel soon diverges from what you want to do, but the real you has no outputs... and is eventually scooped out to be replaced by some biologically inert material, while the jewel lives to be 1000s of years old.
It was several years since I read it, but good stuff all the same.
It wasn'y clear from any of those links on what gounds microsoft objected. It wasn't a port of one of their games, was it, but rather to their platform?
How is this different from apple throwing a hissy fit because I've ported galeon to run native on carbon (which I haven't, but for sake of argument)?
I truly am confused, not just shocked, shocked. Not askign you to justify M$ reasoning, just explain it.
Re:6502 microcode bugs...
on
Pet Bugs?
·
· Score: 1
well you can argue definitions 'till the cows come home, but I always thought that the R in RISC referred to reduced _complexity_ of instructions rather than reduced _number_ of instructions. So risc cpus don't have any get-from-memory-and-jump- indirect style instructions. So this page load bug would be a bug on a load instruction, rather than a jump.
To muddy the waters, the PowerPC has a RLWINM instruction, which does a rotate left for x bits and ands with a mask. Which is kinda ciscy. But it doesn't do any memory indirect addressing, so it passes that test, and gets the job done in one cycle, so passes that as well.
This is basically just interactive arithmetic coding of arbitrary strings... but with the twist that each letter modifies the artithmetic probablities. I've only seen adaptive probabilities done on huffman trees.
BTW. there is a reason why IBM is involved: they own the patent on arithmetic coding.
Re:6502 microcode bugs...
on
Pet Bugs?
·
· Score: 1
The Commodore PET apparently had a design flaw where if you accessed this one memory mapped I/O location in a tight loop, you could make some component on the mainboard overheat and fry.
ooops!
Re:6502 microcode bugs...
on
Pet Bugs?
·
· Score: 2
I seem to remember that one being documented, too. I don't know if it was the official intel dox, or some thid party ones I had lying around, but there was about a page per instruction, and that bug was mentioned.
As for 6502 quirks, I think the hard coded stack page (02) was pretty funky. Ah. good old cisc. IIRC, the 6502 was still pretty snappy, getting most ops done in one or two cycles.
so only the link is banned? Just make it a plaintext url. Or provide handy get-lucky search terms for google. The point is that this is a meaningless victory as it is very easy to comply with the removal of a link while still providing means for the interested reader to get to the web page.
heh. I just "invented" a stabilised pair of binoculars the other day. The idea is that you mount cheap-ass (the kind you use in ABS systems, natch ~ $10 a pop) micro-gyros to sense pitch and yaw of the binoculars (roll is benign, and REALLY hard to compensate for) and correct by micro-actuators attached to the optics.
Maybe I'm really shaky, but I have a really hard time using binoculars, as I just can't keep them very steady. (yes, I am a crap shot with a rifle, too)
1280x720, 24bit, at 25 fps is more like 553 Mbs. Unless they use a JPG-like YCrCB-and-downsample transform. That would get the numbers down to about the right range. I imagine that neither color not spacial fidelity makes much difference for motion video.
If you have a *standard* work-for-hire contract (which is what 99% of programmers do), then the company owns of ALL code you write, be it "on the clock" or at home. Hence, John would be unable to work on sawfish without giving it to apple, which he may not have wanted to do.
Many people don't realize (the 1%) that this is merely just in the standard contract, rather than law, and can be dealt with during employment negotiation. Obviously countries other than the USA will be different.
One of the Perl gurus had this happen to him recently, and wrote a longish writeup about it on the perlmonks (?) website. A brief search should provide more info. The upshot is that the GPL doesn't help you here, because you likely don't have the right to assign copyright terms to code you write in your free time.
Assuming they have Samba, NFS, or AppleShare installed (covers about 95% of OS out there, I think), the OS needs to go too, as these can be used to share music and files. We all recall the hubbub that cablemodems and inadvertently open shares caused a few years ago.
The real idiot here is the Sysadmin who wasn't filtering these packets. Internal p2p programs are just a convenient way to share music.
Perhaps you can do something inbetween: start downgrading the performance of said ports, depending on length of connection. Short connections on a p2p port go through, while longer transfers start getting slower and slower because you're dropping every nth packet.
So instead of making it impossible, illegal, or whatever, just make p2p really inconvenient. If everything else works fine, the culprits can't really complain -- in fact, this will likely make everything else faster.
A net search indicates that I may be the only person to get crappy ext3 / i/o performace under 2.4.18, which suggests I'm incompetent to compile my own kernel, and should just use someone-else's.
So I guess I'll be experiencing the excitement of a semi-backed-up (my mp3s are living on borrowed time) upgrade over the weekend. woo hoo!
It appears to be orthogonal to memory pressure, but related to i/o. Running top suggests that kjournald is the culprit, as it both sucks up a (comparatively) large fraction of cpu time and % (low single digits under i/o load), in addition to bing near the top of the runq when UI lag sucks most.
I have had exactly the opposite experience. RH 7.2 (2.5.7-10 IIRC) with ext3 is snappy and responsive, even under the heavy cpu and i/o load of a background kernel compile.
But I wanted ALSA, so I grabbed 2.4.18, and installed that. It is absoultely HORRIBLE. With any sort of i/o in the background, the mouse is laggy, GUI latency can be measured in large fractions (and sometimes numbers of) seconds. top This is for exactly the same setup otherwise. Low latency patches don't help much. maybe a little, but it still is unacceptably laggy. Forget xmms + pan (one program to feed the other:-)) at the same time.
So there are two possibilities: 1) fsked up my 2.4.18 config, and thus ended up compiling a really crappy kernel. But I've been compiling kernels since 1.2.13, and have yet to have one behave anywhere NEAR this badly. 2) RH have significantly hacked 2.4.7 to make it useful. Does anyone know whether the same hacks have happened for the 7.3 kernel?
pdftex generates nice results when it works, but seems to choke on the more advanced packages; typically the psfrag and listings.
Going to pdf via dvi is more robust, but you can get really shitty pdf quality unless you hack the default dvipdf command to use type 3 fonts. If you're unlucky it looks like it was sent via fax (~ 72 dpi). Not pretty.
but the difference is that if you strip a computer of its preinstalled M$OS, and instead install FreeBSD, then microsoft HAVE some ability to restrict you (the giver) from installing this spare OS on another computer, but they have absolutely NO ability to restrict the recipient from accepting the cleansed computer.
The licence doesn't have to stay with the computer, but it can't be used with any other computers. BIG difference.
Basically, microsoft are completely misrepresenting the burden of copyright verification, hoping (with reason) that educators will be too spineless to question the webpage.
I have the original Radeon AIW, and am not very happy with it. It fails on two counts:
1) from an engineering standpoint, it's hacky. Look at those pictures, esp [site is down, so URL is unverified]the wires. The card should have one coax in (and optionally svideo) and one DRI out. Instead it has a bush of wires. This is--for example--because they don't do sound output oer the bus, but instead send it out a wire, which then must be plugged into your soundcard to be sampled. Getting a soundblaster to co-exist with the AIW is a bitch and a half. And double-decoding my sound just feels wrong. 2) Macrovision. I bought one of these to be able to watch VHS movies on my computer, and every so often, one just doesn't work. I figure it's macrovision. The sound continues, but the video freezes after about 3 secs.
3) I agree with everyone else: The bundled software and drivers are so bad, they should pay me to put up with this crap.
So:
anyone have a better solution? I saw one guy suggesting getting the hauppage or bt442 cards.
Programs (and hence PCs and operating systems) MUST accept illformed inputs. When your PC crashes because you tried to read a CD, well, the CD sucks, but so does your computer.
Reminds me of the polish virus "hey! I'm a virus. Send me to everyone you know!". Now, only an idiot would fall for that, and only a peice-of-crap computer will crash on ill-formed input.
Slashdot Mirror
Greg Egan has a great story (I believe it is called "learning to be me") about this small computer (jewel) that you get implanted in your brain as a small child. The premise is that all other parts of the body can be readily replaced, appart from the brain. Thus, the only obstacle to eternal life is copying the brain.
The jewel sits in your head, monitoring your inputs (sight, sound, tactile...) and your outputs. Eventually, it is consistently able to predict your actions. It has learned how to be you.
Later in life, it is time for your transference, where the jewel is given control over the outputs, and your brain takes the back seat. Of course, being a good fiction short, the jewel soon diverges from what you want to do, but the real you has no outputs... and is eventually scooped out to be replaced by some biologically inert material, while the jewel lives to be 1000s of years old.
It was several years since I read it, but good stuff all the same.
It wasn'y clear from any of those links on what gounds microsoft objected. It wasn't a port of one of their games, was it, but rather to their platform?
How is this different from apple throwing a hissy fit because I've ported galeon to run native on carbon (which I haven't, but for sake of argument)?
I truly am confused, not just shocked, shocked. Not askign you to justify M$ reasoning, just explain it.
well you can argue definitions 'till the cows come home, but I always thought that the R in RISC referred to reduced _complexity_ of instructions rather than reduced _number_ of instructions. So risc cpus don't have any get-from-memory-and-jump-
indirect style instructions. So this page load bug would be a bug on a load instruction, rather than a jump.
To muddy the waters, the PowerPC has a RLWINM instruction, which does a rotate left for x bits and ands with a mask. Which is kinda ciscy. But it doesn't do any memory indirect addressing, so it passes that test, and gets the job done in one cycle, so passes that as well.
This is basically just interactive arithmetic coding of arbitrary strings... but with the twist that each letter modifies the artithmetic probablities. I've only seen adaptive probabilities done on huffman trees.
BTW. there is a reason why IBM is involved: they own the patent on arithmetic coding.
erm. Intel? I said that? Brainfart.
The Commodore PET apparently had a design flaw where if you accessed this one memory mapped I/O location in a tight loop, you could make some component on the mainboard overheat and fry.
ooops!
I seem to remember that one being documented, too. I don't know if it was the official intel dox, or some thid party ones I had lying around, but there was about a page per instruction, and that bug was mentioned.
As for 6502 quirks, I think the hard coded stack page (02) was pretty funky. Ah. good old cisc. IIRC, the 6502 was still pretty snappy, getting most ops done in one or two cycles.
so only the link is banned? Just make it a plaintext url. Or provide handy get-lucky search terms for google. The point is that this is a meaningless victory as it is very easy to comply with the removal of a link while still providing means for the interested reader to get to the web page.
and all this time, I was wondering how Canon's IS series of 35 lenses worked... doh!
heh. I just "invented" a stabilised pair of binoculars the other day. The idea is that you mount cheap-ass (the kind you use in ABS systems, natch ~ $10 a pop) micro-gyros to sense pitch and yaw of the binoculars (roll is benign, and REALLY hard to compensate for) and correct by micro-actuators attached to the optics.
Maybe I'm really shaky, but I have a really hard time using binoculars, as I just can't keep them very steady. (yes, I am a crap shot with a rifle, too)
1280x720, 24bit, at 25 fps is more like 553 Mbs. Unless they use a JPG-like YCrCB-and-downsample transform. That would get the numbers down to about the right range. I imagine that neither color not spacial fidelity makes much difference for motion video.
s/radiation/convection/
and i'd agree with you.
probably not the case.
If you have a *standard* work-for-hire contract (which is what 99% of programmers do), then the company owns of ALL code you write, be it "on the clock" or at home. Hence, John would be unable to work on sawfish without giving it to apple, which he may not have wanted to do.
Many people don't realize (the 1%) that this is merely just in the standard contract, rather than law, and can be dealt with during employment negotiation. Obviously countries other than the USA will be different.
One of the Perl gurus had this happen to him recently, and wrote a longish writeup about it on the perlmonks (?) website. A brief search should provide more info. The upshot is that the GPL doesn't help you here, because you likely don't have the right to assign copyright terms to code you write in your free time.
ah.
Assuming they have Samba, NFS, or AppleShare installed (covers about 95% of OS out there, I think), the OS needs to go too, as these can be used to share music and files. We all recall the hubbub that cablemodems and inadvertently open shares caused a few years ago.
The real idiot here is the Sysadmin who wasn't filtering these packets. Internal p2p programs are just a convenient way to share music.
Perhaps you can do something inbetween: start downgrading the performance of said ports, depending on length of connection. Short connections on a p2p port go through, while longer transfers start getting slower and slower because you're dropping every nth packet.
So instead of making it impossible, illegal, or whatever, just make p2p really inconvenient. If everything else works fine, the culprits can't really complain -- in fact, this will likely make everything else faster.
Thanks for the info, but I get much better performance with the RH kernel than my own -- ALL else being equal, including hdparms.
Or am I misunderstanding something here? Are you suggesting that hdparm tweaks are part of the kernel?
Thanks.
A net search indicates that I may be the only person to get crappy ext3 / i/o performace under 2.4.18, which suggests I'm incompetent to compile my own kernel, and should just use someone-else's.
So I guess I'll be experiencing the excitement of a semi-backed-up (my mp3s are living on borrowed time) upgrade over the weekend. woo hoo!
Oh just to add:
It appears to be orthogonal to memory pressure, but related to i/o. Running top suggests that kjournald is the culprit, as it both sucks up a (comparatively) large fraction of cpu time and % (low single digits under i/o load), in addition to bing near the top of the runq when UI lag sucks most.
I have had exactly the opposite experience. RH 7.2 (2.5.7-10 IIRC) with ext3 is snappy and responsive, even under the heavy cpu and i/o load of a background kernel compile.
:-)) at the same time.
But I wanted ALSA, so I grabbed 2.4.18, and installed that. It is absoultely HORRIBLE. With any sort of i/o in the background, the mouse is laggy, GUI latency can be measured in large fractions (and sometimes numbers of) seconds. top This is for exactly the same setup otherwise. Low latency patches don't help much. maybe a little, but it still is unacceptably laggy. Forget xmms + pan (one program to feed the other
So there are two possibilities:
1) fsked up my 2.4.18 config, and thus ended up compiling a really crappy kernel. But I've been compiling kernels since 1.2.13, and have yet to have one behave anywhere NEAR this badly.
2) RH have significantly hacked 2.4.7 to make it useful. Does anyone know whether the same hacks have happened for the 7.3 kernel?
Thoughts?
pdftex generates nice results when it works, but seems to choke on the more advanced packages; typically the psfrag and listings.
Going to pdf via dvi is more robust, but you can get really shitty pdf quality unless you hack the default dvipdf command to use type 3 fonts. If you're unlucky it looks like it was sent via fax (~ 72 dpi). Not pretty.
A google search will set you straight.
but the difference is that if you strip a computer of its preinstalled M$OS, and instead install FreeBSD, then microsoft HAVE some ability to restrict you (the giver) from installing this spare OS on another computer, but they have absolutely NO ability to restrict the recipient from accepting the cleansed computer.
The licence doesn't have to stay with the computer, but it can't be used with any other computers. BIG difference.
Basically, microsoft are completely misrepresenting the burden of copyright verification, hoping (with reason) that educators will be too spineless to question the webpage.
The problem is that headway is something like a knot. So you're only going to use this gliding to move you between currents, rather than against them.
... not to mention a one-to-one relationship with my computer.
Great! Here I was thinking we had an exclusive relationship all along.
I have the original Radeon AIW, and am not very happy with it. It fails on two counts:
1) from an engineering standpoint, it's hacky. Look at those pictures, esp [site is down, so URL is unverified]the wires. The card should have one coax in (and optionally svideo) and one DRI out. Instead it has a bush of wires. This is--for example--because they don't do sound output oer the bus, but instead send it out a wire, which then must be plugged into your soundcard to be sampled. Getting a soundblaster to co-exist with the AIW is a bitch and a half. And double-decoding my sound just feels wrong.
2) Macrovision. I bought one of these to be able to watch VHS movies on my computer, and every so often, one just doesn't work. I figure it's macrovision. The sound continues, but the video freezes after about 3 secs.
3) I agree with everyone else: The bundled software and drivers are so bad, they should pay me to put up with this crap.
So:
anyone have a better solution? I saw one guy suggesting getting the hauppage or bt442 cards.
Programs (and hence PCs and operating systems) MUST accept illformed inputs. When your PC crashes because you tried to read a CD, well, the CD sucks, but so does your computer.
Reminds me of the polish virus "hey! I'm a virus. Send me to everyone you know!". Now, only an idiot would fall for that, and only a peice-of-crap computer will crash on ill-formed input.