Your points are all valid, and also indicate that the algorithm must be parallel capable. Traversing a link list obviously is not. Linear algebraic equations represent by multi-dimensional arrays can be, depending on the algorithms used to solve them.
Some smart people spent a lot of time on coming up with interesting algorithms to solve such sets of equations, and how to compartmentalize the logic so that there are discrete subsets of work that can be done asynchronously, the key to being able to parallelize a task.
I have no idea what you've been exposed to, but I can assuredly say that you're not looking at the right algorithms if you believe that you won't see a performance increase by dividing a parallizable process across multiple CPUs. I used to run large models on 16 CPU crays. The run time was about 1/14th to 1/15th of what it would have been running on a single CPU. (Real time, actual CPU time increased by about 10%)
nice segue into something that is happening vs what exists. I'm not sure I believe in "moral rights", motals, yes, but rights? Next we'll be having ethical rights too.
I'm not sure moral rights came out of IP rights either, it sounds like something that would more likely have come out of the Church more than anything else.
It will probably remain away from the common desktop for a long time to come. But it does have its uses.
Graphics processing (large amounts of data in multi-dimensional arrays) are a perfect scenario for multi-threading. This is essentially lots of independent little problems.
Another one would be writing real AI. AI running on its own threads could act much more like a real player in a game.
So, there's 2 processes that could benefit from multi-processing. I bet you're saying but that's specialty graphics/games only. Well, what about an intelligent agent handling processing of your email/IM/IRC/web browing? (Not here yet, but everyone thinks eventually there will be some sort of agent technology to help us filter through the chaff)
Copyright holders only have the exclusive right to produce and distribute their works (US only - don't know about other countries)
Once you buy it, you can do anything to it you want privately, and in some cases, publicly. For instance:
Buy a painting, spray paint over it, display it anywhere
Buy several paintings, do a cut and paste montage, display anywhere
Buy a book, selectively cut out words and/or paste in pages from another book, sell book
Buy a film on film (you recall super 8 and the like?), cut out what you want, you can resell it if anyone wants to buy it
Buy a CD or DVD, use a Sharpie to black out certain portions of the disk (let's pretend you know where the physical locations are). It effectively edits it, although your CD/DVD players probably won't play past them, at least not automatically. Then sell that disk, if anyone will buy it.
Note that I can legally do all the above (at least in the US) including reselling what I've done to the "copyrighted" work. What I cannot do is buy the work, duplicate it, and sell the duplicates (that's publication and distribution).
Things get trickier with audio and movies, public/commercial performance of those are restricted. But, you still have many rights as the owner of a copyrighted work. (At least, until the DMCA became law in the US)
so you choose different specs than I pointed out. And who the heck compres 7 year old tech with brand new tech anyways like it is a valid comparison. Granted, it was submitted in 2005, but who cares? What if I were to submit a PII's performance in 2005?
Ok, you have 0 idea what you're talking about. Check out the numbers just in these two specs:
MailSpec
and CFP2000 Rates both indicate that Sun boxes perform at the top or in the top group CPU for CPU. Now, these don't compare prices, I'll agree, but Sun hardware costs have been coming down. Also, note that the Sun hardware ranks at least on par with the truly big players, realms that cheap x86 technology just doesn't play in.
Basically, it all comes down to what you need. Personally, for anything and everything I do at this time, Opteron based systems are the best solution for me. In the past, Sun systems have figured prominently in solutions I worked on, because they were the best fit dollarwise for the requirements.
Because Windows has this interesting habit of loading crap you don't need into RAM, and swapping stuff you do need out into the pagefile. If the DLL that you don't need is loaded into the memory area that's bad, then nothing bad will ever happen, as that particular piece of code will never be attempted to be read.
Because when we shop on the net we're trained to seek out the lowest price, at the expense of quality. And it's not exactly like the places selling dodgy RAM will label it as "Dodgy!".
I thought that we shop on the net to find the quality item we want at the lowest price. I only buy a select few name brand RAM, as I prefer a system that doesn't BSOD or cause kernel panics. I also no longer see the need to overclock really, considering the price you pay for a certain level of performance vs the maximum performance pieces. Stability is much nicer than that extra couple of frames to take you from 100 to 105 or something. Your screen only refreshes a max of 85 times a second (for most), so the most you'll see is 85 fps anyways.
Personally, I find anyone that accepts low-bid purely on the basis of money to be an idiot. You have to look at what you're getting/$.
Sure, the AI doesn't actually have to build anything, generation of items and wonders are done based on a random die percentage.
In Civ 1, if you save the game at every turn (for the truly paranoid) you can effectively stop all your computer opponents from generating any wonders ever by just going back to the last save prior to the computer opponent "building" a wonder.
What I was never able to prove is that on the most difficult setting, the computer's probability of building a wonder within 1 or 2 turns of you building a wonder greatly increased, and if the computer did build a wonder, the odds that it would build the wonder you were building would also increase.
Lastly, it appeared that the computer did not necessarily need the "tech" to achieve a wonder. This was ascertained by checking the save file and confirming that a computer AI did not "build" any of the better units indicative of the tech required to build a wonder.
Too bad I don't have mod points today. Definitely a "nail on the head" post.
I'd actually not thought of design documentation as a series of layers, going from design docs to function calls/interfaces/etc to the actual code implementation. Interesting viewpoint that explicitly states what many of state about the "design" of classes, interfaces, etc.
In simplest terms, here's the flaw with a "not" requirement.
Let's say I want to divide a bank acount into a an integer number of accounts. An alpha would cause a NaN error. Now, since we don't want the NaN error, we make it 0, oops, it would be a divide by 0 error, except somewhere there was another requirement (for whatever reason) that divide by zero's would equate to INT_MAX. Now I have a really big bank account.:)
That's why we do specific requirements instead of "not" requirements (e.g., in case of alpha entry, display error message indicating "only numeric input accepted".)
And, there's the russian wave hugging cargo plane design that's orders of magnitude more fuel efficient than a ship by using ground effects, not to mention a couple of orders of magnitude faster to boot.
It's just the rogue wave or storm that would send it plummeting to the bottom of the ocean that keeps it from being a viable transport.
My problem with this one is that it was listed as a prototype, and the only one they had.
I recall reading about the Moller sky car in Popular Science years ago (5? 10? 15? it was a long time ago;) except then it was a 7 engine beast able to fly 400mph, get 20 mpg with 4 passenagers, along with VTOL. I guess that was merely a paper proposal, although it wasn't presented as such.
Egads! I hope not!!! I have Civ3 (and Civ 1 and 2 for that matter) and played Civ 1 until I could beat it passed out. (The "AI" and movement controls were about as feeble as they come, not to mention that the "AI" cheats.)
I was just about to load Civ 3 and play it, and if it is merely eye candy on top of the Civ 1 engine, I predict I'll win and be done with it in less than 8 hours.
You've got it backwards. NT is a monolithic single proc disaster. On an MP system, only 1 proc runs the kernel (CPU 0).
OS/2 is truly SMP, provided you bought the SMP version. (ie, not the common workstation version)
BTW, what makes you think OS/2 wasn't running multiple threads? It's threading model blows away the NT core's model, and quite possibly BSD/Linux's as well from what I've read (but not enough to form a definite opinion yet)
BTW, the restriction on CPU 0 interrupts is due to the Intel SMP specification for x86, which is why all of these OSes originally, on x86, had CPU 0 restrictions.
And finally, my statement about modern OSes was both sarcasm and mocking - UNIX, for all intents and purposes in all it's flavors, is the current modern OS other than Windows NT/2K/XP, as it's the only real other game in town. (And I'd state that NT/2K/XP is not a modern OS, but a rather hacked together POS, but that's another posting at least.)
Time slicing under the NT core basically gives threads a 32ms slice of time before preempting them (last time I looked, although I believe is is 16ms now based on some multithreaded tests I was running on XP 6 months ago). Under OS/2 and UNIX, a thread could be preempted at any time, not just on 32ms windows.
That is a fine difference, but one that can matter depending on your application.
WinXP isn't designed for MP? At least MS has had some practice with MP, preemptive multitasking and virtual memory.
No. It's not. It's designed with time slicing. time slicing is a quite different beast than preemptive multi-tasking or running on multiple processors (SMP for most of us)
For preemptive multi-tasking, take a look at OS/2's processing. It was preemptive on the same hardware that MS is time-sliced.
For SMP, look at any modern OS (like, for instance, BSD, Linux, Solaris, OS400 - gee, most are older than NT!) and compare against how the NT/2K/XP core really works. The core is a monolithic disaster that pretty much runs on a single processor in an SMP system (which, btw, is what dual core results in). It may have multiple threads, but those threads run on a single processor.
Depends on which benchmark you're going for. Running things designed to run on the x86 single threaded model desktop systems like office apps or games, yes. Try running a massively network app with hundreds of threads though, and a different picture emerges. Don't buy a Sun to play Doom. Do buy a Sun for a Web/Application Server.
actually, in most settings we care about the amount of heat thrown off by Intel's procs. When a single CPU system throws off 100W+, it starts becoming an issue, as the required active cooling starts making the computer loud enough to go beyond merely distracting to annoyance. (And we're not talking about your tame 3.2 GHz CPUs either, we're talking about top-end Intel CPUs near 3.8GHz, or the dual core 3.2 GHz, neither of which can best AMD's offerings which produce significantly less heat)
If you're talking about server rooms, I can tell you that heat is the #1 problem with Intel procs. The limit the CPU density you can pack into a room, as well as the total number (separate issues, density results in concentrated heat sources, total number results in amount of heat to be removed).
Slashdot Mirror
Your points are all valid, and also indicate that the algorithm must be parallel capable. Traversing a link list obviously is not. Linear algebraic equations represent by multi-dimensional arrays can be, depending on the algorithms used to solve them.
Some smart people spent a lot of time on coming up with interesting algorithms to solve such sets of equations, and how to compartmentalize the logic so that there are discrete subsets of work that can be done asynchronously, the key to being able to parallelize a task.
I have no idea what you've been exposed to, but I can assuredly say that you're not looking at the right algorithms if you believe that you won't see a performance increase by dividing a parallizable process across multiple CPUs. I used to run large models on 16 CPU crays. The run time was about 1/14th to 1/15th of what it would have been running on a single CPU. (Real time, actual CPU time increased by about 10%)
nice segue into something that is happening vs what exists. I'm not sure I believe in "moral rights", motals, yes, but rights? Next we'll be having ethical rights too.
I'm not sure moral rights came out of IP rights either, it sounds like something that would more likely have come out of the Church more than anything else.
Yeah, there's more, but they all revolve around this key concept.
Graphics processing (large amounts of data in multi-dimensional arrays) are a perfect scenario for multi-threading. This is essentially lots of independent little problems.
Another one would be writing real AI. AI running on its own threads could act much more like a real player in a game.
So, there's 2 processes that could benefit from multi-processing. I bet you're saying but that's specialty graphics/games only. Well, what about an intelligent agent handling processing of your email/IM/IRC/web browing? (Not here yet, but everyone thinks eventually there will be some sort of agent technology to help us filter through the chaff)
Copyright holders only have the exclusive right to produce and distribute their works (US only - don't know about other countries)
Once you buy it, you can do anything to it you want privately, and in some cases, publicly. For instance:
- Buy a painting, spray paint over it, display it anywhere
- Buy several paintings, do a cut and paste montage, display anywhere
- Buy a book, selectively cut out words and/or paste in pages from another book, sell book
- Buy a film on film (you recall super 8 and the like?), cut out what you want, you can resell it if anyone wants to buy it
- Buy a CD or DVD, use a Sharpie to black out certain portions of the disk (let's pretend you know where the physical locations are). It effectively edits it, although your CD/DVD players probably won't play past them, at least not automatically. Then sell that disk, if anyone will buy it.
Note that I can legally do all the above (at least in the US) including reselling what I've done to the "copyrighted" work. What I cannot do is buy the work, duplicate it, and sell the duplicates (that's publication and distribution).Things get trickier with audio and movies, public/commercial performance of those are restricted. But, you still have many rights as the owner of a copyrighted work. (At least, until the DMCA became law in the US)
1) fixed time lengths could require the 14/28 with a refiling by the *creator* required at the 14 year mark.
2) fixed time lengths protect the hiers
5 years after the death of the artist? Why so long? Why not back down to the more reasonable 14 or 28 year terms?
so you choose different specs than I pointed out. And who the heck compres 7 year old tech with brand new tech anyways like it is a valid comparison. Granted, it was submitted in 2005, but who cares? What if I were to submit a PII's performance in 2005?
UltraSPARCIV's have been out a while now.
But I bought a Mac.
Basically, it all comes down to what you need. Personally, for anything and everything I do at this time, Opteron based systems are the best solution for me. In the past, Sun systems have figured prominently in solutions I worked on, because they were the best fit dollarwise for the requirements.
Because Windows has this interesting habit of loading crap you don't need into RAM, and swapping stuff you do need out into the pagefile. If the DLL that you don't need is loaded into the memory area that's bad, then nothing bad will ever happen, as that particular piece of code will never be attempted to be read.
I thought that we shop on the net to find the quality item we want at the lowest price. I only buy a select few name brand RAM, as I prefer a system that doesn't BSOD or cause kernel panics. I also no longer see the need to overclock really, considering the price you pay for a certain level of performance vs the maximum performance pieces. Stability is much nicer than that extra couple of frames to take you from 100 to 105 or something. Your screen only refreshes a max of 85 times a second (for most), so the most you'll see is 85 fps anyways.
Personally, I find anyone that accepts low-bid purely on the basis of money to be an idiot. You have to look at what you're getting /$.
Sure, the AI doesn't actually have to build anything, generation of items and wonders are done based on a random die percentage.
In Civ 1, if you save the game at every turn (for the truly paranoid) you can effectively stop all your computer opponents from generating any wonders ever by just going back to the last save prior to the computer opponent "building" a wonder.
What I was never able to prove is that on the most difficult setting, the computer's probability of building a wonder within 1 or 2 turns of you building a wonder greatly increased, and if the computer did build a wonder, the odds that it would build the wonder you were building would also increase.
Lastly, it appeared that the computer did not necessarily need the "tech" to achieve a wonder. This was ascertained by checking the save file and confirming that a computer AI did not "build" any of the better units indicative of the tech required to build a wonder.
Too bad I don't have mod points today. Definitely a "nail on the head" post.
I'd actually not thought of design documentation as a series of layers, going from design docs to function calls/interfaces/etc to the actual code implementation. Interesting viewpoint that explicitly states what many of state about the "design" of classes, interfaces, etc.
In simplest terms, here's the flaw with a "not" requirement.
:)
Let's say I want to divide a bank acount into a an integer number of accounts. An alpha would cause a NaN error. Now, since we don't want the NaN error, we make it 0, oops, it would be a divide by 0 error, except somewhere there was another requirement (for whatever reason) that divide by zero's would equate to INT_MAX. Now I have a really big bank account.
That's why we do specific requirements instead of "not" requirements (e.g., in case of alpha entry, display error message indicating "only numeric input accepted".)
And, there's the russian wave hugging cargo plane design that's orders of magnitude more fuel efficient than a ship by using ground effects, not to mention a couple of orders of magnitude faster to boot.
It's just the rogue wave or storm that would send it plummeting to the bottom of the ocean that keeps it from being a viable transport.
My problem with this one is that it was listed as a prototype, and the only one they had.
I recall reading about the Moller sky car in Popular Science years ago (5? 10? 15? it was a long time ago;) except then it was a 7 engine beast able to fly 400mph, get 20 mpg with 4 passenagers, along with VTOL. I guess that was merely a paper proposal, although it wasn't presented as such.
Egads! I hope not!!! I have Civ3 (and Civ 1 and 2 for that matter) and played Civ 1 until I could beat it passed out. (The "AI" and movement controls were about as feeble as they come, not to mention that the "AI" cheats.)
I was just about to load Civ 3 and play it, and if it is merely eye candy on top of the Civ 1 engine, I predict I'll win and be done with it in less than 8 hours.
Windows isn't 7 years behind, it took a wrong fork 15 years ago.
You've got it backwards. NT is a monolithic single proc disaster. On an MP system, only 1 proc runs the kernel (CPU 0).
OS/2 is truly SMP, provided you bought the SMP version. (ie, not the common workstation version)
BTW, what makes you think OS/2 wasn't running multiple threads? It's threading model blows away the NT core's model, and quite possibly BSD/Linux's as well from what I've read (but not enough to form a definite opinion yet)
BTW, the restriction on CPU 0 interrupts is due to the Intel SMP specification for x86, which is why all of these OSes originally, on x86, had CPU 0 restrictions.
And finally, my statement about modern OSes was both sarcasm and mocking - UNIX, for all intents and purposes in all it's flavors, is the current modern OS other than Windows NT/2K/XP, as it's the only real other game in town. (And I'd state that NT/2K/XP is not a modern OS, but a rather hacked together POS, but that's another posting at least.)
Time slicing under the NT core basically gives threads a 32ms slice of time before preempting them (last time I looked, although I believe is is 16ms now based on some multithreaded tests I was running on XP 6 months ago). Under OS/2 and UNIX, a thread could be preempted at any time, not just on 32ms windows.
That is a fine difference, but one that can matter depending on your application.
the 2xx series is only good up to 2 procs. For more than 2, you have to run the 8xx series.
:)
But, with the dual core 2xx series coming out, you can effectively run a 4 proc system. It'd be smoking.
No. It's not. It's designed with time slicing. time slicing is a quite different beast than preemptive multi-tasking or running on multiple processors (SMP for most of us)
For preemptive multi-tasking, take a look at OS/2's processing. It was preemptive on the same hardware that MS is time-sliced.
For SMP, look at any modern OS (like, for instance, BSD, Linux, Solaris, OS400 - gee, most are older than NT!) and compare against how the NT/2K/XP core really works. The core is a monolithic disaster that pretty much runs on a single processor in an SMP system (which, btw, is what dual core results in). It may have multiple threads, but those threads run on a single processor.
Depends on which benchmark you're going for. Running things designed to run on the x86 single threaded model desktop systems like office apps or games, yes. Try running a massively network app with hundreds of threads though, and a different picture emerges. Don't buy a Sun to play Doom. Do buy a Sun for a Web/Application Server.
actually, in most settings we care about the amount of heat thrown off by Intel's procs. When a single CPU system throws off 100W+, it starts becoming an issue, as the required active cooling starts making the computer loud enough to go beyond merely distracting to annoyance. (And we're not talking about your tame 3.2 GHz CPUs either, we're talking about top-end Intel CPUs near 3.8GHz, or the dual core 3.2 GHz, neither of which can best AMD's offerings which produce significantly less heat)
If you're talking about server rooms, I can tell you that heat is the #1 problem with Intel procs. The limit the CPU density you can pack into a room, as well as the total number (separate issues, density results in concentrated heat sources, total number results in amount of heat to be removed).