Indeed. 1000 simple CPUs will fit in a FPGA, though it might require one near the top of the line. (e.g., picoblaze reportedly needs 96 "slices" and 1.5 "block RAMs"; the biggest Virtex-7 FPGA has more than 1400x as many block RAMs and 3100x as many "slices") There's little doubt that you could program a DCT for a picoblaze, if you wanted to.
It's hard to tell what 5.0GBps refers to -- the bitrate of the incoming, uncompressed, RGB video data? If so, that's maybe about 800FPS of 1080P video. In a circa 2002 paper, FPGAs were doing 100FPS HDTV DCT; an improvement of only a factor of 8 in the intervening 8 years would be frankly disheartening. Especially given that DCT of a frame of video is embarassingly parallel. The FPGA I mentioned earlier could hold 180 copies of the DCT from this paper; right there you have 18kFPS without even raising the clock. But the multipliers can probably also be clocked faster now...
You've just described the FPGA. Large areas of an FPGA are devoted to thousands of almost-identical functional blocks ("slices" in xilinx parlance). For instance, in one Xilinx family, a slice contains a 4-input LUT, a flip-flop (1 bit of memory, called an FF), and other specific gates that help implement things like carry chains, shift registers, and some 5+input functions the chip designers thought were commonly encountered.
Other areas contain "block RAMs" and "DSP cores" (basically, dedicated multipliers).
But now you've got yourself a lot of hard problems to solve: how to dice a program into something that is represented by essentially LUTs and FFs. how to recognize when a special function outside the LUT, like a carry chain, should be used. how to efficiently route the signals from where they're produced to where they're consumed. how to actually implement an efficient LUT where the contents are field-programmed. how to figure figure out what speed to clock the whole thing at so that it operates properly. how to read in the configuration to the chip. This is an enormous investment in research and software, and you still have to target the chip with languages that are totally alien to your typical C/C++ programmer.
As far as I know there is no production FPGA that you can write software for without using proprietary software. (though often the software can be obtained at no cost, at least for the non-flagship FPGA chips) This is partially because the details of bitstream structure are trade secrets of the respective FPGA companies, but also partially due to the inherent difficulty of the task.
Two human users (web+email+etc), three websites (around 115k hits/month total), and cvs server for an open source project (a few dozen "cvs up" a day).
If I want to download porn until it chafes, I can easily get to just under my ISP's monthly cap of 30GB/month, after which it's 0.50USD/GB.
Fedora is an upgrade treadmill. With Fedora, you're stuck upgrading every 12 months or so, or you can't get security updates anymore. With Fedora, install an LTS version and you're covered for 5 years on the server. That's why I switched.
And Microsoft Word does? When you type "he took a bow", does Microsoft silently add a possibly-incorrect annotation that you can't see, or does Clippy pop up and ask you how to pronounce the word you just typed? Also, it looks like you mean "homonym" or "homograph", not "homophonic".
It figures---the post with the useless numbers was modded up to 5, but the important one that gave the "methodology" wasn't moderated up past 2, so it went unseen (by me, at any rate). It doesn't look like Windows CALC.EXE even computes the full value of factorials, judging from the way it displays 100! in scientific notiation. This truly is the contender for most worthless benchmark! Though, to bring this back around to increasing performance, I suggest the NOPsort. It's a version of bogosort optimized for overclocked computers. First, NOP is executed, which may sort the array. Then, the order of the array is checked. If the process never completes, just overclock your computer some more. I'll wait for someone else to publish the first NOPsort benchmark, though.
I don't know what method you're using to compute factorials, or whether the time includes converting to decimal and displaying on a terminal. My programs do not. (actually, the last one does convert to decimal and write to a file, and it still runs quite a bit faster than yours)
My test system is a Sempron 1800MHz 64-bit processor, and I wrote my programs in Python
A naive program which calculates 1 * 2 *... * 100000 just as written takes 54s CPU time.
Another program which uses a "divide and conquer" approach takes 2.6s CPU time. Once again, it's clear that algorithm can make a much bigger difference than CPU speed.
The "divide and conquer" method creates a list of all the numbers to be multiplied. At each stage, it divides this list into two sublists of equal length, and multiplies the pairs to create a new list. If there's an odd item, it's simply placed on the new list. This means that the intermediate products stay smaller longer.
If you *are* timing the conversion and output stage, then you should use a math package that is designed to create decimal output. I used "DecInt", and added a step which converted Python longs to DecInts when they appeared in the intermediate results. The run time is a bit longer, but the result is still calculated and written to a file in just 5.5s CPU time.
Here's the best I can do at getting my code into this post without hitting the fitlers:
def f(n): nums = range(2, n+1) while len(nums) > 1: . print len(nums) . hl = len(nums)/2 . if len(nums) % 2: .. a = nums[:hl] .. b = nums[hl:2*hl]; b.reverse() .. odd = nums[2*hl] .. nums = [odd] + [i*j for i, j in zip(a, b)] . else: .. a = nums[:hl] .. b = nums[hl:2*hl]; b.reverse() .. nums = [i*j for i, j in zip(a, b)] return nums[0]
it's nice that your email client can authenticate from your login
You can have this, today, on any modern Linux system. First, set up pam_ssh, so that you log into X with your ssh passphrase, and get an SSH agent automatically configured.
Then, you can start by using sshfs to mount filesystems using the credentials you established at login.
To authenticate to your imap server, use a highly-configurable client like mutt, and configure it to use imap-over-ssh:
set tunnel="ssh mail/usr/sbin/imapd"
Now, the only two passwords I enter are the SSH passphrase at login, and the Master Password for Firefox (and I wish it were possible to get rid of that one too; you could if you used an encrypted home directory)
Inferior software may not be able to do all these things, and there's certainly no point-and-drool interface to configuring all this stuff. But believe me, it works dandy.
One poster writes "it looks like there may already be some interest in Ceragenix Pharmaceuticals' OTC stock", and others note that the amazing new substance is from "a family of compounds called Ceragenins". Gee, looks like somebody accidentally published a press release, and then somebody else "greenlighted" it here on slashdot. In any case, there's no science here.
I hope that we at least beat digg.com on this non-science, non-news item.
You identified some elements that could be written as simple unix scripts.
Serializing cron jobs that access some resource? Use "lockfile". Wrap it with something that claims the necessary locks in the right order, runs the real script, and then removes all the locks.
Only producing output in the case of an error? Write a script that saves output to a temporary file, and then cats the file if the exit value is not "success".
These should both be pretty basic tasks that can be done in shell, python, or perl. And you can take them with you t
but which of those compressed formats is not proprietary? MP3 and AAC are covered by patents. I don't know about Audible formats, but several web pages refer to it as "one of four Audible.com proprietary formats".
Actually, I love our makefiles. We are using GNU make. It's a non-recursive build, so every time you invoke "make", you know the results are right.
get out of RCS and into CVS
While it sounds tempting, there's really no use for CVS in our shop. There are no off-site developers (so no need for cvs-over-ssh or the like), and the makefile-supported "workarea" concept is a better way for us to allow each developer to have independent changes. [see below]
[...] and do the builds there
If only this were a reasonable thing to do. The main problem is that a few header files are both frequently-modifed and unversally-included. On a single 2GHz machine, it takes multiple hours to do a build. (Oh, I should thank g++ for also being a speed demon too) On the development cluster (with a dual 2.8GHz machine running the build), distcc brings this down to 20 minutes instead.
If the header-file catastrophe could be changed, maybe it would make sense to have a tree per developer. But as it is, a tree per developer would mean that whenever you did a "cvs update" you could make a good bet that the rest of your work day will be shot waiting for the next build to complete. (and I don't see how that other favorite compiler-wrapper, ccache, could help here)
ObYearDropping: Programming since 1985, Linux since 1993
My day job is a large C++/tcl application with thousands of source files. We build for Unix and for Windows.
Version control is done via RCS. The build is a non-recursive Makefile with correct dependency-checking. Individual programmers do not have complete copies of the source tree, but can substitute their working files for trunk files when linking the binary.
We use distcc and cross-compilers to build all our platforms.
One build step uses wine to run a Windows program, and another build step invokes msvc++ v6 in wine to build a DLL related to some third-party C++ software.
Our open-source, third-party libraries are rebuilt by a separate non-recursive-makefile, also using distcc and cross-compilers.
The biggest current problem with the setup is that a do-nothing build takes about 10 seconds, build 1 object + link takes nearly 1 minute, and all compiles in the office are serialized since they write to the same repository of object files.
Anyway, in summary, keep the horror that is Windows development far away from you, and build your.EXE file using Open Source software on Unix. And get some other poor fellow to do the testing on Windows.
While I still have a 4-year-old HP inkjet as my home printer, I plan to replace it with a laser. While I originally bought this printer to print photos, nowadays I get photos printed at Walgreens down the street. (1-hour service, "light-jet" style process) Someday I'll do a "digital picture frame" like another poster suggests.
CSS columns will be great for documents printed from html source, but I dread seeing webpages formatted for the screen using columns (all of which are about 100 pixels taller than the browser window, of course).
I honestly don't know exactly how it would work, but I believe that all or nearly all copyrighted material (books, music, software) should be "sell it or lose it": if the copyright owner doesn't care to sell copies of it, for instance because it is not deemed profitable, the copyright lapses.
Clearly, if it's not worth selling, the copyright holder doesn't lose much if the copyright lapses early, 5Nth anniversary precious metal editions notwithstanding. Of course, companies who are in the business of selling the same pile of tripe every few years with a different name would suffer. (Quicken 2005? No thanks, I like Quicken 2000 just fine and don't want to learn anything new. What, you mean I can't buy Quicken 2000? I think some music labels and book publishers would find themselves in the same bind)
This belief is what makes me feel not at all bad when downloading abandonware games to play on my Commodore 64 emulator, for instance.
Failing "Sell It or Lose It Copyright", I'd love to see a non-profit corporation in the business of buying the copyright to abandoned software, particularly games, and releasing it to the public domain. In my mind this would involve finding out what copyrighted items people were most interested in, reaching a deal with the owner, and then raising the money online. I have no idea whether it would work, but I'd love to see it tried. I'd put up a few bucks to see EA's 8-bit software collection enter the public domain, and surely a lot of geeks would do the same. Would it add up to the piles of cash Electronic Arts would demand? Well, I don't know.
You can get a small microcontroller (such as an 8-pin AVR AT90S2323) and create a firmware that speaks enough of the PS/2 protocol to be recognized as a keyboard. The parts count should be fairly low, probably just the PS/2 plug, the microcontroller, and a ceramic resonator. The micrcontroller I mentioned has an internal I/O pullups.
You're likely to end up with a spare I/O port on that 8-pin microcontroller, so you could expand the design to include a button that sends ctrl+alt+del (or, for linux, alt+sysrq sequences to unmount and reboot), or an LED caps-lock indicator that you could control from software on the PC. With the AT90S2343 you would get rid of the external clock and get 2 additional I/O ports
Here's an existing project with source (license unclear) that uses AVR and emulates the PS2 keyboard protocol:
AVR microcontrollers are great. Few external parts, lots of documentation, develop with the GNU toolchain (including a fairly complete C library), and easy-to-build "SPI" programming hardware
When Bind can take 10,000 requests per second on a dual Xeon box (used for MAPS) and not melt into a smoky plastic dog treat, let me know.
10k requests per second doesn't sound like that much.
Just for fun, I wrote a program that calls Python's socket.gethostbyname() 10,000 times, and ran it against a very old (AMD K6, 350MHz, single processor, 384 megs RAM) machine on the local network. The runtime was 13 seconds, and named got about 90% CPU usage during that time.
If your modern dual Xeon box isn't 13 times faster than this years-old box, I kinda feel like you must have gotten ripped off. Y'know, Moore's Law and all.
Slashdot Mirror
Indeed. 1000 simple CPUs will fit in a FPGA, though it might require one near the top of the line. (e.g., picoblaze reportedly needs 96 "slices" and 1.5 "block RAMs"; the biggest Virtex-7 FPGA has more than 1400x as many block RAMs and 3100x as many "slices") There's little doubt that you could program a DCT for a picoblaze, if you wanted to.
It's hard to tell what 5.0GBps refers to -- the bitrate of the incoming, uncompressed, RGB video data? If so, that's maybe about 800FPS of 1080P video. In a circa 2002 paper, FPGAs were doing 100FPS HDTV DCT; an improvement of only a factor of 8 in the intervening 8 years would be frankly disheartening. Especially given that DCT of a frame of video is embarassingly parallel. The FPGA I mentioned earlier could hold 180 copies of the DCT from this paper; right there you have 18kFPS without even raising the clock. But the multipliers can probably also be clocked faster now...
You've just described the FPGA. Large areas of an FPGA are devoted to thousands of almost-identical functional blocks ("slices" in xilinx parlance). For instance, in one Xilinx family, a slice contains a 4-input LUT, a flip-flop (1 bit of memory, called an FF), and other specific gates that help implement things like carry chains, shift registers, and some 5+input functions the chip designers thought were commonly encountered.
Other areas contain "block RAMs" and "DSP cores" (basically, dedicated multipliers).
But now you've got yourself a lot of hard problems to solve: how to dice a program into something that is represented by essentially LUTs and FFs. how to recognize when a special function outside the LUT, like a carry chain, should be used. how to efficiently route the signals from where they're produced to where they're consumed. how to actually implement an efficient LUT where the contents are field-programmed. how to figure figure out what speed to clock the whole thing at so that it operates properly. how to read in the configuration to the chip. This is an enormous investment in research and software, and you still have to target the chip with languages that are totally alien to your typical C/C++ programmer.
As far as I know there is no production FPGA that you can write software for without using proprietary software. (though often the software can be obtained at no cost, at least for the non-flagship FPGA chips) This is partially because the details of bitstream structure are trade secrets of the respective FPGA companies, but also partially due to the inherent difficulty of the task.
Two human users (web+email+etc), three websites (around 115k hits/month total), and cvs server for an open source project (a few dozen "cvs up" a day).
If I want to download porn until it chafes, I can easily get to just under my ISP's monthly cap of 30GB/month, after which it's 0.50USD/GB.
Fedora is an upgrade treadmill. With Fedora, you're stuck upgrading every 12 months or so, or you can't get security updates anymore. With Fedora, install an LTS version and you're covered for 5 years on the server. That's why I switched.
And Microsoft Word does? When you type "he took a bow", does Microsoft silently add a possibly-incorrect annotation that you can't see, or does Clippy pop up and ask you how to pronounce the word you just typed? Also, it looks like you mean "homonym" or "homograph", not "homophonic".
It figures---the post with the useless numbers was modded up to 5, but the important one that gave the "methodology" wasn't moderated up past 2, so it went unseen (by me, at any rate). It doesn't look like Windows CALC.EXE even computes the full value of factorials, judging from the way it displays 100! in scientific notiation. This truly is the contender for most worthless benchmark! Though, to bring this back around to increasing performance, I suggest the NOPsort. It's a version of bogosort optimized for overclocked computers. First, NOP is executed, which may sort the array. Then, the order of the array is checked. If the process never completes, just overclock your computer some more. I'll wait for someone else to publish the first NOPsort benchmark, though.
My test system is a Sempron 1800MHz 64-bit processor, and I wrote my programs in Python
A naive program which calculates 1 * 2 * ... * 100000 just as written takes 54s CPU time.
Another program which uses a "divide and conquer" approach takes 2.6s CPU time. Once again, it's clear that algorithm can make a much bigger difference than CPU speed.
The "divide and conquer" method creates a list of all the numbers to be multiplied. At each stage, it divides this list into two sublists of equal length, and multiplies the pairs to create a new list. If there's an odd item, it's simply placed on the new list. This means that the intermediate products stay smaller longer.
If you *are* timing the conversion and output stage, then you should use a math package that is designed to create decimal output. I used "DecInt", and added a step which converted Python longs to DecInts when they appeared in the intermediate results. The run time is a bit longer, but the result is still calculated and written to a file in just 5.5s CPU time.
Here's the best I can do at getting my code into this post without hitting the fitlers:
You can have this, today, on any modern Linux system. First, set up pam_ssh, so that you log into X with your ssh passphrase, and get an SSH agent automatically configured.
Then, you can start by using sshfs to mount filesystems using the credentials you established at login.
To authenticate to your imap server, use a highly-configurable client like mutt, and configure it to use imap-over-ssh:
Now, the only two passwords I enter are the SSH passphrase at login, and the Master Password for Firefox (and I wish it were possible to get rid of that one too; you could if you used an encrypted home directory)
Inferior software may not be able to do all these things, and there's certainly no point-and-drool interface to configuring all this stuff. But believe me, it works dandy.
One poster writes "it looks like there may already be some interest in Ceragenix Pharmaceuticals' OTC stock", and others note that the amazing new substance is from "a family of compounds called Ceragenins". Gee, looks like somebody accidentally published a press release, and then somebody else "greenlighted" it here on slashdot. In any case, there's no science here.
I hope that we at least beat digg.com on this non-science, non-news item.
... the time when the mark of a real computer was that you couldn't hook it up to your TV, unlike your Commodore 64?
If this "private" flag relies on cooperation from clients, then it is broken.
A few years ago, some friends and I tried combining cold coffee with carbonated water from a soft-drink machine.
Because of the taste of the stuff, we christened it "Coffee NT", which stood for "Not Tasty".
Perhaps we were missing the secret ingredient, though. Corn syrup. Lots of corn syrup.
the subject says it all .. please replace TFA with one written by a clue-holder.
You identified some elements that could be written as simple unix scripts.
Serializing cron jobs that access some resource? Use "lockfile". Wrap it with something that claims the necessary locks in the right order, runs the real script, and then removes all the locks.
Only producing output in the case of an error? Write a script that saves output to a temporary file, and then cats the file if the exit value is not "success".
These should both be pretty basic tasks that can be done in shell, python, or perl. And you can take them with you t
but which of those compressed formats is not proprietary? MP3 and AAC are covered by patents. I don't know about Audible formats, but several web pages refer to it as "one of four Audible.com proprietary formats".
I read the paper. They devote, oh, a page or so to attacks. Proven as secure as AES? bah.
Actually, I love our makefiles. We are using GNU make. It's a non-recursive build, so every time you invoke "make", you know the results are right.
While it sounds tempting, there's really no use for CVS in our shop. There are no off-site developers (so no need for cvs-over-ssh or the like), and the makefile-supported "workarea" concept is a better way for us to allow each developer to have independent changes. [see below]
If only this were a reasonable thing to do. The main problem is that a few header files are both frequently-modifed and unversally-included. On a single 2GHz machine, it takes multiple hours to do a build. (Oh, I should thank g++ for also being a speed demon too) On the development cluster (with a dual 2.8GHz machine running the build), distcc brings this down to 20 minutes instead.
If the header-file catastrophe could be changed, maybe it would make sense to have a tree per developer. But as it is, a tree per developer would mean that whenever you did a "cvs update" you could make a good bet that the rest of your work day will be shot waiting for the next build to complete. (and I don't see how that other favorite compiler-wrapper, ccache, could help here)
ObYearDropping: Programming since 1985, Linux since 1993
My day job is a large C++/tcl application with thousands of source files. We build for Unix and for Windows.
.EXE file using Open Source software on Unix. And get some other poor fellow to do the testing on Windows.
Version control is done via RCS. The build is a non-recursive Makefile with correct dependency-checking. Individual programmers do not have complete copies of the source tree, but can substitute their working files for trunk files when linking the binary.
We use distcc and cross-compilers to build all our platforms.
One build step uses wine to run a Windows program, and another build step invokes msvc++ v6 in wine to build a DLL related to some third-party C++ software.
Our open-source, third-party libraries are rebuilt by a separate non-recursive-makefile, also using distcc and cross-compilers.
The biggest current problem with the setup is that a do-nothing build takes about 10 seconds, build 1 object + link takes nearly 1 minute, and all compiles in the office are serialized since they write to the same repository of object files.
Anyway, in summary, keep the horror that is Windows development far away from you, and build your
While I still have a 4-year-old HP inkjet as my home printer, I plan to replace it with a laser. While I originally bought this printer to print photos, nowadays I get photos printed at Walgreens down the street. (1-hour service, "light-jet" style process) Someday I'll do a "digital picture frame" like another poster suggests.
Why not firewall the chat services, if that is seen as a problem?
Second option, make whatever directories firefox installs extensions into non-writable.
Third option, refresh that directory from a fresh copy each time firefox is installed (don't all extensions require a restart?)
Software called "ALE" has an implementation of this general idea. ALE is GPL software.
ALE homepage
CSS columns will be great for documents printed from html source, but I dread seeing webpages formatted for the screen using columns (all of which are about 100 pixels taller than the browser window, of course).
If you want to know more, you might read
draft of css standard for multicolumn layout
Mozilla bug for this enhancement
I honestly don't know exactly how it would work, but I believe that all or nearly all copyrighted material (books, music, software) should be "sell it or lose it": if the copyright owner doesn't care to sell copies of it, for instance because it is not deemed profitable, the copyright lapses.
Clearly, if it's not worth selling, the copyright holder doesn't lose much if the copyright lapses early, 5Nth anniversary precious metal editions notwithstanding. Of course, companies who are in the business of selling the same pile of tripe every few years with a different name would suffer. (Quicken 2005? No thanks, I like Quicken 2000 just fine and don't want to learn anything new. What, you mean I can't buy Quicken 2000? I think some music labels and book publishers would find themselves in the same bind)
This belief is what makes me feel not at all bad when downloading abandonware games to play on my Commodore 64 emulator, for instance.
Failing "Sell It or Lose It Copyright", I'd love to see a non-profit corporation in the business of buying the copyright to abandoned software, particularly games, and releasing it to the public domain. In my mind this would involve finding out what copyrighted items people were most interested in, reaching a deal with the owner, and then raising the money online. I have no idea whether it would work, but I'd love to see it tried. I'd put up a few bucks to see EA's 8-bit software collection enter the public domain, and surely a lot of geeks would do the same. Would it add up to the piles of cash Electronic Arts would demand? Well, I don't know.
You can get a small microcontroller (such as an 8-pin AVR AT90S2323) and create a firmware that speaks enough of the PS/2 protocol to be recognized as a keyboard. The parts count should be fairly low, probably just the PS/2 plug, the microcontroller, and a ceramic resonator. The micrcontroller I mentioned has an internal I/O pullups.
w ord_is_id=1&keywords=46
You're likely to end up with a spare I/O port on that 8-pin microcontroller, so you could expand the design to include a button that sends ctrl+alt+del (or, for linux, alt+sysrq sequences to unmount and reboot), or an LED caps-lock indicator that you could control from software on the PC. With the AT90S2343 you would get rid of the external clock and get 2 additional I/O ports
Here's an existing project with source (license unclear) that uses AVR and emulates the PS2 keyboard protocol:
http://www.avrfreaks.net/Freaks/freakshow.php?key
http://members.rogers.com/nlange/avrStuff.html
AVR microcontrollers are great. Few external parts, lots of documentation, develop with the GNU toolchain (including a fairly complete C library), and easy-to-build "SPI" programming hardware
10k requests per second doesn't sound like that much.
Just for fun, I wrote a program that calls Python's socket.gethostbyname() 10,000 times, and ran it against a very old (AMD K6, 350MHz, single processor, 384 megs RAM) machine on the local network. The runtime was 13 seconds, and named got about 90% CPU usage during that time.
If your modern dual Xeon box isn't 13 times faster than this years-old box, I kinda feel like you must have gotten ripped off. Y'know, Moore's Law and all.