IPv4 and IPv6 support basic QoS. Also VOIP does not have to run over the public Internet. You can build a separate IP network to get bounded latency characteristics.
While there are no major technological advantages VOIP is certainly not worse than packet switched POTS.
PCS service in LA is $30/mo for 2800 night/weekend minutes, 200 any time, free LD.
I guess Vonage makes sense for business users but wireless still seems like a better deal for individuals interested in emergency and/or basic night/weekend social use.
OSS does not try to extract value from duplication because duplication itself has zero cost.
That fact hardly means that no one is being paid; it's simply a rational business move which accomodates reality.
In the next 20 years there will be more money in OSS than in traditional commercial software. The big players who aren't in denial are already preparing for this future.
Pretty soon, we'll see if Microsoft's treasure chest is big enough to trump inevitability...
Oh, believe me, we tried to force Torvalds into early retirement. But do you have *any* idea how hard it is to hire a puppet saboteur who's also named Linus?
This man's one true gift is product naming.
Nonetheless, victory shall be ours.
Signed,
mcbride@i-hate-linus.com
P.S. Can I borrow $699? The lawyers took my lunch money again, and Bill isn't answering his cell...
1. Fix X to be fast, non-bloated. 2. Fix KDE to be fast, non-bloated. 3. Fix Gnome to be fast, non-bloated. 4. Fix Mozilla to be fast, non-bloated. 5. Fix OpenOffice^W^W Write a new Office Suite.
Not trolling: Go install Fedora and see how it runs on a three year old machine. There's quite a lot of work to do.
Repeatedly re-inventing the wheel creates more work for engineers. So human survival instincts explain why we need thirty slightly different network card architectures.
We'll get more efficient hardware once machines begin to self-propagate. The Matrix told me so.
One problem is that modern computers are very I/O bound. C++ tends to produce larger binaries than an equivalent C implementation, meaning longer loads off HD and longer loads from DRAM into CPU cache.
Your point about buying faster hardware is widely accepted. I just wonder what computing would be like if programmers didn't rely on hardware advances to cover up their own incompetence and laziness.
One of Win32's advantages is that it was largely developed in 1989-1996. KDE and Gnome started around 2000 and don't scale down like Win95 or Win98.
Ever try mozilla on a P90 with 16MB? Swapalicious.
1. Window dragging is clearly slower in XFree86 than Windows. I'm also tired of arguing than F1 car is faster than an Amish horse-drawn wagon so I'm not even gonna try.
2. KDE and Gnome are shiny in the "let's copy Microsoft" way. When you play catch-up, you put yourself in second place and hand them the first place medal.
3. Thankfully Windows C++ code has made it possible for me to buy a supercomputer for $200. Given this situation I do not wish to argue.
4. I've seen plenty of upgrade failures and complete dependency fsckups in apt. Debian's packaging is, in the wild, much less reliable than the zealots believe. dselect is also way too slow (C++ argument).
5. 2-3 seconds is too slow. Computer hardware is fast enough now that there should be no user-perceptible delays. Also, in Linux, when one mozilla window locks up, they all do, and I lose my entire session. Cascading failure is unacceptable but mozilla is so bloated that 1GB of RAM is too little to create a separate process space for each window.
7. OS/2's UI was IMO better than anything for Linux right now. Given the amount of interest in Linux, though, we have a chance to far exceed OS/2.
8. HTML, not ASCII.
9. Packaging does not have to be complicated. Take a look at http://www.u-os.org/ (complete packager in 100k of C)
1. Making it "easy" to build layers-upon-layers of often-useless abstraction means less efficient code and encourages developers to lose sight of overall system architecture. KDE proves this point. How long does it take to compile? How many megs of memory does it use?
2. You can certainly do "object oriented design" in C. C just forces you to abandon the OO concepts which do not map cleanly to hardware, and thus are inefficient and should not be used.
3. With C you end up with portable, efficient code. C++ portability is almost non-existent due to the STL and complexity of the language. C++ is a maintenance nightmare because there are ten approaches to every problem, and if you have ten programmers they will each chose a different one.
4. C is certainly easier to compile than C++ and the result is often more efficient than hand-coded assembly. As compilers improve, C will get even better. Much of C++ is built on constructs which do not map easily to hardware so compiler progress will be much slower.
I've never met a C++ programmer who thinks in terms of cache utilization and pipeline stalls. Maybe the C people just tend to understand the hardware better, and thus produce better code.
Free software developers need to be honest with themselves. No Linux environment is yet on par with Windows for typical desktop use. Here are the missing characteristics:
1. direct-render windowing system. XFree86's architecture is unfixably inefficient for typical desktop users. Window dragging in X is clearly much less responsive than on Win32 due to the silly message encode/decode overhead, context switches, and inter-process copies. DirectFB provides some hope here but their multi-window code is still immature. Network transparency should still be supported (and better than Win32 VNC), but not used for the local desktop.
2. Consistent "ooh, shiny" widget set. The computing public expects computers to make them feel futuristic and sexy, not like dateless engineering nerds. WinXP's advantage over Win2K is purely visual--there were basically no interesting technical changes (proof that Microsoft is a stagnant market-driven company). With ATI releasing specs and NVIDIA losing market share to the point of irrelevance, free software can leverage hardware acceleration to build sleak UIs.
3. Much higher efficiency. Practically speaking that means straight C. Nobody has figured out how to architect, write, maintain, or compile C++ efficiently. Don't use it.
4. Avoid shared library hell. Gnome seems to require about 200 shared libraries, which slows down the dynamic linker and creates a maintenance and installation nightmare (GNUCash).
5. fast, efficient browser. mozilla is a slow bloated tribute to the horror of C++ software engineering. The dillo folks have the right idea but are way behind IE. A free browser can ignore javascript, java, flash, and ActiveX as these are not critical to building a viable desktop.
6. media players. mplayer is better than anything availible for Windows. no problem here.
7. Full Win32 compatability. Wine is making great progress here. Once Win32 apps (especially games) run at native speed it's all over.
8. Office Suite. OpenOffice is a horrid, bloated mess. AbiWord is better. Free software developers are wasting their time on backwards compatability. Develop a better system, and people will switch. It is always possible to dump the useful contents of a Word doc to plaintext or HTML using Word itself, so free software developers should not waste time trying to reverse engineer proprietary formats. Develop something faster, cheaper, more stable, better, and let users deal with the conversion.
9. package management. Instead of forcing developers to write nasty packaging scripts, design a system which takes a tar.gz URL, auto-calculates dependencies based on autoconf, then installs and manages the program transparently. Once that happens you've freed 10,000 Debian maintainers for more useful work.
10. no open ports and state-based firewalling. It's trivial to beat Microsoft here.
When companies have a significant competitive advantage using a free desktop the ones who don't switch will go out of business. The develoeprs of this system will have guaranteed, well-paying jobs. revolution complete.
Your suggested optimization already happens at the operating system level in modern OSes. It's called demand paging.
When a program enters a function in a dynamic shared object (DSO), the ~4kB pages which contain that function are pulled off disk (assuming they weren't already present in the file system cache). In general, user code or data which is not accessed ("faulted in") does not consume physical memory. The average program contains lots of code that is rarely executed, and allocates memory which is never actually used, so this optimization is pretty significant.
Take a look in "ps aux". This behavior shows up as the difference between a process's "virtual" size (VSZ) and its "real" size (RSS).
Xilinx sales guys have been barking up that tree for a while.
There are two practical problems: interchip I/O is extremely expensive in modern computing architectures, and FPGA clock rates are too low. By the time you get the data into and out of an FPGA running at 200 MHz, you typically could have done the whole algorithm in software (3 GHz assuming your working set fits in cache).
Someone with excellent fabs would need to build a CPU+FPGA onto one piece of silicon for this to be worthwhile. Once that happens, I/O to the FPGA could become the same cost as I/O to cache, and reprogramming latency would also become more reasonable...
Slashdot Mirror
"Mouse has moving. Humbly please clicking here to make honorfully rebooted"
Whew, good thing that future didn't happen!
IPv4 and IPv6 support basic QoS. Also VOIP does not have to run over the public Internet. You can build a separate IP network to get bounded latency characteristics.
While there are no major technological advantages VOIP is certainly not worse than packet switched POTS.
I'm surprised the Vonage rates are that high.
PCS service in LA is $30/mo for 2800 night/weekend minutes, 200 any time, free LD.
I guess Vonage makes sense for business users but wireless still seems like a better deal for individuals interested in emergency and/or basic night/weekend social use.
What? You mean it's 1999 again? Thanks, Santa!
All those engineers are working for free?
OSS does not try to extract value from duplication because duplication itself has zero cost.
That fact hardly means that no one is being paid; it's simply a rational business move which accomodates reality.
In the next 20 years there will be more money in OSS than in traditional commercial software. The big players who aren't in denial are already preparing for this future.
Pretty soon, we'll see if Microsoft's treasure chest is big enough to trump inevitability...
Mod +1 (Marketoid Buzzword Compliant)
Call Tyson.
Which one of his songs would you say you like best?
You seem quite happy in the Marketrix.
But I have to ask: Why didn't you just buy a desktop with that $500? I somehow doubt Dell support is better than a _redundant machine_
quit trolling. dare to compare:
http://www.xiph.org/ogg/vorbis/listen.html
You never forget your first--this post's for you, D.
Actually, Microsoft is responsible for the dirt cheap commodity x86 hardware that powers most Linux machines.
Shut Up. Bill Gates is saving you money!
Oh, believe me, we tried to force Torvalds into early retirement. But do you have *any* idea how hard it is to hire a puppet saboteur who's also named Linus?
This man's one true gift is product naming.
Nonetheless, victory shall be ours.
Signed,
mcbride@i-hate-linus.com
P.S. Can I borrow $699? The lawyers took my lunch money again, and Bill isn't answering his cell...
1. Fix X to be fast, non-bloated.
2. Fix KDE to be fast, non-bloated.
3. Fix Gnome to be fast, non-bloated.
4. Fix Mozilla to be fast, non-bloated.
5. Fix OpenOffice^W^W Write a new Office Suite.
Not trolling: Go install Fedora and see how it runs on a three year old machine. There's quite a lot of work to do.
Also:
6. Standardize on one version of Solitaire.
Repeatedly re-inventing the wheel creates more work for engineers. So human survival instincts explain why we need thirty slightly different network card architectures.
We'll get more efficient hardware once machines begin to self-propagate. The Matrix told me so.
Wow. You get paid $4000/day to suggest putting XML in the kernel?
Screw senior software engineer! How do I get started in marketing?
One problem is that modern computers are very I/O bound. C++ tends to produce larger binaries than an equivalent C implementation, meaning longer loads off HD and longer loads from DRAM into CPU cache.
Your point about buying faster hardware is widely accepted. I just wonder what computing would be like if programmers didn't rely on hardware advances to cover up their own incompetence and laziness.
One of Win32's advantages is that it was largely developed in 1989-1996. KDE and Gnome started around 2000 and don't scale down like Win95 or Win98.
Ever try mozilla on a P90 with 16MB? Swapalicious.
1. Window dragging is clearly slower in XFree86 than Windows. I'm also tired of arguing than F1 car is faster than an Amish horse-drawn wagon so I'm not even gonna try.
2. KDE and Gnome are shiny in the "let's copy Microsoft" way. When you play catch-up, you put yourself in second place and hand them the first place medal.
3. Thankfully Windows C++ code has made it possible for me to buy a supercomputer for $200. Given this situation I do not wish to argue.
4. I've seen plenty of upgrade failures and complete dependency fsckups in apt. Debian's packaging is, in the wild, much less reliable than the zealots believe. dselect is also way too slow (C++ argument).
5. 2-3 seconds is too slow. Computer hardware is fast enough now that there should be no user-perceptible delays. Also, in Linux, when one mozilla window locks up, they all do, and I lose my entire session. Cascading failure is unacceptable but mozilla is so bloated that 1GB of RAM is too little to create a separate process space for each window.
7. OS/2's UI was IMO better than anything for Linux right now. Given the amount of interest in Linux, though, we have a chance to far exceed OS/2.
8. HTML, not ASCII.
9. Packaging does not have to be complicated. Take a look at http://www.u-os.org/ (complete packager in 100k of C)
1. Making it "easy" to build layers-upon-layers of often-useless abstraction means less efficient code and encourages developers to lose sight of overall system architecture. KDE proves this point. How long does it take to compile? How many megs of memory does it use?
2. You can certainly do "object oriented design" in C. C just forces you to abandon the OO concepts which do not map cleanly to hardware, and thus are inefficient and should not be used.
3. With C you end up with portable, efficient code. C++ portability is almost non-existent due to the STL and complexity of the language. C++ is a maintenance nightmare because there are ten approaches to every problem, and if you have ten programmers they will each chose a different one.
4. C is certainly easier to compile than C++ and the result is often more efficient than hand-coded assembly. As compilers improve, C will get even better. Much of C++ is built on constructs which do not map easily to hardware so compiler progress will be much slower.
I've never met a C++ programmer who thinks in terms of cache utilization and pipeline stalls. Maybe the C people just tend to understand the hardware better, and thus produce better code.
Free software developers need to be honest with themselves. No Linux environment is yet on par with Windows for typical desktop use. Here are the missing characteristics:
1. direct-render windowing system. XFree86's architecture is unfixably inefficient for typical desktop users. Window dragging in X is clearly much less responsive than on Win32 due to the silly message encode/decode overhead, context switches, and inter-process copies. DirectFB provides some hope here but their multi-window code is still immature. Network transparency should still be supported (and better than Win32 VNC), but not used for the local desktop.
2. Consistent "ooh, shiny" widget set. The computing public expects computers to make them feel futuristic and sexy, not like dateless engineering nerds. WinXP's advantage over Win2K is purely visual--there were basically no interesting technical changes (proof that Microsoft is a stagnant market-driven company). With ATI releasing specs and NVIDIA losing market share to the point of irrelevance, free software can leverage hardware acceleration to build sleak UIs.
3. Much higher efficiency. Practically speaking that means straight C. Nobody has figured out how to architect, write, maintain, or compile C++ efficiently. Don't use it.
4. Avoid shared library hell. Gnome seems to require about 200 shared libraries, which slows down the dynamic linker and creates a maintenance and installation nightmare (GNUCash).
5. fast, efficient browser. mozilla is a slow bloated tribute to the horror of C++ software engineering. The dillo folks have the right idea but are way behind IE. A free browser can ignore javascript, java, flash, and ActiveX as these are not critical to building a viable desktop.
6. media players. mplayer is better than anything availible for Windows. no problem here.
7. Full Win32 compatability. Wine is making great progress here. Once Win32 apps (especially games) run at native speed it's all over.
8. Office Suite. OpenOffice is a horrid, bloated mess. AbiWord is better. Free software developers are wasting their time on backwards compatability. Develop a better system, and people will switch. It is always possible to dump the useful contents of a Word doc to plaintext or HTML using Word itself, so free software developers should not waste time trying to reverse engineer proprietary formats. Develop something faster, cheaper, more stable, better, and let users deal with the conversion.
9. package management. Instead of forcing developers to write nasty packaging scripts, design a system which takes a tar.gz URL, auto-calculates dependencies based on autoconf, then installs and manages the program transparently. Once that happens you've freed 10,000 Debian maintainers for more useful work.
10. no open ports and state-based firewalling. It's trivial to beat Microsoft here.
When companies have a significant competitive advantage using a free desktop the ones who don't switch will go out of business. The develoeprs of this system will have guaranteed, well-paying jobs. revolution complete.
Your suggested optimization already happens at the operating system level in modern OSes. It's called demand paging.
When a program enters a function in a dynamic shared object (DSO), the ~4kB pages which contain that function are pulled off disk (assuming they weren't already present in the file system cache). In general, user code or data which is not accessed ("faulted in") does not consume physical memory. The average program contains lots of code that is rarely executed, and allocates memory which is never actually used, so this optimization is pretty significant.
Take a look in "ps aux". This behavior shows up as the difference between a process's "virtual" size (VSZ) and its "real" size (RSS).
Xilinx sales guys have been barking up that tree for a while.
There are two practical problems: interchip I/O is extremely expensive in modern computing architectures, and FPGA clock rates are too low. By the time you get the data into and out of an FPGA running at 200 MHz, you typically could have done the whole algorithm in software (3 GHz assuming your working set fits in cache).
Someone with excellent fabs would need to build a CPU+FPGA onto one piece of silicon for this to be worthwhile. Once that happens, I/O to the FPGA could become the same cost as I/O to cache, and reprogramming latency would also become more reasonable...
Maybe they just lost the source code and wanted a disassembly.
HURD... Hardly Under Rapid Development
woah, they finally figured out how to read?