The author of the review (and apparently the author of the book as well) are proponents of the view that each tool should be as simple as possible and do one job well. That just doesn't make sense any more. Now we have large, monolithic command line tools that can do many jobs well.
Having one large, all-purpose tool is better because instead of having to learn many simple tools with different syntax and stringing them all together in a massive chain of fork()'s and file handles, we can use a single tool that is more sophisticated, more consistent, and more resource efficient for big tasks. As Rob Pike said in his interview here so recently, the days of one tool doing one job well "are dead and gone and the eulogy was delivered by Perl."
In Microsoft Word, I go to the "Tools" menu and choose "Word Count..."
For my current document, it reads:
Statistics:
Pages 23
Words 10,234
Characters (no spaces) 52,996
Characters (with spaces) 63,140
Paragraphs 107
Lines 660
One of the really nice things about GUI's is that they make rarely used commands (like word count) really easy to discover. You just look in the menu, and there it is. Compare that to a UNIX command prompt, where if you don't know how to write scripts and you don't know that the "wc" command is for word counts, you are lost.
It does have some interesting stuff...
on
MSN Search Roundup
·
· Score: 1
I'm not going to change my search to a beta product; I try to avoid beta products after too much of Netscape a few years ago. If it seems half baked, it could be because it is half baked. Give them time to finish it and maybe it will be worth using.
I am interested in the ranking features, though. Click on "Search Builder" and you then on "Results ranking" and you get some sliders for adjsting the display of pages. That seems like it could be very useful at times, in particular for those searches where there seem to be no good results (in the first few pages out of ten thousand).
I think I will check in on this thing once in a while to see if it improves. Maybe it will be good when it is done.
I've been unwilling to install any Real software on my computer for a long time, so I'm really glad that they ditched that. Windows media is my definite preference; I'm not even sure if my Windows computer plays ogg right now. Obviously I could download an ogg player if necessary, but I'm happy with what I've got now. Getting CBC in streaming media wouldn't count as necessary to me.
Anyone else notice that the Linux distribution they support is Gentoo? A curious choice, although one that makes me happy. I do have a question, though. I know that Red Hat, and perhaps some other distribution makers, modifies or doesn't include packages with questionable licensing (or obvious license violations in some cases). Does the version of mplayer included with every modern distribution support Windows media? It may well be that ogg support is the only way to support some Linux distributions.
They are going through all their old code, but that will only find implementation bugs, not design problems with the security model. They are also revisiting security models, and I expect they will have to change some number of them.
On the other hand, they have also shown quite a bit of innovation (can I use that word?) in dealing with the problem. One of the biggest is the need to run as Administrator in Windows, or lose the ability to run a lot of software. They've fixed that one in Windows 2003. The big reasons for running as Administrator are that applications want to write global registry settings and write potentially dangerous places in the filesystem. In Windows 2003 you can set compatibility mode to redirect writes to global locations so that they write to user specific locations instead. In the next consumer release of Windows that huge flaw in the security model will be gone, and normal users will run as normal users.
It's going to take a lot of time and effort, but I think that Microsoft will make major improvements to security in the next few years, including security model fixes and workarounds where necessary. It's the same now with security as it was with stability. If they don't improve, they will die.
Hmm... I don't think Microsoft would release a Linux distribution. If they did I think they wouldn't use it to promote Windows as the end goal. A theoretical Microsoft Linux distribution would probably be separate from the main Windows group, and would probably have the business goal of selling Microsoft Linux, not of getting people to migrate from Linux to Windows. I'm sure they would agressively target proprietary UNIX customers, though.
Just thinking of what the business case would be, I suppose they would have to implement at least.NET and a Win32 subsystem for Linux. They would have to convince the Office group and various server groups that supporting Microsoft Linux was a good case. The Win32 subsystem would be essential for convincing those groups that a port would be viable, given that supporting a new platform is both difficult and risky. If they could get those ports then Microsoft Linux might be very compelling. It would certainly attract a lot of businesses who are already examining Linux, but are tied to Office. It might also drive new sales of Microsoft server products. They are actually quite decent products, but a lot of people won't trust the Windows platform when they need high reliability. Running SQL Server or Exchange on Linux could be very compelling.
I really don't think they will do it. The NT kernel is quite different from Linux, and matching the capabilities on Linux would be difficult, and translating between them would probably involve a significant performance hit. It's an interesting idea, though, and if I were a strategist at Microsoft I would definitely look into it further. Making a Win32 layer for the Linux kernel might also be an interesting project for a Masters student, perhaps using the Windows CE sources as a starting point.
Microsft did make security a top priority. It's going to take years for Microsft to solve their security problems, but my hunch is that they can do it. They've done a decent job with stability of the operating system, and I think they can do just as well with security. If you compare the early results of their security effort (Windows 2003, IIS 6) with products that didn't get the security effort (Windows XP, IIS 5, IE 6), you will notice a very significant improvement. Microsoft is making a big effort on security and it is getting results. It's not there yet, but chances are it will get there.
Overall, I wasn't tremendously impressed with the email. I didn't see anything that was literally wrong, but parts of it were stupid or misleading or based on third party studies that weren't necessarily accurate. There were also some things that were completely true. If you can refute every single point in that email then you are making stuff up, and you should re-evaluate your arguments to determine which ones are true and which ones are nonsense.
Linux is a growing factor worldwide, but from what I've seen, Microsoft's doom is in the very distant future. For the moment it seems that they are still increasing their market share.
Apple continues to lose market share, as they have for a long time. Even with the introduction of Power Macintosh G5s, new Power Macintosh G5s, and the new iMac G5, growth of sales of Macintosh computers haven't kept up with the overall growth of the PC market. Apple is a profitable company in little danger of disappearing in the near future, but it is in even less danger of becoming a dominant force in the overall PC market.
Linux is gaining traction in industry, but it seems that it is still mostly in customers moving from other UNIX systems. Sun, HP, SGI and IBM are losing the low-end UNIX workstation market to Linux. The mass migration to Windows has stopped, but Windows is still taking a little bit of that. A more common path I've seen is a move from proprietary UNIX to Linux. With Linux comes cheap x86 machines. Shortly after the x86 machines arrive people start running Windows. That is, the migration goes proprietary UNIX->Linux->Windows. Overall I have seen more Linux to Windows migrations than the other way around.
I don't mean this as doom and gloom to Linux. Proportionally, Linux is growing much faster than Windows. Linux is also improving much faster than Windows. As the installed base grows bigger, more services will become available, and Windows to Linux migrations will become more viable.
Hmm... let me make a WAG. Microsoft is really concentrating on security now. I think that Microsoft will handle security issues just as well as they have handled stability issues. That is to say in ten years Windows security jokes will be a cliche that out-of-touch Slashdot readers make, and get corrected on by the more normal users.
At that point, there will be some other big issue. I propose (another WAG) that it will be how well the operating system supports new hardware models, and highly parallel personal computers in particular. Hardware will move from todays fast single-threaded processors to processors that run a single thread not much faster than today's processors, but can run many threads in parallel. Windows and Linux both have trouble with scalability now. In this future scenario it is quite possible that one operating system will be four times faster than the other. If you want Linux to beat Microsoft, make sure that Linux is the one that is four times faster.
I didn't know the compute nodes ran Linux as well; I just assumed that they ran a completely custom OS. It's not only virtual memory and fork() that is lacking; they don't have a scheduler or normal I/O facilities either.
The nodes share main memory, but are not cache coherent in L1 or L2. It really is an interesting design. They trade off a lot for maximum efficiency, with respect to space and performance. I suspect power was mainly a concern only as it affected the maximum possible density.
I suspect that all the specialization means that BlueGene/L will be harder to code for, and probably very special-purpose, even compared to other really big supercomputers. It is a very interesting system, though.
This result was from the partially completed cluster, at the beginning of October. At that time only 16 of the 20 machines were online. When the result is taken again with all 20 of the machines there will be a sizeable increase in that lead.
There's also a dark horse in the supercomputer race; a cluster of low-end IBM servers using PPC970 chips that is in between the BlueGene/L prototype and the Earth Simulator. That pushes the last Alpha machine off the top 5 list, and gives Itanium and PowerPC each two spots in the top 5. It's amazing to see the Earth Simulator's dominance broken so thoroughly. After so long on top, in one list it goes from first to fourth, and it will drop at least two more spots in 2005.
hybrid system with multiple kernels
on
Cray XT-3 Ships
·
· Score: 3, Informative
From the documents, it looks like it runs Linux on the management nodes and Catamount on the compute nodes. The idea is you can do what you like with the general purpose nodes, but for the compute nodes, you run a lightweight operating system that has low overhead, minimal services and predictable scheduling. BlueGene/L works the same way; it runs Linux on the management nodes and a custom operating system on the compute nodes. Compute nodes likely provide scheduling for only the number of threads that run on the node, communication through MPI and some proprietary API, and basic debugging facilities. Compute nodes probably lack normal OS services like network, disk, or even a console.
I ran a simple experiment to see how bad heat is. I left my Google branded lava lamp on until the temperature stabilized. It uses a 40 Watt light bulb, or a bit more than half what a brand new AMD processor puts out under sustained heavy use. Older AMD processors, and newer Intel processors, put out even more than that. Anyway, when it reached thermal equilibrium, the cool part at the top of the lamp was too hot to touch.
Compare that 350 square cm to the area of a processor core, about 1 square cm, with twice the total heat dissipation. You can see how removing the heat is a huge problem. Compare the hot spots of the core, like the scheduler and active functional units, which use much more energy per area than the average, and you can see the problem getting much worse.
A peltier integrated into a core would be all about moving heat away from the tiny core as quickly as possible to a larger surface, such as a heat sink contact plate and a heat sink, so that it can be dissipated quickly. Even if heat dissipation in future cores went down, with smaller processes leading to smaller cores, the problem of heat dissipation would still get bigger.
I know this is so late that everyone has moved on to the next story, but curious about the idea, I wrote a program to simulate the idea using my shadow mask CRT monitor, and compare it to downsampling and true 640x480. It may work on aperture grill and LCD monitors as well, but it probably won't look as good. Download here. Sorry; it's a windows binary only, and it requires.NET.
For best results set your resolution low, otherwise it has very visible moire patterns. As a side effect of the conversion, the image gets darker. My program also has a colour cast, which the article claims is due to adding the white pixel. The article also says that Samsung has overcome this problem.
It works by setting up the subpixels as a 640x480 square grid, with each pixel consisting of a starting pixel, and the right, lower, and lower right subpixels. Subpixel values are calculated using the average intensity of the corresponding colour value in each of the four pixels the subpixel is a part of.
Visually, aside from the darkness and colour cast which are artifacts of the simulation and wouldn't appear in the real product, it looks decent. It's blurrier than a true 640x480 display, but retains more detail than the 320x240 downsampled version.
Perhaps Konqueror is better than other browsers, or perhaps the involvement of Apple means that Safari is better tested than Mozilla or Opera.
Unfortunately, handling the test cases provided with the post doesn't mean it's ok. Browsers were run with randomly generated test cases, and the article says that usually a hundred or so tests were required to cause a crash. The probability of a poorly written browser crashing on any given case is low.
The chances of any single test case being problematic are low, so passing the cases provided doesn't mean Safari is safe. If someone downloaded the cgi that automatically generated tests and left Safari with that for a few hours without crashes, we could then say that it lacks the flaws that Mozilla and Opera have.
This post on BugTraq suggests that Mozilla and Firefox will have security problems when they get popular enough. The evidence is pretty compelling, too. Current testing procedures for Mozilla are obviously inadequate.
Slashdot Mirror
I'm containing my hope for sequels until I see that the first one doesn't suck. I fear that the sequels are much more likely.
Having one large, all-purpose tool is better because instead of having to learn many simple tools with different syntax and stringing them all together in a massive chain of fork()'s and file handles, we can use a single tool that is more sophisticated, more consistent, and more resource efficient for big tasks. As Rob Pike said in his interview here so recently, the days of one tool doing one job well "are dead and gone and the eulogy was delivered by Perl."
For my current document, it reads:
Statistics:
Pages 23
Words 10,234
Characters (no spaces) 52,996
Characters (with spaces) 63,140
Paragraphs 107
Lines 660
One of the really nice things about GUI's is that they make rarely used commands (like word count) really easy to discover. You just look in the menu, and there it is. Compare that to a UNIX command prompt, where if you don't know how to write scripts and you don't know that the "wc" command is for word counts, you are lost.
I am interested in the ranking features, though. Click on "Search Builder" and you then on "Results ranking" and you get some sliders for adjsting the display of pages. That seems like it could be very useful at times, in particular for those searches where there seem to be no good results (in the first few pages out of ten thousand).
I think I will check in on this thing once in a while to see if it improves. Maybe it will be good when it is done.
Anyone else notice that the Linux distribution they support is Gentoo? A curious choice, although one that makes me happy. I do have a question, though. I know that Red Hat, and perhaps some other distribution makers, modifies or doesn't include packages with questionable licensing (or obvious license violations in some cases). Does the version of mplayer included with every modern distribution support Windows media? It may well be that ogg support is the only way to support some Linux distributions.
On the other hand, they have also shown quite a bit of innovation (can I use that word?) in dealing with the problem. One of the biggest is the need to run as Administrator in Windows, or lose the ability to run a lot of software. They've fixed that one in Windows 2003. The big reasons for running as Administrator are that applications want to write global registry settings and write potentially dangerous places in the filesystem. In Windows 2003 you can set compatibility mode to redirect writes to global locations so that they write to user specific locations instead. In the next consumer release of Windows that huge flaw in the security model will be gone, and normal users will run as normal users.
It's going to take a lot of time and effort, but I think that Microsoft will make major improvements to security in the next few years, including security model fixes and workarounds where necessary. It's the same now with security as it was with stability. If they don't improve, they will die.
Just thinking of what the business case would be, I suppose they would have to implement at least .NET and a Win32 subsystem for Linux. They would have to convince the Office group and various server groups that supporting Microsoft Linux was a good case. The Win32 subsystem would be essential for convincing those groups that a port would be viable, given that supporting a new platform is both difficult and risky. If they could get those ports then Microsoft Linux might be very compelling. It would certainly attract a lot of businesses who are already examining Linux, but are tied to Office. It might also drive new sales of Microsoft server products. They are actually quite decent products, but a lot of people won't trust the Windows platform when they need high reliability. Running SQL Server or Exchange on Linux could be very compelling.
I really don't think they will do it. The NT kernel is quite different from Linux, and matching the capabilities on Linux would be difficult, and translating between them would probably involve a significant performance hit. It's an interesting idea, though, and if I were a strategist at Microsoft I would definitely look into it further. Making a Win32 layer for the Linux kernel might also be an interesting project for a Masters student, perhaps using the Windows CE sources as a starting point.
Overall, I wasn't tremendously impressed with the email. I didn't see anything that was literally wrong, but parts of it were stupid or misleading or based on third party studies that weren't necessarily accurate. There were also some things that were completely true. If you can refute every single point in that email then you are making stuff up, and you should re-evaluate your arguments to determine which ones are true and which ones are nonsense.
Apple continues to lose market share, as they have for a long time. Even with the introduction of Power Macintosh G5s, new Power Macintosh G5s, and the new iMac G5, growth of sales of Macintosh computers haven't kept up with the overall growth of the PC market. Apple is a profitable company in little danger of disappearing in the near future, but it is in even less danger of becoming a dominant force in the overall PC market.
Linux is gaining traction in industry, but it seems that it is still mostly in customers moving from other UNIX systems. Sun, HP, SGI and IBM are losing the low-end UNIX workstation market to Linux. The mass migration to Windows has stopped, but Windows is still taking a little bit of that. A more common path I've seen is a move from proprietary UNIX to Linux. With Linux comes cheap x86 machines. Shortly after the x86 machines arrive people start running Windows. That is, the migration goes proprietary UNIX->Linux->Windows. Overall I have seen more Linux to Windows migrations than the other way around.
I don't mean this as doom and gloom to Linux. Proportionally, Linux is growing much faster than Windows. Linux is also improving much faster than Windows. As the installed base grows bigger, more services will become available, and Windows to Linux migrations will become more viable.
Hmm... let me make a WAG. Microsoft is really concentrating on security now. I think that Microsoft will handle security issues just as well as they have handled stability issues. That is to say in ten years Windows security jokes will be a cliche that out-of-touch Slashdot readers make, and get corrected on by the more normal users.
At that point, there will be some other big issue. I propose (another WAG) that it will be how well the operating system supports new hardware models, and highly parallel personal computers in particular. Hardware will move from todays fast single-threaded processors to processors that run a single thread not much faster than today's processors, but can run many threads in parallel. Windows and Linux both have trouble with scalability now. In this future scenario it is quite possible that one operating system will be four times faster than the other. If you want Linux to beat Microsoft, make sure that Linux is the one that is four times faster.
The nodes share main memory, but are not cache coherent in L1 or L2. It really is an interesting design. They trade off a lot for maximum efficiency, with respect to space and performance. I suspect power was mainly a concern only as it affected the maximum possible density.
I suspect that all the specialization means that BlueGene/L will be harder to code for, and probably very special-purpose, even compared to other really big supercomputers. It is a very interesting system, though.
There's also a dark horse in the supercomputer race; a cluster of low-end IBM servers using PPC970 chips that is in between the BlueGene/L prototype and the Earth Simulator. That pushes the last Alpha machine off the top 5 list, and gives Itanium and PowerPC each two spots in the top 5. It's amazing to see the Earth Simulator's dominance broken so thoroughly. After so long on top, in one list it goes from first to fourth, and it will drop at least two more spots in 2005.
From the documents, it looks like it runs Linux on the management nodes and Catamount on the compute nodes. The idea is you can do what you like with the general purpose nodes, but for the compute nodes, you run a lightweight operating system that has low overhead, minimal services and predictable scheduling. BlueGene/L works the same way; it runs Linux on the management nodes and a custom operating system on the compute nodes. Compute nodes likely provide scheduling for only the number of threads that run on the node, communication through MPI and some proprietary API, and basic debugging facilities. Compute nodes probably lack normal OS services like network, disk, or even a console.
Compare that 350 square cm to the area of a processor core, about 1 square cm, with twice the total heat dissipation. You can see how removing the heat is a huge problem. Compare the hot spots of the core, like the scheduler and active functional units, which use much more energy per area than the average, and you can see the problem getting much worse.
A peltier integrated into a core would be all about moving heat away from the tiny core as quickly as possible to a larger surface, such as a heat sink contact plate and a heat sink, so that it can be dissipated quickly. Even if heat dissipation in future cores went down, with smaller processes leading to smaller cores, the problem of heat dissipation would still get bigger.
For best results set your resolution low, otherwise it has very visible moire patterns. As a side effect of the conversion, the image gets darker. My program also has a colour cast, which the article claims is due to adding the white pixel. The article also says that Samsung has overcome this problem.
It works by setting up the subpixels as a 640x480 square grid, with each pixel consisting of a starting pixel, and the right, lower, and lower right subpixels. Subpixel values are calculated using the average intensity of the corresponding colour value in each of the four pixels the subpixel is a part of.
Visually, aside from the darkness and colour cast which are artifacts of the simulation and wouldn't appear in the real product, it looks decent. It's blurrier than a true 640x480 display, but retains more detail than the 320x240 downsampled version.
Unfortunately, handling the test cases provided with the post doesn't mean it's ok. Browsers were run with randomly generated test cases, and the article says that usually a hundred or so tests were required to cause a crash. The probability of a poorly written browser crashing on any given case is low.
The chances of any single test case being problematic are low, so passing the cases provided doesn't mean Safari is safe. If someone downloaded the cgi that automatically generated tests and left Safari with that for a few hours without crashes, we could then say that it lacks the flaws that Mozilla and Opera have.
This post on BugTraq suggests that Mozilla and Firefox will have security problems when they get popular enough. The evidence is pretty compelling, too. Current testing procedures for Mozilla are obviously inadequate.