They have ported konsole... so the burning question is will Qt-console run correctly inside that? If so there is no longer any need for screen(1) or even window managers - konsole's screen-switching buttons can do the job of changing between windows.
Or maybe Emacs's M-x term, if anyone ported Emacs to Qt.
There was a similar thing done for Java's AWT toolkit a while ago (article in Dr Dobbs). And I know that aalib comes with an ASCII X server. I wonder which of these three gives the best results? Probably the Qt-console port because it is the most high-level.
A version of KDE using text-mode Qt would be very interesting to behold... or how about a 'desktop' using the aalib ASCII X server (for GTK, Motif etc. programs) plus Qt.
All we have to do now is persuade the Mozilla team to add text-mode display to their bloated 'chrome' rendering system. I'm sure they will be happy to add it as a new feature before 1.0, after all they've thrown in pretty much everything else.
I thought it was a reference to the Robin Williams character in 'Santa Claus: The Movie'. Which makes just about as much sense as any reference to Harry Potter or LoTR, given that we're supposed to be discussing Mac OS X.
I agree, it is a bit of an endless cycle. My feeling is just that Serial ATA is (a) not quite good enough to compete with SCSI in the 'let's build an intelligent disk subsystem and take load of the host processor' market, and (b) too complex for the 'let's build something really cheap and dirty' market. The original ATA disk controller could be built for about 50 cents in an ISA bus machine. Could there be an equivalent for PCI?
You could indeed stick it into a drive bay, if you can connect the PCI bus over the distance required. Some of the early PS/2 models had a so-called 'Direct Bus Attachment' drive which plugged straight into the MCA bus. There was a ribbon cable with a dirty great ferrite ring around it plugging straight into the riser (which also contained the ordinary MCA slots). Then no separate disk controller was necessary.
The same might not be possible for PCI because of the faster bus speed (33MHz or higher, rather than MCA's 10MHz), unless you have an even bigger ferrite or something.
As well as Blackbox IceWM is small, fast and pleasant to use. I run it everywhere, from Athlons down to a 16MHz 386SX box, and it's acceptably fast on all of them:-).
I was thinking of 2.5 inch disks; if they can be bashed around in a laptop they can surely be mounted on a PCI card in a stationary machine. The disks might have to be thinner, just reduce the number of platters (so the capacity is 40 gigabytes rather than 80 - so what). And the card would probably have to be a bit chunkier than an ordinary PCB. But it's surely doable.
Richard Stallman will be giving a talk on 'Software patents: obstacles to software development' on the 25th of March in Cambridge. I expect he will talk with reference to _all_ software developers, not just free software, because he has said in the past that both free and proprietary software developers have a common interest here.
That's Cambridge, Cambridgeshire rather than Cambridge, MA BTW. Send mail to rmstalk@fipr.org for details.
This is important right now because of the proposed EU patent directive; it would be good for the mainstream press to attend.
I'm not sure you can blame Intel for trying to kill things like SCSI which reduce load on the host CPU. Because such cards require their own separate CPU, which is obviously good news for Intel! Intel chips like the i960, StrongARM, even the good ol' 80186 are all suitable for building intelligent disk controllers.
A better reason for Intel to dislike a technology is that it's controlled by patent holders who happen not to be Intel.
Why bother with ATA at all when you could just make 'hard cards' and plug a disk straight into the PCI bus? It seems kinda pointless to go to all that trouble of defining a complex protocol just for that ten centimetre cable between the motherboard and the disk.
With SCSI I can see some point to it, because you have potentially many disks over longer distances and you want to mix-and-match and have disks support the same features. But since ATA is for the low end anyway, why not make a really basic interface where any fancy stuff is done in software on the host machine? As long as you have DMA for transfers to/from the disk, and some way to move the disk head, that's it really. Today's machines are fast enough to do all the rest for themselves.
I realize this would create a whole nightmarish set of WinDisks... but I'm surprised the 'hard card' concept hasn't caught on again since they were first introduced on XT-type machines.
I think software isn't like engineering because the same code should never be written twice. If you build a bridge, you pick one of a standard set of designs (eg suspension bridge) and follow well-known techniques. There may be some innovation required for this particular bridge, but on the whole you're following established wisdom. If another similar bridge were then required then you could apply the same method.
This would also be true in software if people needed to build the same thing over and over again. If I needed to implement a linked list then I would follow the standard way of doing it, explained in dozens of textbooks, and I would follow an engineering-like process to design, implement and test it.
But that's really not the way it should work. Any competent programmer would use someone else's linked list code, and lots of other libraries besides. (Even if you're some kind of weird C hacker who never reuses code, you still use the C library.) The code you do write is something that has never been done before. Furthermore, market forces mean that it's not worth doing the job perfectly when 'good enough' is most profitable.
I think this may be what you were saying, that life-critical systems are more worth 'engineering' than ordinary applications. I think I have to agree with you, but I'd be very careful about adopting a too-rigid engineering method. These can often turn into impediments to productivity. You have to trust the programmers to come up with new techniques in new situations.
Do you think anyone should be able to call themselves 'Dr Joe Soap' without having a doctorate? How about 'Joe Soap MD'? I don't think the engineers are hijacking the English language. There are plenty of other words to use like 'software developer'.
I also saw a letter to a computer magazine complaining about the unjustified use of the word 'architect'. That is slightly less justified IMO, because you cannot actually qualify as an architect in software, while you can get an engineering degree in that field. So there isn't so much scope for confusion or people claiming fancy titles they don't really hold.
FWIW, the concept of 'software engineer' is a bit inappropriate because writing software is _not_ an engineering discipline, and I don't think it ever will be.
That history is quite amusing to read from bottom to top. It shows how highly the Unisys marketroids rate their own importance.
So we have in 1946 'ENIAC developed', and in 1997, with about the same amount of emphasis: 'Lawrence A. Weinbach named chairman, president, and CEO. Unisys Windows NT servers lead industry in price/performance.'.
I know that's not how it works in the real world, the idea of perfect competition is something that never actually occurs. But it is useful as a benchmark to see how closely real-world markets approximate perfect competition. So in a sense what I'm saying is a tautology - the mainframe market isn't perfectly competitive, duh - but OTOH we can compare it to some other markets, where there is stronger competition and no scope for loss-leaders or market segmentation, to see that IBM is quite a powerful player.
WRT memory modules: point taken, I should have said 'every _working_ memory module at a given speed is interchangeable'. Which is true by definition...
'Turning on a feature introduces a small but definite risk' - so in other words there is a marginal cost to adding a new feature. It's in cases where adding a feature *costs nothing extra* but the firm can still get away with charging extra for it, that you suspect strong market power or a monopoly.
'Because the market will bear the price' - well this shows that there isn't perfect competition. If there were then your customers could get an identical product from a competitor, who'd be happy to turn on the features for free.
The other reasons you give are good ones but again would not apply in a perfectly competitive market. Because in such a market all the products would be indentical or at least interchangeable (eg one 128Mbyte memory module is pretty much identical to another at the same speed, there are no additional bells and whistles that can be offered).
'ACSJ provides enterprise customers with a scalable, reliable platform for ACS deployments' - LOL.
To my knowledge there has never been any site succesfully built using ACS Java - sorry, ACSJ - and I doubt there ever will be. I don't know why RedHat bought the company - what were they thinking of? Did Greylock tell them to?
IBM may have a monoploy in mainframes but that isn't the same as a monopoly in the market for ' serious uptime, serious processing or serious throughput'
Exactly. Just as MS has a monopoly in Windows but not a (total) monopoly in operating systems. MS does arguably have a monopoly in 'Windows-compatible operating systems'. This is one reason why monopolies/mergers/antitrust law is so tricky.
I stand by what I said - that IBM has lots of market power, if not a 100% monopoly, in the mainframe market, and we can tell this by looking at its prices. However because of the large fixed costs in producing mainframes (so that marginal costs, ie manufacturing costs, are only a small part of the total price), it's likely that any other company in IBM's position would do the same.
Hmm, why do you not simply turn on all features by default? If you split the company in ten, all selling the same product, and you end up in control of one of these ten competitors, what pricing strategy would you adopt?
This is not to doubt that IBM is a monopoly. I know that they used to be. It's just that this isn't proof (though it is evidence).
It is not proof of an absolute monopoly. It is evidence of market power. In practice no company has an absolute monopoly because there are always alternatives. Government doesn't intervene only in cases of an absolute monopoly, but also if one company has a near-monopoly or a dominant market position. In the UK a monopoly is legally defined as 25% or more market share, which sounds stupid but makes sense for economic policy.
And, to an extent, the price of mainframe tools is justifible on the basis that they can't expect to sell many of them. Large mainframes are rather like electricity distribution. They are a "natural monopoly" because the entry costs are huge, and there isn't a demand for a large number of them.
Agreed. If you accept that there will only be a few mainframe suppliers, then the hardware-crippling can be seen as a good thing, letting smaller customers afford mainframes.
(I remember an article in an industry publication describing Amdahl's breakthrough in allowing you to limit the amount of CPU or memory your machine uses, in order to pay lower software licence fees. It's ironic that a way to deliberately make your computer perform worse is called a 'breakthrough' - but that's the way this market works. The software companies make their money by pricing according to usage, because marginal costs are zero and there are no direct competitors selling exactly the same piece of software.)
Consider the market of producing "Metallica" albums. That is basically a monopoly (at least if I got the name right). But in this case the government hasn't decided to insist that the market be protected against the monopoly.
The government actively created the monopoly by passing copyright legislation. Doing so is actually in the public interest because it gives incentives to create more music. At least, that's the theory.
Slashdot Mirror
They have ported konsole... so the burning question is will Qt-console run correctly inside that? If so there is no longer any need for screen(1) or even window managers - konsole's screen-switching buttons can do the job of changing between windows.
Or maybe Emacs's M-x term, if anyone ported Emacs to Qt.
There was a similar thing done for Java's AWT toolkit a while ago (article in Dr Dobbs). And I know that aalib comes with an ASCII X server. I wonder which of these three gives the best results? Probably the Qt-console port because it is the most high-level.
A version of KDE using text-mode Qt would be very interesting to behold... or how about a 'desktop' using the aalib ASCII X server (for GTK, Motif etc. programs) plus Qt.
All we have to do now is persuade the Mozilla team to add text-mode display to their bloated 'chrome' rendering system. I'm sure they will be happy to add it as a new feature before 1.0, after all they've thrown in pretty much everything else.
Surely people's web browsing sessions are getting shorter because they've read most of it by now.
Can you explain why the PS/2 mouse causes latency problems? Even when you're not moving it?
I thought it was a reference to the Robin Williams character in 'Santa Claus: The Movie'. Which makes just about as much sense as any reference to Harry Potter or LoTR, given that we're supposed to be discussing Mac OS X.
I agree, it is a bit of an endless cycle. My feeling is just that Serial ATA is (a) not quite good enough to compete with SCSI in the 'let's build an intelligent disk subsystem and take load of the host processor' market, and (b) too complex for the 'let's build something really cheap and dirty' market. The original ATA disk controller could be built for about 50 cents in an ISA bus machine. Could there be an equivalent for PCI?
You could indeed stick it into a drive bay, if you can connect the PCI bus over the distance required. Some of the early PS/2 models had a so-called 'Direct Bus Attachment' drive which plugged straight into the MCA bus. There was a ribbon cable with a dirty great ferrite ring around it plugging straight into the riser (which also contained the ordinary MCA slots). Then no separate disk controller was necessary.
The same might not be possible for PCI because of the faster bus speed (33MHz or higher, rather than MCA's 10MHz), unless you have an even bigger ferrite or something.
As well as Blackbox IceWM is small, fast and pleasant to use. I run it everywhere, from Athlons down to a 16MHz 386SX box, and it's acceptably fast on all of them :-).
I was thinking of 2.5 inch disks; if they can be bashed around in a laptop they can surely be mounted on a PCI card in a stationary machine. The disks might have to be thinner, just reduce the number of platters (so the capacity is 40 gigabytes rather than 80 - so what). And the card would probably have to be a bit chunkier than an ordinary PCB. But it's surely doable.
There's a 'part 2' of the article where Transmeta is mentioned. Link at the bottom.
Richard Stallman will be giving a talk on 'Software patents: obstacles to software development' on the 25th of March in Cambridge. I expect he will talk with reference to _all_ software developers, not just free software, because he has said in the past that both free and proprietary software developers have a common interest here.
That's Cambridge, Cambridgeshire rather than Cambridge, MA BTW. Send mail to rmstalk@fipr.org for details.
This is important right now because of the proposed EU patent directive; it would be good for the mainstream press to attend.
I'm not sure you can blame Intel for trying to kill things like SCSI which reduce load on the host CPU. Because such cards require their own separate CPU, which is obviously good news for Intel! Intel chips like the i960, StrongARM, even the good ol' 80186 are all suitable for building intelligent disk controllers.
A better reason for Intel to dislike a technology is that it's controlled by patent holders who happen not to be Intel.
Why bother with ATA at all when you could just make 'hard cards' and plug a disk straight into the PCI bus? It seems kinda pointless to go to all that trouble of defining a complex protocol just for that ten centimetre cable between the motherboard and the disk.
With SCSI I can see some point to it, because you have potentially many disks over longer distances and you want to mix-and-match and have disks support the same features. But since ATA is for the low end anyway, why not make a really basic interface where any fancy stuff is done in software on the host machine? As long as you have DMA for transfers to/from the disk, and some way to move the disk head, that's it really. Today's machines are fast enough to do all the rest for themselves.
I realize this would create a whole nightmarish set of WinDisks... but I'm surprised the 'hard card' concept hasn't caught on again since they were first introduced on XT-type machines.
I don't get it, why not just commoditize SCSI and make it cheaper? Nice features like power and data in the same cable can be done that way too.
Alternatively, if Serial ATA is really good, does that mean SCSI will become obsolete?
Are there good technical reasons for having both? Or just marketing reasons to do with charging more for SCSI drives, and keeping ATA for the low-end?
I think software isn't like engineering because the same code should never be written twice. If you build a bridge, you pick one of a standard set of designs (eg suspension bridge) and follow well-known techniques. There may be some innovation required for this particular bridge, but on the whole you're following established wisdom. If another similar bridge were then required then you could apply the same method.
This would also be true in software if people needed to build the same thing over and over again. If I needed to implement a linked list then I would follow the standard way of doing it, explained in dozens of textbooks, and I would follow an engineering-like process to design, implement and test it.
But that's really not the way it should work. Any competent programmer would use someone else's linked list code, and lots of other libraries besides. (Even if you're some kind of weird C hacker who never reuses code, you still use the C library.) The code you do write is something that has never been done before. Furthermore, market forces mean that it's not worth doing the job perfectly when 'good enough' is most profitable.
I think this may be what you were saying, that life-critical systems are more worth 'engineering' than ordinary applications. I think I have to agree with you, but I'd be very careful about adopting a too-rigid engineering method. These can often turn into impediments to productivity. You have to trust the programmers to come up with new techniques in new situations.
Do you think anyone should be able to call themselves 'Dr Joe Soap' without having a doctorate? How about 'Joe Soap MD'? I don't think the engineers are hijacking the English language. There are plenty of other words to use like 'software developer'.
I also saw a letter to a computer magazine complaining about the unjustified use of the word 'architect'. That is slightly less justified IMO, because you cannot actually qualify as an architect in software, while you can get an engineering degree in that field. So there isn't so much scope for confusion or people claiming fancy titles they don't really hold.
FWIW, the concept of 'software engineer' is a bit inappropriate because writing software is _not_ an engineering discipline, and I don't think it ever will be.
I believe many of O'Reilly's books are produced with troff. Their books about TeX use a different method, obviously.
Well, the JVM segfaults occasionally... but that was written in C I suppose.
That history is quite amusing to read from bottom to top. It shows how highly the Unisys marketroids rate their own importance.
So we have in 1946 'ENIAC developed', and in 1997, with about the same amount of emphasis: 'Lawrence A. Weinbach named chairman, president, and CEO. Unisys Windows NT servers lead industry in price/performance.'.
I know that's not how it works in the real world, the idea of perfect competition is something that never actually occurs. But it is useful as a benchmark to see how closely real-world markets approximate perfect competition. So in a sense what I'm saying is a tautology - the mainframe market isn't perfectly competitive, duh - but OTOH we can compare it to some other markets, where there is stronger competition and no scope for loss-leaders or market segmentation, to see that IBM is quite a powerful player.
WRT memory modules: point taken, I should have said 'every _working_ memory module at a given speed is interchangeable'. Which is true by definition...
'Turning on a feature introduces a small but definite risk' - so in other words there is a marginal cost to adding a new feature. It's in cases where adding a feature *costs nothing extra* but the firm can still get away with charging extra for it, that you suspect strong market power or a monopoly.
'Because the market will bear the price' - well this shows that there isn't perfect competition. If there were then your customers could get an identical product from a competitor, who'd be happy to turn on the features for free.
The other reasons you give are good ones but again would not apply in a perfectly competitive market. Because in such a market all the products would be indentical or at least interchangeable (eg one 128Mbyte memory module is pretty much identical to another at the same speed, there are no additional bells and whistles that can be offered).
The press release is just corporate bullshit.
'ACSJ provides enterprise customers with a scalable, reliable platform for ACS deployments' - LOL.
To my knowledge there has never been any site succesfully built using ACS Java - sorry, ACSJ - and I doubt there ever will be. I don't know why RedHat bought the company - what were they thinking of? Did Greylock tell them to?
Exactly. Just as MS has a monopoly in Windows but not a (total) monopoly in operating systems. MS does arguably have a monopoly in 'Windows-compatible operating systems'. This is one reason why monopolies/mergers/antitrust law is so tricky.
I stand by what I said - that IBM has lots of market power, if not a 100% monopoly, in the mainframe market, and we can tell this by looking at its prices. However because of the large fixed costs in producing mainframes (so that marginal costs, ie manufacturing costs, are only a small part of the total price), it's likely that any other company in IBM's position would do the same.
Hmm, why do you not simply turn on all features by default? If you split the company in ten, all selling the same product, and you end up in control of one of these ten competitors, what pricing strategy would you adopt?
It is not proof of an absolute monopoly. It is evidence of market power. In practice no company has an absolute monopoly because there are always alternatives. Government doesn't intervene only in cases of an absolute monopoly, but also if one company has a near-monopoly or a dominant market position. In the UK a monopoly is legally defined as 25% or more market share, which sounds stupid but makes sense for economic policy.
Agreed. If you accept that there will only be a few mainframe suppliers, then the hardware-crippling can be seen as a good thing, letting smaller customers afford mainframes.
(I remember an article in an industry publication describing Amdahl's breakthrough in allowing you to limit the amount of CPU or memory your machine uses, in order to pay lower software licence fees. It's ironic that a way to deliberately make your computer perform worse is called a 'breakthrough' - but that's the way this market works. The software companies make their money by pricing according to usage, because marginal costs are zero and there are no direct competitors selling exactly the same piece of software.)
The government actively created the monopoly by passing copyright legislation. Doing so is actually in the public interest because it gives incentives to create more music. At least, that's the theory.