CPU specific optimizations would be a benefit, but having looked at TCC I doubt it currently does anything of the sort:-) And of course I doubt the tiny performance improvement you could get out of it would make a difference if you're running off a CD anyway...:)
Comparing it to Perl is a bit unfair considering we don't exactly have a full fledged Unix like kernel written in Perl to boot...
I prefer to see this as a great proof of concept that kernel compilation can be made fast enough to do "on the fly". Considering that driver installation for Linux still often requires a kernel recompile, if this system can be made solid enough it could make things like that a lot easier for end users, though I think I'd prefer to have it done at package installation time rather than boot time:-)
However, that doesn't make the hardware MPEG-2 playback a good thing. It doesn't do hardware MPEG-1 in any of the chipsets I've heard about. And as I said in my previous post, you can do any video filtering before the video is displayed, so it's still quite a lowsy compromise. You're much better off getting a system that doesn't choke on software MPEG-2 playback. I could easily find an old Intel processor that's more than fast enough for software MPEG-2, and as low power as VIA's EPIA processors.
I'm sure you could find an old Intel processor that fast enough, but thats irellevant. What IS relevant is whether you can find a Mini-ITX or smaller form factor motherboard that can take that CPU and that has TV out, SP-DIF, IR support, USB, firewire, compact flash slots, and all the other stuff people buy the Via system for, and that can be easily passively cooled with a setup small enough to fit into one of the small Mini-ITX cases designed for fanless operation.
In my experience, the Via MPEG-2 playback is more than sufficient to generate high quality output - in fact I prefer it over a most of the software decoders I've seen. And for a home cinema setup, MPEG-1 is mostly useless, as DVB (digital TV standard used practically everywhere but the US) uses MPEG-2 native (I suspect the US digital standard does too, but I don't know for sure), as does DVDs. I don't have a single movie encoded using MPEG-1
No, they aren't. If he was asking about the quality of the video card, that would be something. Instead, he wants the benchmarks to be highly biased torwards VIA, and against every other processor. With a computer, you want general-purpose benchmarks, not ones that exaggerate the importance of it's few strengths.
Including MPEG2 playback and other uses typical for the use of Mini-ITX boards in a review of performance of Mini-ITX boards is hardly making them highly biased towards one of the manufacturers. Including Divx and MPEG1 decoding would be good too, and Via isn't likely to do well there, which is worth knowing for a large segment of Mini-ITX purchasers.
With a computer, you want benchmarks that reflect the use people are likely to put the system to, and pure CPU performance is a useless indicator of performance for most "consumer" type uses except games - which isn't even remotely on the radar for Mini-ITX systems - and even CPU performance is generally of secondary importance after 3D graphics performance.
For Mini-ITX, set top boxes/home cinema setups is a HUGE chunk of the market, and any benchmark that ignores this is useless for a significant part of the potential customer base. (Yes, I know they say the Transmeta based board isn't intended for that market, but if you're comparing it to Via it matters to include what Via's market cares about).
I would agree with you if the suggestion was to give an overall performance rating based on some aggregate of CPU performance and specific application performance such as MPEG2 decoding, as that would be useless to everyone. But you can perfectly well take application performance into consideration by including some typical application results separately.
On VIA EPIA boards, the hardware MPEG2 playback is vital if you want smooth playback at reasonable resolutions at least up to the M12000 (latest I've tested), and provide good quality output. If this machine is fast enough not to have problems with doing software decoding then great, but it the questions about what the benchmarks measured that the previous poster asked are still valid - CPU performance is far from everything on a system where a significant number of the customers are likely to put them in cases with no posibility of adding a separate graphics card or other peripherals.
Firefox as a platform works in two ways: You use XUL with the security restrictions of remotely loaded javascript, or you install an XPI that may contain javascript with more rights. The first case is all you need to be able to do what you suggest, and gives no more access to the system than a normal HTML page with javascript. The latter is what brings security implications.
Using Sender: to indicate the actual account and leave From: as the account he wants people to consider it as "from" would be better, as that leaves Reply-To: free to be used to redirect replies somewhere else on a case by case basis.
If the US spent the same amount it does on wars on promoting peace, instead of starting wars, there would at least be a few wars less and a lot fewer people hating the US enough to want to blow people up.
There's a huge leap between not spending money on wars where you are the agressor, and perhaps cutting down on military spending so that the US doesn't spend more than the next 25 countries combined (including China, Russia, India and other countries making up more than half the worlds population), and not spending anything on the military.
Hey, maybe the US would manage to balance its budget again too...
That isn't an excuse. It is a reality that doesn't attract many people, because it is contrary to most musmlims interpretations of Islam.
If using the Quaran was a sufficient excuse to get the attention and support the terrorist organisations wanted, they would have no need to use the US as an excuse in all their propaganda. Hatred only "works" if it has a target that people dislike.
It's not about blaming the US, but about stating the reality of why the US is facing decades of war unless it addresses the reasons why these organisations can get money and people so easily: Millions of people hate the US for it's agressive foreign policy. Some of them enough to risk dying for it.
Perhaps, and to some people that IS an issue. A lot of amputee's do choose not to use a prosthetic, or at least not to always use it because it draws attention to their disability.
If RTFA'd you'd see that the iBOT was explicitly mentioned, but that the main point of the article is that a lot of people don't want to use a wheelchair, find it more comfortable standing up instead of staring at peoples asses, and/or didn't want to draw attention to their disabilities.
Yes, the iBOT is great, but it's targetted at a different market segment.
Fuck you too. I walk "a lot". Generally at least half an hour each day related to my commute, and quite often I go for long walks. I do all my food shopping on foot. I don't even own a car, because I prefer to walk when I can and use public transport whenever I'm going to far to walk. I've more than once chosen to walk 50-60 minutes rather than take the bus because I enjoy it.
However, my knees sometimes act up, particularly in cold weather, and it gets significantly worse if I then keep walking, to the point where I will have problems sitting down because it's painful to bend my leg.
Today I solve it by taking the bus when my knees feel bad, and avoiding long walks until they're ok again, but I'd much rather use a Segway and avoid the waiting and the bloody traffic, and be more flexible about buying groceries in larger quantities than practical on food.
I'm sure some people want it out of pure lazyness, but there's a lot of people that want it not to replace walking, but to replace a car or to enable them to keep moving around without being tied to a wheelchair or restricted by bus routes.
It's about practicality. I don't like cycling in traffic, because I don't want to go through maneuvering crowds with a device where my balance gets progressively worse the slower I go. Also, I'd prefer not arriving at work all sweaty. Additionally, in my case going to work involves two walks and a train ride - I'd certainly not want the hassle of taking a bike on the train, though they do allow it on most routes.
Most times I walk to and from the train station on my route to work, but if I'm in a hurry it would be great to have an alternatie that doesn't involve a bike, and doesn't involve car (don't have one, never bothered getting a license)
The shops and the train station where I live are a 15 minute walk away. I can take the bus, but worst case I have to wait 10-12 minutes. Being able to just grab a Segway whenever I don't feel like the walk would be great, particularly as my knees act up in cold weather.
You miss the point. It's not about feeling guilty. It's about whether or not there are people out there who will see the US as acceptable targets and as a result decides to volunteer for terrorist operations.
US foreign policy is sufficiently divisive to allow terrorist organisations to capitalize on the hatred it evokes and used US actions as a way of getting people to volunteer.
No matter how abhorrent it is, the US needs to face facts: The US is a target of terror because it maintains a foreign policy that incite hatred in large groups of people, to the point where not only Islamic fundamentalists but even non-muslims all over the world get disgusted.
Fix the US foreign policy and the vast majority of recruits to terrorist organisations that wants to target the US will disappear, because the excuses they use to justify their decisions have disappeared and exposed them for what they are.
That IS the real issue of militant Islamic terrorism. Take away their excuses, and you've cut off their air supply. As a side effect perhaps the US would actually be respected again.
Your logic is invalid. That some groups of people haven't resorted to violence as a result of being subjected to violence and oppression doesn't mean that other groups won't.
In any case, the most important issue is that US foreign policy is creating excuses. If someone wants to carry out a terrorist attack and needs to attract volunteers to carry out the attacks, it's a damn lot easier when the country is widely despised for it's support someone most of your target audience considers enemies.
THAT is what the US should be really worried about - each time a civilian is killed by US troops, no matter how much it's an accident or no matter whether it was an unfortunate result of a fully justified operation, someone out there will hate the US more. Some of them will be easy prey for fanatical terror groups. Some of them will go on to recruit and conduct propaganda for such groups themselves. The same happens each time the US veto's a UN resolution condemning Israel, or in other ways does anything that is seen by huge groups of people as interfering with their lives and supporting oppression.
These groups would have a vastly harder time recruiting anyone if the US focused more on being seen as fair and balanced and respectful of cultural and religious differences, and less on military superiority.
As I pointed out, unless the license explicitly grants you a license to the patent, you don't get a license to use the patent just because you have a license that grants you the right to distribute the software.
The BSD license, for instance, would NOT in any shape or form give you the right to use software that contains patented material, though you might still be able to redistribute the software (depending on laws in your jurisdiction). The GPL is amiguous on the point. The preamble and section 7 deals briefly with patents, but IANAL and don't want to guess at what problems would arise.
The point is that there is nothing preventing the release of software as open source while still protecting the value of the patent by requiring a paid license to use the patented invention. Such software could still be useful - it might allow open source products to include the patented software as an option that you can legally use only with a license (FreeType have already been doing something similar for hinting).
The right to limit use is the primary protection granted by patents. If you wish to limit distribution, you should be looking at copyright, not patents.
You seem to be misunderstanding it too. The point of patents is that you get the time limited monopoly in return for explaining how your invention works, so that when the patent expires everyone can start using the same invention.
Releasing your patented software under an open source license would only make your patent meaningless if the license explicitly includes a license of the patent as well as the code - otherwise the patent would prevent you from legally using the software.
One of the many problems with patents today is that the patent claims usually are so vague they can apply to a very broad class of implementations, and can be stretched to cover ideas that wasn't conceived when the patent was filed.
Requiring an "open source" implementation would at least solve that one problem, if the patent protection is limited to exactly the methods demonstrated by the programs. It would also make it hard to obscure how the invention works, and therefore make patents much more useful, and make it easier to try to work around them.
Your logic is completely screwed up. SCO isn't using patents, they're trying to use a contract regarding Unix licenses, and at some point claimed they'd be raising copyright claims.
But even if they had any patents to sue over, there is a clear difference between using patents to make a profit from other peoples work and using patents to protect yourself from people trying to use patents to make a profit from your work.
In the UK, contracted project managers can expect to get up to around 250 UKP a day / 60k UKP a year unless they have particularly marketable skills apart from the generic project management skills, or are particularly senior. Not bad, but nowhere near the levels a good analyst can get as a contractor (or if you've got the qualifications and the masochist tendencies required to land a senior analyst position in a bank).
Expect +/- 40k (UKP) a year for the same level permanent jobs.
<blatant-plug>
If you want flexible working hours (another way to say we sometimes work late, but then we make up for it by starting late) and are interested in working on the whole lifecycle and not just design, take a look here, or mail me a CV. I'm looking for two engineers based in London - someone with analysis background would definitively be appreciated as long as they've got the skills and the will to code as well.
VIA has released a version of Xine with support for the MPEG2 hardware built in. For me that one never worked very well. If you want to try it, search for VeXP.
I've used both Xine and Mplayer without problems, and yes, playback is smooth with the hardware decoder. With the M12000 it's not smooth for any reasonable resolutions if you don't use the hardware decoder...
The main problem was getting the right X drivers installed... Take a look here for one approach. And in particular, note that if you run into weird problems with the colour balance, you've installed an old buggy driver - I almost went crazy before I stumbled across the explanation somewhere.
If you haven't bought a VIA EPIA board yet, research it thoroughly before doing so - the different boards can have quite different chipsets on them, so the level of support for Linux (particularly video and audio) may wary quite a lot.
I bought a M12000 based machine in an e-Otonashi case to use for streaming to my TV, and it works great. The VIA CPU's are slow, but their Mini-ITX motherboards make up for a lot of it by having hardware MPEG2 decoding on most of them. I'd be careful (that is, test before you buy) about the M12000 for systems where you need support for other video formats, or high (as in higher than TV) resolutions, but I guess the newer VIA CPU's might be getting there.
The main appeal to me was that with the e-Otonashi case, I can run the system completely fanless. I was originally planning to tear out the HD and replace it with a 512MB compact flash, but after hearing (or rather not hearing) the system in action I decided to use the compact flash for my digital camera instead - I can barely notice the HD when the disk is heavily trashing and I'm reasonably close to the machine, but even then it's quieter than my old Sony DVD player.
Word of warning though: Setting up Linux on the VIA Mini-ITX boards can be a real pain. Getting the basic stuff up and running is trivial (For people in the UK I can recommend LinItx.com - they delivered my system quickly and with Fedora Core 1 pre-installed, but getting the best out of the TV out and hardware MPEG2 decoder was painful (thanks to VIA not being particularly forthcoming with specs) when I did it a few months ago.
So what , its still just a shell! So you can download some parser to parse it. Oh well great, that saves a weeks development time. And slows down the whole product whenever its run. Hmm , great tradeoff. Not.
Even making the assumtion that what you say above is true (saving a week of development time and slows down the whole product whenever it's run), it still makes a lot of sense for many cases:
As an example, a typical senior developer in the UK will cost their company somewhere between a USD 8k and USD 12k a month (salary + benefits + national insurance contributions + office costs and others). Assuming 8k a month, saving about a week and allowing that developer to go on to other projects means that unless the servers needs an upgrade worth more than 2k because of being slowed down by the XML parsing, it's a net benefit.
However for typical business applications that are processing transaction volumes large enough that performance is much of an issue, the savings in development time over using proprietary formats and having someone write parsers for it are likely to be more on the order of months.
And given how bad most software developers are at writing parsers, if you go with proprietary formats, however, they're more likely than not to write something slow, brittle and difficult to extend...
All the server would need to do would be to close the tags. Seeing as a server using this approach would likely be maintaining a list of open tags, closing them and leaving the recipient with well formed XML but possibly semantic errors is trivial.
This is no different than a protocol using size indicators, such as HTTP with Content-Length: or using chunked transfer, or SMTP or any protocol using a special end marker - if you pass on data before it is completely received, you ALWAYS need to be prepared to either clean things up or break off the connection to the recipient on the other side if the sender doesn't send the complete message.
Sometimes that is a problem, and you will need to keep the message until it is complete before you pass it on, but in most cases where the recipient is a human it's an acceptable trade off and you'll just choose whatever mechanism is suitable to terminate the message as cleanly as possible. The choice of whether to break the connection or terminate depends on the importance of the integrity of the data, and the cost of reestablishing the connection, but is usually a fairly trivial assessment to make.
And if you, as you suggest, start complicating the framing etc. with routing information, you're just increasing the cost of handling the framing - SAX parsing is cheap, and it will take very little extra overhead added from additional framing before any saving from reduced XML parsing will be lost.
Sigh. Why are you assuming I'd only have one parser instance? What part of "pseudo" code was it you blatantly decided to ignore? So imagine an array subscript in the call to the sax parser. And as for associating the events to a session, that's why any parser worth anything would let you associate user data to the parser instance, such as for instance the setProperty() method of the XMLReader interface in SAX 2.0 parsers. Are you going to complain that my pseudo code wouldn't run because I called undefined functions too?
As for your claim it's a "problem" that you have to parse everything, if your server is CPU bound and the SAX parser is dominating your profiling output you're doing something seriously wrong.
You're assuming that the cheap tokenization work a SAX parser has to do is massively more expensive than handling other types of framing, and enough for it to become a problem. In my work it's never even showed up in profiling runs as a factor worth optimizing. Tuning the system to maximize IO throughput and minimize the frequency of system calls (to reduce context switches) is generally much more worthwhile.
Slashdot Mirror
CPU specific optimizations would be a benefit, but having looked at TCC I doubt it currently does anything of the sort :-) And of course I doubt the tiny performance improvement you could get out of it would make a difference if you're running off a CD anyway... :)
I prefer to see this as a great proof of concept that kernel compilation can be made fast enough to do "on the fly". Considering that driver installation for Linux still often requires a kernel recompile, if this system can be made solid enough it could make things like that a lot easier for end users, though I think I'd prefer to have it done at package installation time rather than boot time :-)
I'm sure you could find an old Intel processor that fast enough, but thats irellevant. What IS relevant is whether you can find a Mini-ITX or smaller form factor motherboard that can take that CPU and that has TV out, SP-DIF, IR support, USB, firewire, compact flash slots, and all the other stuff people buy the Via system for, and that can be easily passively cooled with a setup small enough to fit into one of the small Mini-ITX cases designed for fanless operation.
In my experience, the Via MPEG-2 playback is more than sufficient to generate high quality output - in fact I prefer it over a most of the software decoders I've seen. And for a home cinema setup, MPEG-1 is mostly useless, as DVB (digital TV standard used practically everywhere but the US) uses MPEG-2 native (I suspect the US digital standard does too, but I don't know for sure), as does DVDs. I don't have a single movie encoded using MPEG-1
No, they aren't. If he was asking about the quality of the video card, that would be something. Instead, he wants the benchmarks to be highly biased torwards VIA, and against every other processor. With a computer, you want general-purpose benchmarks, not ones that exaggerate the importance of it's few strengths.
Including MPEG2 playback and other uses typical for the use of Mini-ITX boards in a review of performance of Mini-ITX boards is hardly making them highly biased towards one of the manufacturers. Including Divx and MPEG1 decoding would be good too, and Via isn't likely to do well there, which is worth knowing for a large segment of Mini-ITX purchasers.
With a computer, you want benchmarks that reflect the use people are likely to put the system to, and pure CPU performance is a useless indicator of performance for most "consumer" type uses except games - which isn't even remotely on the radar for Mini-ITX systems - and even CPU performance is generally of secondary importance after 3D graphics performance.
For Mini-ITX, set top boxes/home cinema setups is a HUGE chunk of the market, and any benchmark that ignores this is useless for a significant part of the potential customer base. (Yes, I know they say the Transmeta based board isn't intended for that market, but if you're comparing it to Via it matters to include what Via's market cares about).
I would agree with you if the suggestion was to give an overall performance rating based on some aggregate of CPU performance and specific application performance such as MPEG2 decoding, as that would be useless to everyone. But you can perfectly well take application performance into consideration by including some typical application results separately.
On VIA EPIA boards, the hardware MPEG2 playback is vital if you want smooth playback at reasonable resolutions at least up to the M12000 (latest I've tested), and provide good quality output. If this machine is fast enough not to have problems with doing software decoding then great, but it the questions about what the benchmarks measured that the previous poster asked are still valid - CPU performance is far from everything on a system where a significant number of the customers are likely to put them in cases with no posibility of adding a separate graphics card or other peripherals.
Firefox as a platform works in two ways: You use XUL with the security restrictions of remotely loaded javascript, or you install an XPI that may contain javascript with more rights. The first case is all you need to be able to do what you suggest, and gives no more access to the system than a normal HTML page with javascript. The latter is what brings security implications.
Using Sender: to indicate the actual account and leave From: as the account he wants people to consider it as "from" would be better, as that leaves Reply-To: free to be used to redirect replies somewhere else on a case by case basis.
There's a huge leap between not spending money on wars where you are the agressor, and perhaps cutting down on military spending so that the US doesn't spend more than the next 25 countries combined (including China, Russia, India and other countries making up more than half the worlds population), and not spending anything on the military.
Hey, maybe the US would manage to balance its budget again too...
If using the Quaran was a sufficient excuse to get the attention and support the terrorist organisations wanted, they would have no need to use the US as an excuse in all their propaganda. Hatred only "works" if it has a target that people dislike.
It's not about blaming the US, but about stating the reality of why the US is facing decades of war unless it addresses the reasons why these organisations can get money and people so easily: Millions of people hate the US for it's agressive foreign policy. Some of them enough to risk dying for it.
Perhaps, and to some people that IS an issue. A lot of amputee's do choose not to use a prosthetic, or at least not to always use it because it draws attention to their disability.
No, he means the iBOT, which is manufactured and available from Johnson and Johnson.
Yes, the iBOT is great, but it's targetted at a different market segment.
However, my knees sometimes act up, particularly in cold weather, and it gets significantly worse if I then keep walking, to the point where I will have problems sitting down because it's painful to bend my leg.
Today I solve it by taking the bus when my knees feel bad, and avoiding long walks until they're ok again, but I'd much rather use a Segway and avoid the waiting and the bloody traffic, and be more flexible about buying groceries in larger quantities than practical on food.
I'm sure some people want it out of pure lazyness, but there's a lot of people that want it not to replace walking, but to replace a car or to enable them to keep moving around without being tied to a wheelchair or restricted by bus routes.
Most times I walk to and from the train station on my route to work, but if I'm in a hurry it would be great to have an alternatie that doesn't involve a bike, and doesn't involve car (don't have one, never bothered getting a license)
The shops and the train station where I live are a 15 minute walk away. I can take the bus, but worst case I have to wait 10-12 minutes. Being able to just grab a Segway whenever I don't feel like the walk would be great, particularly as my knees act up in cold weather.
US foreign policy is sufficiently divisive to allow terrorist organisations to capitalize on the hatred it evokes and used US actions as a way of getting people to volunteer.
No matter how abhorrent it is, the US needs to face facts: The US is a target of terror because it maintains a foreign policy that incite hatred in large groups of people, to the point where not only Islamic fundamentalists but even non-muslims all over the world get disgusted.
Fix the US foreign policy and the vast majority of recruits to terrorist organisations that wants to target the US will disappear, because the excuses they use to justify their decisions have disappeared and exposed them for what they are.
That IS the real issue of militant Islamic terrorism. Take away their excuses, and you've cut off their air supply. As a side effect perhaps the US would actually be respected again.
In any case, the most important issue is that US foreign policy is creating excuses. If someone wants to carry out a terrorist attack and needs to attract volunteers to carry out the attacks, it's a damn lot easier when the country is widely despised for it's support someone most of your target audience considers enemies.
THAT is what the US should be really worried about - each time a civilian is killed by US troops, no matter how much it's an accident or no matter whether it was an unfortunate result of a fully justified operation, someone out there will hate the US more. Some of them will be easy prey for fanatical terror groups. Some of them will go on to recruit and conduct propaganda for such groups themselves. The same happens each time the US veto's a UN resolution condemning Israel, or in other ways does anything that is seen by huge groups of people as interfering with their lives and supporting oppression.
These groups would have a vastly harder time recruiting anyone if the US focused more on being seen as fair and balanced and respectful of cultural and religious differences, and less on military superiority.
Uhm, they have no choice but to comply when it's done under court order, unless the individual employees want to end up in jail.
The BSD license, for instance, would NOT in any shape or form give you the right to use software that contains patented material, though you might still be able to redistribute the software (depending on laws in your jurisdiction). The GPL is amiguous on the point. The preamble and section 7 deals briefly with patents, but IANAL and don't want to guess at what problems would arise.
The point is that there is nothing preventing the release of software as open source while still protecting the value of the patent by requiring a paid license to use the patented invention. Such software could still be useful - it might allow open source products to include the patented software as an option that you can legally use only with a license (FreeType have already been doing something similar for hinting).
The right to limit use is the primary protection granted by patents. If you wish to limit distribution, you should be looking at copyright, not patents.
Releasing your patented software under an open source license would only make your patent meaningless if the license explicitly includes a license of the patent as well as the code - otherwise the patent would prevent you from legally using the software.
One of the many problems with patents today is that the patent claims usually are so vague they can apply to a very broad class of implementations, and can be stretched to cover ideas that wasn't conceived when the patent was filed.
Requiring an "open source" implementation would at least solve that one problem, if the patent protection is limited to exactly the methods demonstrated by the programs. It would also make it hard to obscure how the invention works, and therefore make patents much more useful, and make it easier to try to work around them.
But even if they had any patents to sue over, there is a clear difference between using patents to make a profit from other peoples work and using patents to protect yourself from people trying to use patents to make a profit from your work.
Expect +/- 40k (UKP) a year for the same level permanent jobs.
<blatant-plug>
If you want flexible working hours (another way to say we sometimes work late, but then we make up for it by starting late) and are interested in working on the whole lifecycle and not just design, take a look here, or mail me a CV. I'm looking for two engineers based in London - someone with analysis background would definitively be appreciated as long as they've got the skills and the will to code as well.
Or for other Yahoo jobs in Europe (most in London).
</blatant-plug>
I've used both Xine and Mplayer without problems, and yes, playback is smooth with the hardware decoder. With the M12000 it's not smooth for any reasonable resolutions if you don't use the hardware decoder...
The main problem was getting the right X drivers installed... Take a look here for one approach. And in particular, note that if you run into weird problems with the colour balance, you've installed an old buggy driver - I almost went crazy before I stumbled across the explanation somewhere.
If you haven't bought a VIA EPIA board yet, research it thoroughly before doing so - the different boards can have quite different chipsets on them, so the level of support for Linux (particularly video and audio) may wary quite a lot.
The main appeal to me was that with the e-Otonashi case, I can run the system completely fanless. I was originally planning to tear out the HD and replace it with a 512MB compact flash, but after hearing (or rather not hearing) the system in action I decided to use the compact flash for my digital camera instead - I can barely notice the HD when the disk is heavily trashing and I'm reasonably close to the machine, but even then it's quieter than my old Sony DVD player.
Word of warning though: Setting up Linux on the VIA Mini-ITX boards can be a real pain. Getting the basic stuff up and running is trivial (For people in the UK I can recommend LinItx.com - they delivered my system quickly and with Fedora Core 1 pre-installed, but getting the best out of the TV out and hardware MPEG2 decoder was painful (thanks to VIA not being particularly forthcoming with specs) when I did it a few months ago.
Even making the assumtion that what you say above is true (saving a week of development time and slows down the whole product whenever it's run), it still makes a lot of sense for many cases:
As an example, a typical senior developer in the UK will cost their company somewhere between a USD 8k and USD 12k a month (salary + benefits + national insurance contributions + office costs and others). Assuming 8k a month, saving about a week and allowing that developer to go on to other projects means that unless the servers needs an upgrade worth more than 2k because of being slowed down by the XML parsing, it's a net benefit.
However for typical business applications that are processing transaction volumes large enough that performance is much of an issue, the savings in development time over using proprietary formats and having someone write parsers for it are likely to be more on the order of months.
And given how bad most software developers are at writing parsers, if you go with proprietary formats, however, they're more likely than not to write something slow, brittle and difficult to extend...
This is no different than a protocol using size indicators, such as HTTP with Content-Length: or using chunked transfer, or SMTP or any protocol using a special end marker - if you pass on data before it is completely received, you ALWAYS need to be prepared to either clean things up or break off the connection to the recipient on the other side if the sender doesn't send the complete message.
Sometimes that is a problem, and you will need to keep the message until it is complete before you pass it on, but in most cases where the recipient is a human it's an acceptable trade off and you'll just choose whatever mechanism is suitable to terminate the message as cleanly as possible. The choice of whether to break the connection or terminate depends on the importance of the integrity of the data, and the cost of reestablishing the connection, but is usually a fairly trivial assessment to make.
And if you, as you suggest, start complicating the framing etc. with routing information, you're just increasing the cost of handling the framing - SAX parsing is cheap, and it will take very little extra overhead added from additional framing before any saving from reduced XML parsing will be lost.
As for your claim it's a "problem" that you have to parse everything, if your server is CPU bound and the SAX parser is dominating your profiling output you're doing something seriously wrong.
You're assuming that the cheap tokenization work a SAX parser has to do is massively more expensive than handling other types of framing, and enough for it to become a problem. In my work it's never even showed up in profiling runs as a factor worth optimizing. Tuning the system to maximize IO throughput and minimize the frequency of system calls (to reduce context switches) is generally much more worthwhile.