I'm not really up-to-date with the current state of BSD ports that are running on the Alpha architecture. But I'm surely willing to give it a go: in the end, all I want are stable and secure machines in production.
If you have any pointers, I would very much appreciate it if you could contact me on bram_at_e-wareness_dot_be.
Thanks.
Okay... I'll do the stupid things first, then you shy people follow.
Slackware doesn't use a silly configuration tool with umpteen configuration files to keep the settings... It has a limited number of start-up scripts which are pretty straight-forward, so that you can customize them easily at your situation.
Since they are that simple and straightforward, I've learned quite a lot about how everything is arranged at boot-up, which is a lot more difficult with Red Hat, SuSe or Debian, where there seem to be hundreds of directories with files and symlinks... Especially newbies who want to go a bit further than playing with their GUI config tool find this very overwhelming.
Basically, slackware is a do-it-yourself distro, since their package management system isn't really that good. But if you know your way around, I think it has a lot to offer in system stability and insight in the inner workings.
It's a pity it's only x86... I would really love to have a comparable distro for my Alphas:-)
Okay... I'll do the stupid things first, then you shy people follow.
Commenting on the question from tbray, a number of these patents seem really silly to me (I must say that I didn't take the time to read 'em all), but the titles alone make me shiver:
23. Nielsen, J.: Tooltips on webpages, U.S. Patent 5,937,417 (1999)
11. Nielsen, J.: Password helper using a client-side master password which automatically presents the appropriate server-side password to a particular remote server, U.S. Patent 6,006,333 (1999) [don't Windows.PWL files do something similar?]
50. Nielsen, J.: System and method for temporally varying pointer icons, U.S. Patent 5,784,056 (1998)
A whole number of them are from 1999, which means that you're frantically inventing great new things and spend tens of thousands of $$$ just patenting the things... There are about 30 patents listed in 1999, which means 2 patents every three weeks, for an amount of at least 500$K patenting costs and patent attorney fees to get them patented...
This all seems very Amazon-like, and either I've missed something really important, or is something else going on?
Okay... I'll do the stupid things first, then you shy people follow.
... unlike a lot of you are saying... NetBEUI isn't dead, but is a very useful feature for all those little SOHO networks around there...
I know litterally hundreds of little businesses with 2/3 people, who only really want to share files and printers... There are thousands of users out there that have the computer in their own little business just because they need it, but wouldn't know a mouse from a keyboard if they didn't read the manuals. They don't want to spend any money more than really necessary on computer stuff, let alone that they would hire someone to set it up and troubleshoot it for them...
Even with dhcp, setting up tcp/ip is a real nightmare for them under Win95/98. (The TCP/IP protocol isn't installed automatically with windows 95, remember). The sheer number of options is very confusing, let alone that one could do it with static IPs without knowing what the heck these numbers mean. They want to give these computers names, not numbers.
For those computer illiterates, an appliance like the Cobalt Qube, with NetBEUI support would be a real plug-n-play solution, because there really is no setting up NetBEUI. Just to store their files, and get to use one printer from different computers. That's all they really need, and all they should be worried about.
Okay... I'll do the stupid things first, then you shy people follow.
IOI SCSI has not just the source tarballs of the Linux drivers for their SCSI cards, they also provided the necessary bootdisks for a number of distros on the driver CDs which is - as I need not explain - uber cool.
Their cards are based on the ignitio chipsets, which makes it quite performant and stable under Linux... And they've been doing so long before the linux hyped.
Okay... I'll do the stupid things first, then you shy people follow.
The downside is that the solution will involve extra processing steps, extra stuff to be implemented, and impose on you a development model that might not always be convenient (not everything wants to be a document, or a conversion or transcoding between document formats).
At a certain point, the trouble to do everything in HTML for 15 different browser versions, including Psion, Palm & WAP-enabled phones, the sheer amount of work of doing that and keeping things updated - this is not a static website, but an on-line application - is so overwhelming that unless the application is extraordinary complex, more manpower is being put into the getting the UI in every supported browser right. This energy should be going into the application and its functionality.
There is indeed a lot of work to be done on a framework from which one can develop these applications, but its components should be recycled very easily, especially when your company's main business is developing on-line applications;-)
Okay... I'll do the stupid things first, then you shy people follow.
Ideally, browsers should develop to the point where they understand XML as well as HTML and XSL as well as CSS.
That's the ideal situation...
The current situation however is that there is a plethora of browsers, which is growing rapidly, with big differences among them, between OSes and even between versions... To develop a number of websites, one can simply not assume that users will have a specific browsers on a specific OS of a specific version...
We might evolve to a pure XML/XSL/CSS browser eventually, but until then, there has to be a different solution that can serve today... You would be amazed how much people still use Netscape 3, just because they don't have the urge to upgrade...
Java servlets are a technique, but again: it's built into the server. There are a number of servers out there, that don't have these servlets, so that another solution would come in reaallllyyyy handy.
Okay... I'll do the stupid things first, then you shy people follow.
Stability is of course one of the very basics in a server, and I'm not the person to criticize a distribution for not having the last minor revision number, but kernel 2.2 has been around for quite some time and is very stable indeed. It features a number of enhancements that I really like to have for some linux based routers: ipchains, quality of service and traffic shaping...
Kernel 2.2 is no "minor" revision, but a major one, and one which has proven to be stable.
I don't mind to "not have" the latest release of everything, but I just can't imagine that it isn't possible of getting a solid distribution out the door faster than what's being done now.
Debian touts to be the distribution for "politically correct" linuxfans, which makes it even more a pity that it seems to be so problematic to get the release cycle straightened out. Being the biggest non-commercial distribution, I wouldn't like to see it fall behind just because at the organisational level things aren't working out as they should be.
I'm quite new to debian (as I explained earlier), so I may have missed some points that are obvious to all that have been involved with Debian sooner than me, but there are some things that just don't make sense to me... I don't know why there are "closed mailing lists" by an "open source" distribution, neither do I know howcome there are so many questions from the how-can-we-get-involved-department here while at first glance Debian doesn't get around to get these people involved in an organized manner. Might it be that it's time to do some more thinking to get Debian on track to be what it should be: a great distribution for everyone, with help from everyone.
disclaimer: I'm not trying to start the next distro-flame-war-thingie, since I'm using Debian and actually liking it... I only want to make clear how I see it -an open source opinion, if you wish-;-)
Okay... I'll do the stupid things first, then you shy people follow.
Personally, I'm a slackware man who has just recently started using debian due to the lack of another distro that I like on DEC alpha based machines.
The current stable debian distribution "slink", is still based on the 2.0.x kernel, while the other big players couldn't adopt 2.2.x fast enough just because of the version number. With the 2.4 kernel just around the corner, how will Debian be able to compete with i.e. Red Hat, as they even might launch the next version with the 2.4 kernel as the Debian maintainers are still working on getting potato out the door?
What are the plans to do something about it? Hon the lot of geeks that are eager to help out join in, or what are the criteria or needed skills? Does one have to be an über-geek to get in? Okay... I'll do the stupid things first, then you shy people follow.
Watchguard, with their hip red FireBoxes based on a Linux kernel also do this: the source of their hardened Linux Kernel is not on the installation disks, but the clients can download it from their website of FTP server if they wish.
Before that, the source wasn't online, but anyone who grabbed a phone or sent an email to their tech support crew, could get that kernel source as well.
Okay... I'll do the stupid things first, then you shy people follow.
The drivers for the Digi board are GPLed, which of course doesn't make it illegal to port it, but the result has to be GPLed as well. *BSD does not work with the GPL, they have their own BSD-licence, which is different from the GPL in a lot of ways, and porting the code from the Digi drivers to *BSD would require the resulting *BSD code to be GPLed as well, which is against the religion of the BSDers out there;-)
Delphi however, is Borland's own code with their own licence, and they do with it as they please, so there is no licencing problem by just compiling/porting it on/to a different OS. They do a binary-release with the Borland licence attached, and those who don't just want free beer and/or free speech, can get their wits together to code a good alternative, the others are happy with the closed-source compiler Borland is selling. It is a quality product, even if it's just binaries...
So he has a point in telling that it's impossible to port the GPLed code for digi himself (since it is GPLed), while Borland can easily recompile their pascal compiler on every OS they want, without having to follow strict rules as is the case with GPL.
Okay... I'll do the stupid things first, then you shy people follow.
Stairs are a no-no, unless you cheat when "climbing" or "descending".
Since each stair is about 20cm above the other, so for a realistic simulation, the exoskeletonfeetthingies should also lock with that height difference. Climbing a staircase would effectively have the robotarms lock up 20cm higher every time, so the player is really going upward.
This goes for all movements that aren't happening on a leveled surface, but which needs some amount of climbing.
One possible solution for this is just by cheating a little bit by letting the robotarms adjust the "ground level" slowly: a bit like the movement that is the result of climbing an escalator that goes down at the exact same speed as the escalator is moving. The sum of movement is zero, and after a hundred steps or so, you're really convinced you climbed those steps. Throw in that the visual input of the player also suggests climbing, so that it can be perfectly possible to get a real stairclimbing experience. I won't even begin to think how utterly fast this pseudo-escalator-VR-controller has to be and how this could be realised, but that's a nice challenge to the physics/CS/AI/... ppl out there. Maybe do the prototype with Lego mindstorm?;-)
Of course this technique can be problematic when the movements go too fast, or even when people don't really react on it: in stead of a true experience the player gets nausea. Acceleration en decelaration are also issues that surface when the players are walking a bit faster or even trying to run... Equilibrium (or whoever you English-speaking ppl call that) is something tricky, and you need the coöperation of that organ in your ear to get that sense perfect. Sticking a couple electrodes in your brain to manipulate somethings up there may be the ultimate answer here *grin* Cronenberg already knew that, ofc;-)
Either way, it will be a lot of thinking and hard work before we can get that controller for our playstation at home...
PS: For those who don't like the almighty metric system: swap your favorite stairheight in feet or inches:-)
Okay... I'll do the stupid things first, then you shy people follow.
A XOR of two bits is completely random, as long as at least one of the bits is completely random.
The whole idea of a stream cipher is built on that concept:
You take a bit of the message you want to encrypt.
You XOR it with a bit from a stream of "random" bits
voila!: the bit is encrypted.
Anyone who doesn't know with which random bit the original bit was encrypted, has no way to "crack" the encryption.
The drawbacks:
The recipient needs to know which "random bits" to XOR with the encrypted bit.
This can be done by using the whole stream as key, so that you have a OTP encryption: unbreakable, but with horrendous key distribition problems...
Or it can be done by using a random number generator that consequently generates the same stream of bits, dependant on an initialization vector (aka: a Key)...
The alignment should be perfect, so the loss of 1 bit in transmission causes the rest of the message to be intellegible. There are numerous error correction codes for this.
By reading a lot of these comments, there aren't that many readers that know what exactly a "stream cipher" is... It's not encryption on a stream of data (like SSH and stuff)... Bram Okay... I'll do the stupid things first, then you shy people follow.
Anything on those Domino vulnerabilities that were making the rounds of Bugtraq a while back? A client of mine was bitten by one of these a few months ago...
Most of these things should be called "do not ever use domino out of the box", since the "flaws" exposed in BugTraq only consist of pretty good examples why you should have a skilled administrator handling your servers.
Lotus Notes/Domino is one of the more secure systems I have seen, provided that:
Your administrators do a good job on securing the machine and the ACLs...
Your developers know a little bit of the security bits too...
The underlying OS is properly secured (which hopefully should become a littlebit better with Linux support;-)...
The only nasty tidbits on Notes 4.5/4.6 is the notoriously weak SMTP-implementation, in which you can easily DOS a server or use it as a relay, and some hiccups in the connection manager, also providing the possibility of DOS'ing the server. The first can be avoided by using a decent SMTP gateway between the notes server and the outside world (linux+sendmail for example), the latter was fixed in one of the QMRs.
There was also a glitch in which the client didn't encrypt sent messages as it should, but that's also fixed, and I have yet to discover any bug which made it possible to compromise security... There's no such thing as a "Notes Melissa", and all the code that is executed on either client or server has to be signed by the developers, so macro virii won't stand a chance in a good administered environment.
Notes is far from perfect, but in the hands of the right people, it does a great job security-wise...
Anyway, that's my 2 eurocent.
I've got a message for all the beautiful people of the world... THERE ARE A LOT MORE OF US UGLY MOTHERF*CKERS AROUND THAN YOU ARE!
"It's interesting to note that there is not a single TPC result on any database running on Linux, and therefore Linux has yet to demonstrate their capabilities as a database server."
According to Siemens, a Linux-based SAP/3 quad Xeon put down the fastest benchmark set on an 4-way Intel system. Read it here, or check out the story on/. september 13th.
I don't think Siemens is the kind of company to give "anecdotal" evidence about "the capabilities of Linux as a database server". SAP is probably one of the most robust ERP/TP systems out there.
I've got a message for all the beautiful people of the world... THERE ARE A LOT MORE OF US UGLY MOTHERF*CKERS AROUND THAN YOU ARE!
ext2 works perfectly with 64bit, just ask anyone who has alphaLinux running. The problem is not in the ext2fs filesystem, but in the File I/O API. That API uses signed integers as argument (and returncode), so on a 32-bit system you've got only 2^31 bits, which is 2Gig.
On a 64-bit system the current ext2fs limit is 2Terabyte (if I'm not mistaken:-)
With the upcoming xfs from SGI to Linux, the filesize problem won't be solved on 32-bits systems, just because the problem isn't in the filesystem, but in the file I/O API itself...
I wouldn't want to be working with applications that require 2G+ files on intel anyhow. Especially not databases.
I've got a message for all the beautiful people of the world... THERE ARE A LOT MORE OF US UGLY MOTHERF*CKERS AROUND THAN YOU ARE!
Slashdot Mirror
If you have any pointers, I would very much appreciate it if you could contact me on bram_at_e-wareness_dot_be.
Thanks.
Okay... I'll do the stupid things first, then you shy people follow.
I have a number of Alpha machines in production, currently running Debian, and I would love to have the BSD-style rc.d system on these...
Does anyone have experience with this?
Okay... I'll do the stupid things first, then you shy people follow.
Slackware doesn't use a silly configuration tool with umpteen configuration files to keep the settings...
It has a limited number of start-up scripts which are pretty straight-forward, so that you can customize them easily at your situation.
Since they are that simple and straightforward, I've learned quite a lot about how everything is arranged at boot-up, which is a lot more difficult with Red Hat, SuSe or Debian, where there seem to be hundreds of directories with files and symlinks... Especially newbies who want to go a bit further than playing with their GUI config tool find this very overwhelming.
Basically, slackware is a do-it-yourself distro, since their package management system isn't really that good. But if you know your way around, I think it has a lot to offer in system stability and insight in the inner workings.
It's a pity it's only x86... I would really love to have a comparable distro for my Alphas :-)
Okay... I'll do the stupid things first, then you shy people follow.
The 800Mhz Athlon has a 2/5 cache divider, which yields a cache speed of 340 Mhz.
The 1Ghz Athlon has a 1/3 cache divider, which yields a cache speed of 333Mhz.
The effective cache speed of the 1Ghz chip is thus slower than the cache of the 800Mhz chip.
Your claim is thus blatantly wrong.
Okay... I'll do the stupid things first, then you shy people follow.
23. Nielsen, J.: Tooltips on webpages, U.S. Patent 5,937,417 (1999)
11. Nielsen, J.: Password helper using a client-side master password which automatically presents the appropriate server-side password to a particular remote server, U.S. Patent 6,006,333 (1999) [don't Windows .PWL files do something similar?]
50. Nielsen, J.: System and method for temporally varying pointer icons, U.S. Patent 5,784,056 (1998)
A whole number of them are from 1999, which means that you're frantically inventing great new things and spend tens of thousands of $$$ just patenting the things... There are about 30 patents listed in 1999, which means 2 patents every three weeks, for an amount of at least 500$K patenting costs and patent attorney fees to get them patented...
This all seems very Amazon-like, and either I've missed something really important, or is something else going on?
Okay... I'll do the stupid things first, then you shy people follow.
I know litterally hundreds of little businesses with 2/3 people, who only really want to share files and printers... There are thousands of users out there that have the computer in their own little business just because they need it, but wouldn't know a mouse from a keyboard if they didn't read the manuals. They don't want to spend any money more than really necessary on computer stuff, let alone that they would hire someone to set it up and troubleshoot it for them...
Even with dhcp, setting up tcp/ip is a real nightmare for them under Win95/98. (The TCP/IP protocol isn't installed automatically with windows 95, remember). The sheer number of options is very confusing, let alone that one could do it with static IPs without knowing what the heck these numbers mean. They want to give these computers names, not numbers.
For those computer illiterates, an appliance like the Cobalt Qube, with NetBEUI support would be a real plug-n-play solution, because there really is no setting up NetBEUI. Just to store their files, and get to use one printer from different computers. That's all they really need, and all they should be worried about.
Okay... I'll do the stupid things first, then you shy people follow.
Their cards are based on the ignitio chipsets, which makes it quite performant and stable under Linux...
And they've been doing so long before the linux hyped.
Okay... I'll do the stupid things first, then you shy people follow.
Quantum computing, DNA computing,... which method would you prefer?
Cracking encryption could become unfeasible with quantul computing as well, provided we jack up the keylength a bit more...
Or would distributed.net have those quantum CPU idle times working on the case by then? Oh my ;-)
Okay... I'll do the stupid things first, then you shy people follow.
There is indeed a lot of work to be done on a framework from which one can develop these applications, but its components should be recycled very easily, especially when your company's main business is developing on-line applications ;-)
Okay... I'll do the stupid things first, then you shy people follow.
The current situation however is that there is a plethora of browsers, which is growing rapidly, with big differences among them, between OSes and even between versions...
To develop a number of websites, one can simply not assume that users will have a specific browsers on a specific OS of a specific version...
We might evolve to a pure XML/XSL/CSS browser eventually, but until then, there has to be a different solution that can serve today... You would be amazed how much people still use Netscape 3, just because they don't have the urge to upgrade...
Java servlets are a technique, but again: it's built into the server. There are a number of servers out there, that don't have these servlets, so that another solution would come in reaallllyyyy handy.
Okay... I'll do the stupid things first, then you shy people follow.
It features a number of enhancements that I really like to have for some linux based routers: ipchains, quality of service and traffic shaping...
Kernel 2.2 is no "minor" revision, but a major one, and one which has proven to be stable.
I don't mind to "not have" the latest release of everything, but I just can't imagine that it isn't possible of getting a solid distribution out the door faster than what's being done now.
Debian touts to be the distribution for "politically correct" linuxfans, which makes it even more a pity that it seems to be so problematic to get the release cycle straightened out. Being the biggest non-commercial distribution, I wouldn't like to see it fall behind just because at the organisational level things aren't working out as they should be.
I'm quite new to debian (as I explained earlier), so I may have missed some points that are obvious to all that have been involved with Debian sooner than me, but there are some things that just don't make sense to me... I don't know why there are "closed mailing lists" by an "open source" distribution, neither do I know howcome there are so many questions from the how-can-we-get-involved-department here while at first glance Debian doesn't get around to get these people involved in an organized manner.
Might it be that it's time to do some more thinking to get Debian on track to be what it should be: a great distribution for everyone, with help from everyone.
Okay... I'll do the stupid things first, then you shy people follow.
The current stable debian distribution "slink", is still based on the 2.0.x kernel, while the other big players couldn't adopt 2.2.x fast enough just because of the version number. With the 2.4 kernel just around the corner, how will Debian be able to compete with i.e. Red Hat, as they even might launch the next version with the 2.4 kernel as the Debian maintainers are still working on getting potato out the door?
What are the plans to do something about it? Hon the lot of geeks that are eager to help out join in, or what are the criteria or needed skills? Does one have to be an über-geek to get in?
Okay... I'll do the stupid things first, then you shy people follow.
Before that, the source wasn't online, but anyone who grabbed a phone or sent an email to their tech support crew, could get that kernel source as well.
Okay... I'll do the stupid things first, then you shy people follow.
*BSD does not work with the GPL, they have their own BSD-licence, which is different from the GPL in a lot of ways, and porting the code from the Digi drivers to *BSD would require the resulting *BSD code to be GPLed as well, which is against the religion of the BSDers out there
Delphi however, is Borland's own code with their own licence, and they do with it as they please, so there is no licencing problem by just compiling/porting it on/to a different OS. They do a binary-release with the Borland licence attached, and those who don't just want free beer and/or free speech, can get their wits together to code a good alternative, the others are happy with the closed-source compiler Borland is selling. It is a quality product, even if it's just binaries...
So he has a point in telling that it's impossible to port the GPLed code for digi himself (since it is GPLed), while Borland can easily recompile their pascal compiler on every OS they want, without having to follow strict rules as is the case with GPL.
Okay... I'll do the stupid things first, then you shy people follow.
Since each stair is about 20cm above the other, so for a realistic simulation, the exoskeletonfeetthingies should also lock with that height difference. Climbing a staircase would effectively have the robotarms lock up 20cm higher every time, so the player is really going upward.
This goes for all movements that aren't happening on a leveled surface, but which needs some amount of climbing.
One possible solution for this is just by cheating a little bit by letting the robotarms adjust the "ground level" slowly: a bit like the movement that is the result of climbing an escalator that goes down at the exact same speed as the escalator is moving. The sum of movement is zero, and after a hundred steps or so, you're really convinced you climbed those steps. Throw in that the visual input of the player also suggests climbing, so that it can be perfectly possible to get a real stairclimbing experience. I won't even begin to think how utterly fast this pseudo-escalator-VR-controller has to be and how this could be realised, but that's a nice challenge to the physics/CS/AI/... ppl out there. ;-)
Maybe do the prototype with Lego mindstorm?
Of course this technique can be problematic when the movements go too fast, or even when people don't really react on it: in stead of a true experience the player gets nausea. ;-)
Acceleration en decelaration are also issues that surface when the players are walking a bit faster or even trying to run... Equilibrium (or whoever you English-speaking ppl call that) is something tricky, and you need the coöperation of that organ in your ear to get that sense perfect. Sticking a couple electrodes in your brain to manipulate somethings up there may be the ultimate answer here *grin* Cronenberg already knew that, ofc
Either way, it will be a lot of thinking and hard work before we can get that controller for our playstation at home...
PS: For those who don't like the almighty metric system: swap your favorite stairheight in feet or inches :-)
Okay... I'll do the stupid things first, then you shy people follow.
The whole idea of a stream cipher is built on that concept:
You take a bit of the message you want to encrypt.
You XOR it with a bit from a stream of "random" bits
voila!: the bit is encrypted.
Anyone who doesn't know with which random bit the original bit was encrypted, has no way to "crack" the encryption.
The drawbacks:
The recipient needs to know which "random bits" to XOR with the encrypted bit.
The alignment should be perfect, so the loss of 1 bit in transmission causes the rest of the message to be intellegible. There are numerous error correction codes for this.
By reading a lot of these comments, there aren't that many readers that know what exactly a "stream cipher" is... It's not encryption on a stream of data (like SSH and stuff)... Bram
Okay... I'll do the stupid things first, then you shy people follow.
Most of these things should be called "do not ever use domino out of the box", since the "flaws" exposed in BugTraq only consist of pretty good examples why you should have a skilled administrator handling your servers.
Lotus Notes/Domino is one of the more secure systems I have seen, provided that:
Your administrators do a good job on securing the machine and the ACLs...
Your developers know a little bit of the security bits too...
The underlying OS is properly secured (which hopefully should become a littlebit better with Linux support ;-)...
The only nasty tidbits on Notes 4.5/4.6 is the notoriously weak SMTP-implementation, in which you can easily DOS a server or use it as a relay, and some hiccups in the connection manager, also providing the possibility of DOS'ing the server. The first can be avoided by using a decent SMTP gateway between the notes server and the outside world (linux+sendmail for example), the latter was fixed in one of the QMRs.
There was also a glitch in which the client didn't encrypt sent messages as it should, but that's also fixed, and I have yet to discover any bug which made it possible to compromise security... There's no such thing as a "Notes Melissa", and all the code that is executed on either client or server has to be signed by the developers, so macro virii won't stand a chance in a good administered environment.
Notes is far from perfect, but in the hands of the right people, it does a great job security-wise...
Anyway, that's my 2 eurocent.
I've got a message for all the beautiful people of the world...
THERE ARE A LOT MORE OF US UGLY MOTHERF*CKERS AROUND THAN YOU ARE!
According to Siemens, a Linux-based SAP/3 quad Xeon put down the fastest benchmark set on an 4-way Intel system. Read it here, or check out the story on /. september 13th.
I don't think Siemens is the kind of company to give "anecdotal" evidence about "the capabilities of Linux as a database server". SAP is probably one of the most robust ERP/TP systems out there.
I've got a message for all the beautiful people of the world...
THERE ARE A LOT MORE OF US UGLY MOTHERF*CKERS AROUND THAN YOU ARE!
ext2 works perfectly with 64bit, just ask anyone who has alphaLinux running. The problem is not in the ext2fs filesystem, but in the File I/O API. That API uses signed integers as argument (and returncode), so on a 32-bit system you've got only 2^31 bits, which is 2Gig.
On a 64-bit system the current ext2fs limit is 2Terabyte (if I'm not mistaken :-)
With the upcoming xfs from SGI to Linux, the filesize problem won't be solved on 32-bits systems, just because the problem isn't in the filesystem, but in the file I/O API itself...
I wouldn't want to be working with applications that require 2G+ files on intel anyhow. Especially not databases.
I've got a message for all the beautiful people of the world...
THERE ARE A LOT MORE OF US UGLY MOTHERF*CKERS AROUND THAN YOU ARE!