The google paper said that there were SMART parameters that did indicate failure (but only a few parameters have a strong correlation with failure). The problem is that those parameters do not change in MOST failure cases - i.e. your disk can die without any warnings from SMART (StorageMojo summarises the Google paper and here's the original Google Failure Trends in Disks PDF).
If (for example) the reallocated sector count is high I don't think it's a matter of if but when your drive will fail. A count of 1 doesn't guarantee failure but indicates a higher probability than usual of imminent failure. From page 7 of the PDF:
After their first reallocation, drives are over 14 times more likely to fail within 60 days than drives without reallocation counts, making the critical threshold for this parameter also one.
According to this NetApp reply to an open letter on storagemojo while the electronics of the drive beyond the interface may be the same on consumer and enterprise drives, the way the firmware behaves is not. The consumer drive firmware apparently do all it can to try and read data back even if it makes the drive temporarily unavailable and trusts additional information less that enterprise firmware.
The sort of "scheduling" you are talking about sound like block reordering. This is where you try and group requests for blocks that you guess are going to be in a similar part of the disk together in the hopes of speeding things up. It's absolutely true that today's disks bear less and less resemblance to the old cylinders, sectors and heads of old disks and most disks have their own cache which can do reordering (not to mention the silent remapping that modern disks do when sectors go bad). Unless the disk cache queue is very deep though, I suspect there is some advantage to doing I/O scheduling because you might be able to wait longer before making a decision as to what to read thus avoiding the worst case scenario where you seek between disparate files continuously.
This sort of I/O scheduling also goes further. For example, with cfq you can arrange for a daily file scan to have it's I/O queued up behind anything else so your web browser is not made to stall (so long) when writing to disk. When disks become capable of this type reordering (I believe MS are pushing for this) then again there will be less need for the OS to do it.
I wonder, if the originating process' priority is taken into account at all... It has always annoyed me, that the "nice" (and especially the idle-only) processes are still treated equally, when it comes to I/O...
Best effort. This is the default scheduling class for any process that hasn't asked for a specific io priority. Programs inherit the CPU nice setting for io priorities.
Answer to 2) The f after 0.5 and the 1.5 simply means "use the floating point version of this number". Multiplying two numbers which have different types (e.g. multiplying an int and a float) or doing assignment between different types means that one of the numbers will have to be automatically converted to the other's type and which conversion happens depends on the types of both numbers. By default, I think type of the literal 0.5 (without any letters after it) is a double. By adding the f (assuming x is a float) the types of the operands are the same so there is no implicit type conversion thus eliminating the chance of the wrong conversion happening automatically. You care about what type of multiply is done because certain types can lead to unwanted loss of precision.
As for 1) Well if you iterate by one step you don't need a loop. Presumably most looping iteration happens because your first guess wasn't accurate enough. If your first guess WAS accurate enough why would you needlessly calculate more accuracy? I'm only speculating though - I don't know if this is really the case.
On Linux, address space randomisation is done by various kernel patches which aren't selinux related. Red Hat uses execshield (which went mainline around 2.6.11) but there have been patches from PaX which also implemented it. See RH Mag on Execshield and Wikipedia's address space layout randomisation article.
Online hard disk resizing. This has only recently (last two years) been possible on Linux and is filesystem dependent (on ext3 you need to have made your partitions within LVM, I think reiser3 can online grow even with regular partitions) I've not been aware of an NTFS online resizer until now though. I will also point out that your school (and probably you) are at some point paying for all that "free" MS software though.
As for ACPI under Linux - there can be a whole stack of issues, some which are related to various BIOSes only looking for XP (and now Vista) and enabling/disabling various features depending on what they find. It is true that putting devices into low power mode doesn't tend to happen by default on Linux though. Power consumption will also be helped when the tickless/dyntick kernel patches arrive.
I know it does because I've tried it myself but I think you get some sort of warning right? Yahoo "grade" which browsers they support based upon browser version and platform (there are four categories: A, C, X and unsupported). I was going to say that Firefox on the Mac is a little bit unusual and thus would be a lower grade but according to this Yahoo browser grades table Firefox on OSX is a grade A platform. Hard to know what's up there...
I'm just another opensuse user (actually I'm relatively recent user having avoided SUSE for years due to the inability to download the ISOs for free). As I'm an administrator of a few (60+) opensuse desktops too, I'm going to guess that the not-ready component you included was novell-zmd right? That thing is painful... Is it as slow for you as it is for me? Why does it go to sleep? People who think yum is slow should give this thing a whirl just after boot... Plus I never thought I'd say this but I'm glad that yast was around or I would have struggled to fetch those updates (yeah I know - fetch the RPMs and update like the bad old days)...
It has to be said that your company is doing some extremely interesting (well to me) work. Little things like the opensuse firefox print dialog displaying CUPS printers (yes I know about the bug where if a printer disappears it makes firefox crash when you go to the print dialog but so do your folks because I found it in your bugzilla). Other things like the ability to delegate the ability to install and update software to a user are absolutely fantastic at a site like the university where I work. I have been enjoying using XGL (although it took a bit of futzing with config files to get it going under KDE which is what we are still using here and why oh why did you change the way the zoom keybinding worked? We had folks who loved the way the zoom stuck because it acted like a magnifying glass that you didn't have to hold...). Our laptop users like the NetworkManager setup although I suspect getting the Wifi to work with the University's PPTP VPN is still going to be a step too far. I even noticed that OOo started much faster than I'd ever seen it in other distros (although I do now see there is some sort of autoloader). Global proxy configuration is also good.
However there are a couple of gotchas like gaim always using a proxy for jabber (!) if a global proxy is set (already known to your folks), the mysterious and yet repeated kmail crashes that I keep having (I haven't had a chance to look that one up) and the bizzare choice to pull up the current kernel when a new kernel is installed (this led to USB harddisks no longer working in the running kernel and has now scared users off applying updates).
Where was I going with this? Oh yeah, you do seem to include a lot of closed stuff in your distro but for the most part people seem to want it (acrobat, realplayer and flash are all popular and the sneaky packaging of a 32 bit firefox side steps the 64 bit owners wrath for now). Your documentation isn't bad (the wiki is somewhat useful on issues like packaging). However what you are doing with SLED (and indeed Opensuse) feels remiencient of Red Hat do with their RHEL which in my book is the right way to go. So long as the SRPMs for the open source parts and you don't start bundling closed source drivers (keep 'em as a seperate download) I feel branding it as wholely proprietary is unfair.
3rd Party Frontmotion Firefox MSI installer. Pros - readily available. Cons - from a 3rd party (trustworthiness, how long will it be supported for, will they start charging...)
I think it depends on which platform you are on but when you go to google video and look at the far right there is a download option that lets you download the video in MPEG4 format if you select Windows/PC.
Yes I can arrange for administrative rights on all the target machines (I believe due to their set up anyone who can access the Administrator account on one will have the same rights on another). And yes they are all part of the same domain so if that is not workable a user with access to all of them can be created.
I've cast a quick eye over unattended and it certainly DOES look interesting (the points discussing other installation processes were enlightening too).
To the person who suggested frontmotion - that's interesting but I would like a en_GB build, it throws away the Mozilla branding (because it's not officially blessed) and has the disadvantage of not being a trusted source.
To the person who posted a bugzilla link - thanks, a quick skim is already proving informative.
I happen to be an admin at a UK university and the thing that bugs me is how to keep Firefox up to date on Windows (on Linux this is a non issue). Because of this sole point, I am unlikely to roll it out across our Windows labs. What are folks doing when the people using the machines don't have the rights to install software globally? More explicitly, what are people doing when they don't have Zenworks or Active Directory for software distribution? Do you just reimage/ghost all your machines?
The answer is doubtless obvious but I'm more than happy to be clued in.
In the slashes case I can think of one good reason why google wouldn't do it - to make the page smaller (few people serve so many pages that this is actually a valid excuse but I can well believe saving one byte would save Google money on heavily trafficked pages). However that typo is less explicable - that just seems wrong (unless someone is making use of a parser error or something). Google's XHTML mobile page doesn't seem to have too many errors in but yeah, there's room for improvement.
I am not a regular SUSE user but I have used SUSE in the past and as an administrator at a site with a small (few 100) Linux desktops I have recently been testing SUSE 10.1 as the site has always used SUSE.
Take all of these replies with a grain of salt. I haven't filed problems in Novell's bugzilla and anyone complaining about things but not filing bugs probably isn't interested in helping to make things better.
1. Package management. This is a curious one as Yast's dep solving seems to now be done by the Novell zmd daemon. This daemon seems to have a XML-RPC interface and consequently can be driven by many interfaces. Yast is one GUI front end, rug is a text driven front end and there are some GTK zmd-installer/zmd-updater tools too. Alas Yast's package groups don't yet appear to be rug bundles which is a little frustrating. In SUSE 10.1 there is also a bug which renders the zmd* and rug unable to resolve dependencies. You will have to use Yast to get an update that fixes this. 2. SUSE has alway had the option within the Yast control centre to upgrade to a new version of SUSE. I haven't tested rug's ability to do this but after patching it seems fairly capable. The bad news is that the dep solver is slower than Red Hat/Fedora's yum (interesting yum is written in Python and novell-zmd is written in mono). The daemon can go to sleep and takes time to wake up (I also wonder if it refetches the list of packages on remote sources every time it wakes up). It eats a lot of memory and CPU when solving but might have lots of interesting features like being able to have updates "pushed" to it (handy when you have lots of machines). 3. There are drop shadows in compiz under XGL when I used it. 4. This is true. Compiz lacks metacity's focus stealing prevention, you can't drag windows off the top and bottom of the screen. You can configure keybindings and it seems to follow GNOME's theme (but not KDEs). It's very usuable and I believe was based off metacity originally (for Wm decisions). 5. Dunno. Perhaps they want to let it stablise (Ubuntu carries a bunch of patches to stabalise its version of 2.14).
There are things in SUSE 10.1 that definitely make it more attractive for large installations (proxy management is far better than the other distros I've used). I need more time to evaulate beagle, XGL works fairly well for me even with KDE (getting XGL going on the integrated Intel graphics cards with open source drivers was a dream compared to the ATI binary drivers). However there are also things broken within SUSE that are not broken elsewhere due to their patches (gaim + jabber + proxy = have to use proxy?!) but other places where their patches are absolutely amazing (I've never seen openoffice start so quickly from cold boot, evolution has right mouse button spelling suggestions!). Until I get round to filing bugs I'm going to leave my criticism there though.
I used an Ubuntu Dapper Flight text install CD to fix up broken FC5test software RAID partitioning (the bug was fixed in the FC5 final anaconda). Ubuntu definitely understands LVM and Linux software RAID and be made to create and mount such partitions (the LVM scripts are definitely there). I don't know what the GUI software support is like after the install though.
If you cut the firmware out of your Windows drivers you should be able to get the native linux driver working doing away with ndiswrapper. See the Wiki page on BCM43xx on dapper for details
(The firmware can't be shipped with the distro due to license issues)
All I can say having quickly flipped through the book is that there are send and recv and notify primatives. Notify fiddles a bitmap and send/recv work on fixed size messages and require the receiver to be waiting when the sender sends a message. I'm guessing that bytes are still copied.
Any chance you can narrow where this would be down to a filename? : )
First up I helped develop a Linux USB webcam driver. These things can sometimes push a serious amount of data about the place and the camera won't wait for you before sending more. Being at the mercy of scheduler because the decoding is in userspace could well mean performance is poor (where I was bayer decoding was done in kernel precisely for speed concerns). The other thing is that if the USB driver is in the kernel the output can be exposed as a video device rather than forcing each program to implement each camera's USB dialect. If you want to use established video playing software this last point is fairly key.
As for stable kernel ABIs... they will only make things worse as fewer people are able to debug your problem. The only people who have the resources to cope with support for a binary driver platform (which are the only winners from a fixed ABI) for hardware with as many permuations as your typical PC are Microsoft (and on Windows people even circumvent measure designed to improve drivers). You need more not fewer people as the amount of hardware possiblities go up and a kernel ABI actively discourages this.
As for the space requirements for shipping all those drivers... quite frankly they are negligble compared to the size of disks in your modern desktop computer. Saving even 500Mbytes of hard disk is probably a false economy when your typical disk size goes over 50Gbytes.
Things like the O(1) scheduler and general scheduler improvements, kernel pre-emption (good for audio), ipv6 stack improvements, major changes to the USB subsystem including increased robustness, udev and better hotplug support, VM improvements, major SMP improvements, adoption of ALSA for improved soundcard support (e.g. surround and software mixing), improved build procedure, FAR better ACPI support, suspend to disk support, samba filesytem performance improvements in CIFS, TCP/IP stack improvements...
The Wonderful World of Linux 2.6 page has a more comprehensive list of Linux 2.6 improvements over Linux 2.4 than what I have just mentioned above. Just because the machine I am currently sitting at is fairly old I would not want to be stuck with an old 2.4 kernel because I still benefit from 2.6's changes.
Without going as far as far as DBSD's threading it would appear you could reduce the number of locks currently used in the Linux TCP stack at the expense of adding a bit of the TCP stack to userland. Take a look at these slides on Van Jacobson's Net Channels where Linux's TCP performance is improved by implementing channels (seen via Dave Miller's blog).
This is what microkernels try to do by forcing things through message passing (as much of a driver as possible runs in a different lower priviledged address). However to actually get at the hardware you HAVE to have the top level permission This seems to be just down to current architectures having various rings of security.
Further, some drivers of hardware need access to bits of memory that overlap with the memory of other processes or the kernel itself thus providing room for stomping over someone else's stuff.
Slashdot Mirror
Wasn't this posted a few weeks ago?
If (for example) the reallocated sector count is high I don't think it's a matter of if but when your drive will fail. A count of 1 doesn't guarantee failure but indicates a higher probability than usual of imminent failure. From page 7 of the PDF:
According to this NetApp reply to an open letter on storagemojo while the electronics of the drive beyond the interface may be the same on consumer and enterprise drives, the way the firmware behaves is not. The consumer drive firmware apparently do all it can to try and read data back even if it makes the drive temporarily unavailable and trusts additional information less that enterprise firmware.
Scott James Remnant gave an upstart talk at this year's linuxconf.au. There's mention of launchd, initng and other sysv init replacements along with a discussion to moving to event based launching.
The sort of "scheduling" you are talking about sound like block reordering. This is where you try and group requests for blocks that you guess are going to be in a similar part of the disk together in the hopes of speeding things up. It's absolutely true that today's disks bear less and less resemblance to the old cylinders, sectors and heads of old disks and most disks have their own cache which can do reordering (not to mention the silent remapping that modern disks do when sectors go bad). Unless the disk cache queue is very deep though, I suspect there is some advantage to doing I/O scheduling because you might be able to wait longer before making a decision as to what to read thus avoiding the worst case scenario where you seek between disparate files continuously.
This sort of I/O scheduling also goes further. For example, with cfq you can arrange for a daily file scan to have it's I/O queued up behind anything else so your web browser is not made to stall (so long) when writing to disk. When disks become capable of this type reordering (I believe MS are pushing for this) then again there will be less need for the OS to do it.
Answer to 2)
The f after 0.5 and the 1.5 simply means "use the floating point version of this number". Multiplying two numbers which have different types (e.g. multiplying an int and a float) or doing assignment between different types means that one of the numbers will have to be automatically converted to the other's type and which conversion happens depends on the types of both numbers. By default, I think type of the literal 0.5 (without any letters after it) is a double. By adding the f (assuming x is a float) the types of the operands are the same so there is no implicit type conversion thus eliminating the chance of the wrong conversion happening automatically. You care about what type of multiply is done because certain types can lead to unwanted loss of precision.
As for 1)
Well if you iterate by one step you don't need a loop. Presumably most looping iteration happens because your first guess wasn't accurate enough. If your first guess WAS accurate enough why would you needlessly calculate more accuracy? I'm only speculating though - I don't know if this is really the case.
On Linux, address space randomisation is done by various kernel patches which aren't selinux related. Red Hat uses execshield (which went mainline around 2.6.11) but there have been patches from PaX which also implemented it. See RH Mag on Execshield and Wikipedia's address space layout randomisation article.
Online hard disk resizing. This has only recently (last two years) been possible on Linux and is filesystem dependent (on ext3 you need to have made your partitions within LVM, I think reiser3 can online grow even with regular partitions) I've not been aware of an NTFS online resizer until now though. I will also point out that your school (and probably you) are at some point paying for all that "free" MS software though.
As for ACPI under Linux - there can be a whole stack of issues, some which are related to various BIOSes only looking for XP (and now Vista) and enabling/disabling various features depending on what they find. It is true that putting devices into low power mode doesn't tend to happen by default on Linux though. Power consumption will also be helped when the tickless/dyntick kernel patches arrive.
I know it does because I've tried it myself but I think you get some sort of warning right? Yahoo "grade" which browsers they support based upon browser version and platform (there are four categories: A, C, X and unsupported). I was going to say that Firefox on the Mac is a little bit unusual and thus would be a lower grade but according to this Yahoo browser grades table Firefox on OSX is a grade A platform. Hard to know what's up there...
Hello Ted,
I'm just another opensuse user (actually I'm relatively recent user having avoided SUSE for years due to the inability to download the ISOs for free). As I'm an administrator of a few (60+) opensuse desktops too, I'm going to guess that the not-ready component you included was novell-zmd right? That thing is painful... Is it as slow for you as it is for me? Why does it go to sleep? People who think yum is slow should give this thing a whirl just after boot... Plus I never thought I'd say this but I'm glad that yast was around or I would have struggled to fetch those updates (yeah I know - fetch the RPMs and update like the bad old days)...
It has to be said that your company is doing some extremely interesting (well to me) work. Little things like the opensuse firefox print dialog displaying CUPS printers (yes I know about the bug where if a printer disappears it makes firefox crash when you go to the print dialog but so do your folks because I found it in your bugzilla). Other things like the ability to delegate the ability to install and update software to a user are absolutely fantastic at a site like the university where I work. I have been enjoying using XGL (although it took a bit of futzing with config files to get it going under KDE which is what we are still using here and why oh why did you change the way the zoom keybinding worked? We had folks who loved the way the zoom stuck because it acted like a magnifying glass that you didn't have to hold...). Our laptop users like the NetworkManager setup although I suspect getting the Wifi to work with the University's PPTP VPN is still going to be a step too far. I even noticed that OOo started much faster than I'd ever seen it in other distros (although I do now see there is some sort of autoloader). Global proxy configuration is also good.
However there are a couple of gotchas like gaim always using a proxy for jabber (!) if a global proxy is set (already known to your folks), the mysterious and yet repeated kmail crashes that I keep having (I haven't had a chance to look that one up) and the bizzare choice to pull up the current kernel when a new kernel is installed (this led to USB harddisks no longer working in the running kernel and has now scared users off applying updates).
Where was I going with this? Oh yeah, you do seem to include a lot of closed stuff in your distro but for the most part people seem to want it (acrobat, realplayer and flash are all popular and the sneaky packaging of a 32 bit firefox side steps the 64 bit owners wrath for now). Your documentation isn't bad (the wiki is somewhat useful on issues like packaging). However what you are doing with SLED (and indeed Opensuse) feels remiencient of Red Hat do with their RHEL which in my book is the right way to go. So long as the SRPMs for the open source parts and you don't start bundling closed source drivers (keep 'em as a seperate download) I feel branding it as wholely proprietary is unfair.
OK I'm going to collect all the information I've found on this topic into this post for posterity.
Bugzila bug discussing Firefox MSI installer. Discusses Mozilla's view on things and what different people want in order to deploy Firefox on large scale desktops.
Instructions on how to package official Firefox releases into an MSI.
UK University admin blog talking about Firefox and System Administration. Talks about deployment of Firefox on enterprise desktops, issues with GPO and links to projects and resources and pros and cons with various packaging attempts.
3rd Party Frontmotion Firefox MSI installer. Pros - readily available. Cons - from a 3rd party (trustworthiness, how long will it be supported for, will they start charging...)
I think it depends on which platform you are on but when you go to google video and look at the far right there is a download option that lets you download the video in MPEG4 format if you select Windows/PC.
I have a few minutes before I rush off to work...
Yes I can arrange for administrative rights on all the target machines (I believe due to their set up anyone who can access the Administrator account on one will have the same rights on another). And yes they are all part of the same domain so if that is not workable a user with access to all of them can be created.
I've cast a quick eye over unattended and it certainly DOES look interesting (the points discussing other installation processes were enlightening too).
To the person who suggested frontmotion - that's interesting but I would like a en_GB build, it throws away the Mozilla branding (because it's not officially blessed) and has the disadvantage of not being a trusted source.
To the person who posted a bugzilla link - thanks, a quick skim is already proving informative.
I happen to be an admin at a UK university and the thing that bugs me is how to keep Firefox up to date on Windows (on Linux this is a non issue). Because of this sole point, I am unlikely to roll it out across our Windows labs. What are folks doing when the people using the machines don't have the rights to install software globally? More explicitly, what are people doing when they don't have Zenworks or Active Directory for software distribution? Do you just reimage/ghost all your machines?
The answer is doubtless obvious but I'm more than happy to be clued in.
In the slashes case I can think of one good reason why google wouldn't do it - to make the page smaller (few people serve so many pages that this is actually a valid excuse but I can well believe saving one byte would save Google money on heavily trafficked pages). However that typo is less explicable - that just seems wrong (unless someone is making use of a parser error or something). Google's XHTML mobile page doesn't seem to have too many errors in but yeah, there's room for improvement.
I am not a regular SUSE user but I have used SUSE in the past and as an administrator at a site with a small (few 100) Linux desktops I have recently been testing SUSE 10.1 as the site has always used SUSE.
Take all of these replies with a grain of salt. I haven't filed problems in Novell's bugzilla and anyone complaining about things but not filing bugs probably isn't interested in helping to make things better.
1. Package management. This is a curious one as Yast's dep solving seems to now be done by the Novell zmd daemon. This daemon seems to have a XML-RPC interface and consequently can be driven by many interfaces. Yast is one GUI front end, rug is a text driven front end and there are some GTK zmd-installer/zmd-updater tools too. Alas Yast's package groups don't yet appear to be rug bundles which is a little frustrating. In SUSE 10.1 there is also a bug which renders the zmd* and rug unable to resolve dependencies. You will have to use Yast to get an update that fixes this.
2. SUSE has alway had the option within the Yast control centre to upgrade to a new version of SUSE. I haven't tested rug's ability to do this but after patching it seems fairly capable. The bad news is that the dep solver is slower than Red Hat/Fedora's yum (interesting yum is written in Python and novell-zmd is written in mono). The daemon can go to sleep and takes time to wake up (I also wonder if it refetches the list of packages on remote sources every time it wakes up). It eats a lot of memory and CPU when solving but might have lots of interesting features like being able to have updates "pushed" to it (handy when you have lots of machines).
3. There are drop shadows in compiz under XGL when I used it.
4. This is true. Compiz lacks metacity's focus stealing prevention, you can't drag windows off the top and bottom of the screen. You can configure keybindings and it seems to follow GNOME's theme (but not KDEs). It's very usuable and I believe was based off metacity originally (for Wm decisions).
5. Dunno. Perhaps they want to let it stablise (Ubuntu carries a bunch of patches to stabalise its version of 2.14).
There are things in SUSE 10.1 that definitely make it more attractive for large installations (proxy management is far better than the other distros I've used). I need more time to evaulate beagle, XGL works fairly well for me even with KDE (getting XGL going on the integrated Intel graphics cards with open source drivers was a dream compared to the ATI binary drivers). However there are also things broken within SUSE that are not broken elsewhere due to their patches (gaim + jabber + proxy = have to use proxy?!) but other places where their patches are absolutely amazing (I've never seen openoffice start so quickly from cold boot, evolution has right mouse button spelling suggestions!). Until I get round to filing bugs I'm going to leave my criticism there though.
Although that quote is used by John Thompson I suspect he didn't create it. There are usenet posts back to at least 1998 saying it is in statistics textbooks. A further look across the web suggests the torture data quote was originated by Ronald Coase.
I used an Ubuntu Dapper Flight text install CD to fix up broken FC5test software RAID partitioning (the bug was fixed in the FC5 final anaconda). Ubuntu definitely understands LVM and Linux software RAID and be made to create and mount such partitions (the LVM scripts are definitely there). I don't know what the GUI software support is like after the install though.
If you cut the firmware out of your Windows drivers you should be able to get the native linux driver working doing away with ndiswrapper. See the Wiki page on BCM43xx on dapper for details
(The firmware can't be shipped with the distro due to license issues)
All I can say having quickly flipped through the book is that there are send and recv and notify primatives. Notify fiddles a bitmap and send/recv work on fixed size messages and require the receiver to be waiting when the sender sends a message. I'm guessing that bytes are still copied.
Any chance you can narrow where this would be down to a filename? : )
First up I helped develop a Linux USB webcam driver. These things can sometimes push a serious amount of data about the place and the camera won't wait for you before sending more. Being at the mercy of scheduler because the decoding is in userspace could well mean performance is poor (where I was bayer decoding was done in kernel precisely for speed concerns). The other thing is that if the USB driver is in the kernel the output can be exposed as a video device rather than forcing each program to implement each camera's USB dialect. If you want to use established video playing software this last point is fairly key.
As for stable kernel ABIs... they will only make things worse as fewer people are able to debug your problem. The only people who have the resources to cope with support for a binary driver platform (which are the only winners from a fixed ABI) for hardware with as many permuations as your typical PC are Microsoft (and on Windows people even circumvent measure designed to improve drivers). You need more not fewer people as the amount of hardware possiblities go up and a kernel ABI actively discourages this.
Having to keep even the kernel API stable only encourages resistance to change and layers of glue in the kernel in an attempt to keep old binary drivers working (thus hurting future developments).
As for the space requirements for shipping all those drivers... quite frankly they are negligble compared to the size of disks in your modern desktop computer. Saving even 500Mbytes of hard disk is probably a false economy when your typical disk size goes over 50Gbytes.
To counter the people who keep saying "No stats! Where are the stats?" (which is a very fair point) take a look at this diary entry by Dave Jones (one of Red Hat's kernel hackers) talking about bugzilla numbers when rebasing kernels.
Things like the O(1) scheduler and general scheduler improvements, kernel pre-emption (good for audio), ipv6 stack improvements, major changes to the USB subsystem including increased robustness, udev and better hotplug support, VM improvements, major SMP improvements, adoption of ALSA for improved soundcard support (e.g. surround and software mixing), improved build procedure, FAR better ACPI support, suspend to disk support, samba filesytem performance improvements in CIFS, TCP/IP stack improvements...
The Wonderful World of Linux 2.6 page has a more comprehensive list of Linux 2.6 improvements over Linux 2.4 than what I have just mentioned above. Just because the machine I am currently sitting at is fairly old I would not want to be stuck with an old 2.4 kernel because I still benefit from 2.6's changes.
Without going as far as far as DBSD's threading it would appear you could reduce the number of locks currently used in the Linux TCP stack at the expense of adding a bit of the TCP stack to userland. Take a look at these slides on Van Jacobson's Net Channels where Linux's TCP performance is improved by implementing channels (seen via Dave Miller's blog).
This is what microkernels try to do by forcing things through message passing (as much of a driver as possible runs in a different lower priviledged address). However to actually get at the hardware you HAVE to have the top level permission This seems to be just down to current architectures having various rings of security.
Further, some drivers of hardware need access to bits of memory that overlap with the memory of other processes or the kernel itself thus providing room for stomping over someone else's stuff.