That's not actually wrong, depending on the database.
Filesystem snapshot backups work the same way. It's not the same as powering off/on a machine; you don't have to contend with filesystem corruption which is the leading cause of DB corruption - not sudden crashes.
They call this "Crash-consistent" backups.
A modern database engine is a transactional one. Transactions (any change to the database) are first written to a transaction log. The log is periodically written to the database. The log is then checkpointed to indicate that the transaction has been successfully written to the database.
If the database engine should crash or if the machine should crash and restart, the database engine will run a recovery because the database was not shut down cleanly. The recovery checks the transaction log to make sure all of the transactions have been written to the database. If there was a transaction being written to the database when the crash occurred, that transaction will be scrubbed and replayed from the log. If a transaction was being written to the log, that transaction is cancelled.
So, the worst case scenario is that you lose a couple transactions that were JUST being written to the log.
The likelyhood of a database being corrupted due to this process is very, very low. So, for most normal databases, this method is actually acceptable. For something very critical (such as order processing databases or something) you'll want to make sure you initiate some sort of database aware quiescence of the database file before taking the snapshot. This will tell the database engine to write all current transactions to disk and stop writing new ones for a moment so we can get our snapshot. This still leaves us with a crash-consistent copy of the database, but it's nearly guaranteed to be consistent when we start up.
Do you actually think the Sidekick server will be that big?
For one, it's on Windows. Danger is owned by Microsoft, and these servers are housed at a Microsoft data center. For two, you really don't need huge hardware anymore.. just redundant hardware.
A single dual-socket quad-core box with enough RAM could probably handle a million Sidekicks, if not more. There's such little data and it's requested over a cell phone data network. Not to mention they don't synchronize everything all the time. Just little changes.
The technology is available to get good, solid backups for anything. They just didn't use it, test it, verify it, etc. And in the case of this, users cannot back up their own data. And what they lost isn't backups.
I used to have one of these things.
The phone is (like someone above pointed out) a local cache of what's on the server side. The live database/back end is what crashed. When you make a change on the phone, it immediately sends that change to the server. You can login to the sidekick web site and make changes there, which appear quickly on your phone. If you reboot your phone, it will retrieve anything it needs from the server side. Apparently, the phone doesn't even keep a permanent local copy on some sort of non-volatile storage (hence "Don't turn off your phone.")
It's like someone that uses Google apps and stores all their documents on their system. If that system should go down, you'd be screwed, except that you COULD back up your documents locally. With this case, you can not.
I don't really like the term "cloud computing." All it means is server storage somewhere on the Internet. Under this term you could call any web site a "Cloud." It's ambiguous at best.
I'll agree with you on Far Cry. I loved that game! Until the monsters came around from every direction and turned it into a cheezy alien shoot'em'up that you couldn't win without cheating - unless you are an FPS master. I'm not bad at shooters, but I always preferred ones where you had to use tactics. I liked Counter Strike (well, Tactical Ops..) because of that. Far Cry was real cool how you'd sneak around and stuff in this ridiculously large, contiguous area.
I actually agree with the article when it comes to ad placement. We see them all around us - on billboards, on the store fronts themselves, on busses, trains.. I do like the satirical ones used in some games but ad placement is required then doing it in a real-world kind of way is okay with me and I think it probably would make a game that's aiming to be realistic even more so. GTA 4 is a good example of where real ad placement would make sense.
So you're sitting in traffic... and all you can hear is a thousand cars making a thousand "personalized" sounds (only $29.99 each, available on iTunes!) all around you.
Woah, hold on there, you missed the point entirely.
WHEN Windows 2000 was released, it didn't have enough driver support for a mass release. Microsoft would have taken an exorbitant amount of flack for this; probably worse than they did with Vista.
It was a TIMING thing, not a compatibility with XP. XP didn't come out until almost three years later, and by that time, driver support was much better BECAUSE of Windows 2000's availability for that time.
you really misunderstood the point on the Macintosh. The difference between a PC and a Mac these days is a different BIOS, but obviously I'm not talking about a Hackintosh. Apple takes off the shelf components, customized an EFI BIOS for it, slaps on an Apple sticker and off they go. Sure, they sometimes have custom board designs made to fix their cases, but the components are no different.
If you've never had a RAID card fail then you're lucky, but 50 servers isn't exactly a good sampling. I've worked in data centers with a thousand servers, and I've worked in a few of them. These days, with VMware stressing servers in a way that's probably never been done on commodity hardware, you get a lot more failure out of RAM and Boards. Well, the failure rate is probably the same, but these problems might never surface on your average 2% utilization server.
My experience with drives hasn't mirrored yours. Maybe 7200.11 are bad quality drives, or maybe you got a bad batch, but I've had just as many SCSI and Fiber Channel disks go bad as SATA disks.
For terms of PSU's, well, sure, many average desktops from Dell or whomever will probably use cheaper components. That doesn't mean all ATX power supplies are bad. The form factor doesn't mean anything, you know. There's some really excellent PSU's available, and Backblaze used two of them in their storage machine.
Yea, but the whole POINT of Windows ME was the ability to use VXD drivers, so people's old hardware would still work. If they didn't allow them, they might as well have just pushed Windows 2000, since WinME looked like Windows 2000, didn't allow you to drop to DOS, etc.
If they left WinME out in the wild for awhile and delayed Windows XP, it would have been more popular, but before it gained any traction at all, they released Windows XP and the rest is history.
In the end, they should have just not released ME, and let Windows 98 stick around until XP. Windows 2000 was a much better OS than WinME, but it lacked in a lot of consumer hardware support. It's amazing what two more years of vendor support can do to a system, and by the time Windows XP was released most vendors finally had NT drivers available so the switch to NT was finally viable. I was actually pretty excited at the prospect of *everyone* using NT finally. And worried at the same time =)
I used Windows 98 for a little while but I mostly ran NT4 and then 2000. I never thought 9x would die..
Totally. I still hate the "dancing" top bar. The idea is that you just smash the mouse up to the top of the screen to click menus, so you don't have to actually make a coordinated effort to click a specific menu item. Maybe that was good for 1985 when the Mouse was still a new thing, but people are pretty good at using them now.
OSX would have been the opportunity to finally use traditional menus (like every other system in the entire world) but they stuck with old MacOS menu.
No, it was never completely fixed. They put in all sorts of work-arounds but memory was still basically unprotected and multitasking was still very Windows 3.1ish.
I hated MacOS with a passion - all of the old MacOS versions would bring me to tears, and not tears of joy. I still don't love it, but 10.4 was pretty good.
Certainly; it's a great idea to have redundant PSU's on critical hardware if available, but I wouldn't call your average beige box server unreliable because of that single commission.
But really, that's the biggest difference between "server" computers and "desktop" ones when it comes to fault tolerance. That, and the aforementioned RAID controller of some sort. Those aren't even present on low end servers..
Sure, a decent server from HP will probably be more reliable, but it's usually because they chose components conservatively. RAM is never clocked to the extreme, it's always ECC, and they use very well tested chipsets. However, you can make a good selection of a commodity board and get all of these things too; you just have to be choosy.
The #1 reason for server downtime in my experience has been RAM or a bad component on a main board. #2 is RAID cards, and a far distant #3 is disk failure.
There's some excellent servers out there that provide very good levels of reliability utilizing things like mirrored/hot plug RAM, hot plug PCI cards, etc. Of course, these systems are extremely expensive; that's why most SAN systems simply use two "servers" (Like a Clariion; it's just two linux boxes) and customized software/hardware to facilitate seamless failover.
So no, a Backblaze type of system isn't ideal for mission critical storage but it provides a similar level of redundancy to most servers while being orders of magnitude cheaper.
It irks the shit out of me when I see people over and over saying desktop hard drives are crap and servers are somehow better. These people don't realize that it's all the same stuff. It's like a Macintosh versus a normal PC. The only difference is a 512Kbit BIOS update.
PCI isn't that slow, and it seems as though their main concern is more about access time versus transfer rates. In fact, most enterprise computing looks at disks from an IOPS perspective and you don't even need to worry about transfer.
So, your average PCI bus won't be a bottleneck here, the disks will, and even if we were using PCIe x8 instead of PCI, it's unlikely the IOPS performance would be any different.
They put 15 disks in each array. Having that many spindles in the array will increase IOPS performance greatly. And, you underestimate Linux RAID performance.
What IS "Server-Grade" equipment, anyways? Because for the thousands of servers I've touched in the last decade, the only differences seem to be multiple power supplies and RAID cards. Many servers have dual-CPU sockets, sometimes four.
So really, the only real difference here is that they didn't put in *redundant* power supplies. Big whoop. Last time I had a PSU on a server go bad was 2001.
PSU A powers the mainboard. And? You do realize that these hard drives don't actually use much power and they've used 750W supplies?
And again with the unsubstantiated claim that desktop hard drives are not as reliable as "enterprise" hard drives, and you even used MTBF. MTBF is more about warranty than anything else. The manufacturer wants to sell you the much more expensive "server" drives. They're made on the same assembly lines with most of the same parts. They weigh reliability against performance - Desktop drives usually have more space and slower rotational speeds, server drives spin faster with less space.
Silent data corruption isn't a problem with Linux RAID 6. It will routinely "scrub" the disk data to make sure there's no parity errors. This effectively prevents this problem. All decent RAID cards do this, and so does Linux RAID. If parity errors are found the disk goes offline and you're alerted. Crisis averted.
Please, stop reading from a text book and get some experience in the field.
Woah woah woah. Hold on there. What do you mean, frequent failures because of "consumer grade" hardware?
That's utter bullshit and you should know that. Desktop hard drives have not been proven to be any less or more reliable than "enterprise" disks. Ever. At my last job, we used a shit ton of desktop hard drives in our data center and the failure rate was no different. Sure, a few hundred disks isn't a great sampling but nobody else has proven it either.
Besides, the BackBlaze is set up into four sets of RAID 6 on ZFS. A lot of disks would have to fail, here. And the redundancy is NO different than any server you'd find from ANY manufacturer of comodity PC's.
He's not asking for mainframe class hot-site backups here (although we're not exactly sure what he's asking for, but that would have been something to mention) he's apparently looking for a cheap place to store a bunch of terrabytes in case the building burns down.
No, you're not wrong. And I don't even think the bearings are any different. Enterprise class hard drives generally spin faster, but they are almost always lower capacity. So, that's the trade-off - space or speed. It's always been that way. Because manufacturing hard drives has improved so much in recent years, there's less of a gap, but you still can't get a 10K 1.5TB disk. I think Hitachi is the only company making a 10K 1TB disk?
The MTBF is more about warranty than quality of the product. I guess some people think desktop hard drives come from a "less clean" room assembly plant?
Once the GPU can do general purpose computing (read: Everything) then we can talk about a paltry 3x improvement. The GPU is a specialized processor and therefore can do that one thing very efficiently.
They could make cars extremely fuel efficient if they they only didn't have to turn, accelerate and brake, and carry passengers. But they do, and the CPU also has to be able to process all kinds of math.
Just like how some Oil exec last year said we'd see $1 gas at the pump this year, we don't see a nearly six hundred fold increase in GPU peformance in the next six.
The computing world is different these days. Any system - including any Linux system and any MacOS system - require reboots on occasion for system updates.
There's a security flaw in the kernel and you need to patch it, and that means reboots.
This has nothing to do with reliability. The system will keep on running if you don't apply the security update, so you can choose to run the update or not. You can design your IT environment so reboots won't cause downtime.
Sure, you have to reboot your Linux system LESS but these days even Solaris is susceptible to the new age of security conscious, immediate updates.
Yea, in the good 'ol days of Netware 4.11, you could leave that system online forever. But now, everything is connected to the Internet and people are trying a lot harder to break in to your systems. A six-month or yearly update schedule is no longer viable for today's IT systems.
I remember running NT4 boxes running Exchange 5.0 or 5.5 that would run for a year with no problems. But they weren't connected to the Internet.
I do see your point, and I've also seen disadvantages of big IT as you have.
The problem, as I see it, isn't that the IT department becomes closed off or the people arrogant about it. It's that many companies don't scale the IT staff as much as they should. Small groups of IT engineers are expected to handle an ever-growing number of users, and you simply can't accommodate everyone's personal desires. You have to do everything you can to centrally manage as much as possible, so you can provide the best user experience that you can.
Even with a properly staffed IT group though, if you allow users to have everything they want you will be left with an unimaginable mess and it can happen very rapidly.
I'm in the middle, really. Some IT people assume all users are dipshits. I don't. I give people more credit, and I've been successful in this approach then the rigid, "lock everything down to shit" approach. I have happier users and therefore an easier time getting the big things done.
It really doesn't matter if Windows 7 is a lot like Vista. You can't assume anything, so you have to re-test, re-configure and get new approvals all they down the line. New images, new deployment strategy.. everything from ground zero, except perhaps hardware compatibility. But, there's no guarantee that just because a hardware item works on Vista it will work on W7 so you have to test that again too.
Windows 7 is a lot like Vista, I agree, but the interface (taskbar) is quite a bit different so there will be even more re-training issues. Vista looks and behaves mostly like XP from an end-user perspective, so it's easier on the non-computer savvy users. (Don't get me wrong, it's not an excuse or anything; people will learn to live with the W7 UI changes eventually.)
Microsoft doesn't say skip Vista. What Gartner says is if you haven't even started planning a Vista upgrade, it might make sense at this point to move forward with Windows 7. It makes NO sense do so if you're already in the process of moving on Vista.
Slashdot Mirror
That's not actually wrong, depending on the database.
Filesystem snapshot backups work the same way. It's not the same as powering off/on a machine; you don't have to contend with filesystem corruption which is the leading cause of DB corruption - not sudden crashes.
They call this "Crash-consistent" backups.
A modern database engine is a transactional one. Transactions (any change to the database) are first written to a transaction log. The log is periodically written to the database. The log is then checkpointed to indicate that the transaction has been successfully written to the database.
If the database engine should crash or if the machine should crash and restart, the database engine will run a recovery because the database was not shut down cleanly. The recovery checks the transaction log to make sure all of the transactions have been written to the database. If there was a transaction being written to the database when the crash occurred, that transaction will be scrubbed and replayed from the log. If a transaction was being written to the log, that transaction is cancelled.
So, the worst case scenario is that you lose a couple transactions that were JUST being written to the log.
The likelyhood of a database being corrupted due to this process is very, very low. So, for most normal databases, this method is actually acceptable. For something very critical (such as order processing databases or something) you'll want to make sure you initiate some sort of database aware quiescence of the database file before taking the snapshot. This will tell the database engine to write all current transactions to disk and stop writing new ones for a moment so we can get our snapshot. This still leaves us with a crash-consistent copy of the database, but it's nearly guaranteed to be consistent when we start up.
So there you have it. Your engineer wasn't wrong.
Do you actually think the Sidekick server will be that big?
For one, it's on Windows. Danger is owned by Microsoft, and these servers are housed at a Microsoft data center. For two, you really don't need huge hardware anymore.. just redundant hardware.
A single dual-socket quad-core box with enough RAM could probably handle a million Sidekicks, if not more. There's such little data and it's requested over a cell phone data network. Not to mention they don't synchronize everything all the time. Just little changes.
Woah, Millennium?
Here I was thinking that computers have only been around for 50 years..
The technology is available to get good, solid backups for anything. They just didn't use it, test it, verify it, etc. And in the case of this, users cannot back up their own data. And what they lost isn't backups.
I used to have one of these things.
The phone is (like someone above pointed out) a local cache of what's on the server side. The live database/back end is what crashed. When you make a change on the phone, it immediately sends that change to the server. You can login to the sidekick web site and make changes there, which appear quickly on your phone. If you reboot your phone, it will retrieve anything it needs from the server side. Apparently, the phone doesn't even keep a permanent local copy on some sort of non-volatile storage (hence "Don't turn off your phone.")
It's like someone that uses Google apps and stores all their documents on their system. If that system should go down, you'd be screwed, except that you COULD back up your documents locally. With this case, you can not.
I don't really like the term "cloud computing." All it means is server storage somewhere on the Internet. Under this term you could call any web site a "Cloud." It's ambiguous at best.
I'll agree with you on Far Cry. I loved that game! Until the monsters came around from every direction and turned it into a cheezy alien shoot'em'up that you couldn't win without cheating - unless you are an FPS master. I'm not bad at shooters, but I always preferred ones where you had to use tactics. I liked Counter Strike (well, Tactical Ops..) because of that. Far Cry was real cool how you'd sneak around and stuff in this ridiculously large, contiguous area.
I actually agree with the article when it comes to ad placement. We see them all around us - on billboards, on the store fronts themselves, on busses, trains.. I do like the satirical ones used in some games but ad placement is required then doing it in a real-world kind of way is okay with me and I think it probably would make a game that's aiming to be realistic even more so. GTA 4 is a good example of where real ad placement would make sense.
So you're sitting in traffic... and all you can hear is a thousand cars making a thousand "personalized" sounds (only $29.99 each, available on iTunes!) all around you.
Fantastic.
Woah, hold on there, you missed the point entirely.
WHEN Windows 2000 was released, it didn't have enough driver support for a mass release. Microsoft would have taken an exorbitant amount of flack for this; probably worse than they did with Vista.
It was a TIMING thing, not a compatibility with XP. XP didn't come out until almost three years later, and by that time, driver support was much better BECAUSE of Windows 2000's availability for that time.
Sheesh.
you really misunderstood the point on the Macintosh. The difference between a PC and a Mac these days is a different BIOS, but obviously I'm not talking about a Hackintosh. Apple takes off the shelf components, customized an EFI BIOS for it, slaps on an Apple sticker and off they go. Sure, they sometimes have custom board designs made to fix their cases, but the components are no different.
If you've never had a RAID card fail then you're lucky, but 50 servers isn't exactly a good sampling. I've worked in data centers with a thousand servers, and I've worked in a few of them. These days, with VMware stressing servers in a way that's probably never been done on commodity hardware, you get a lot more failure out of RAM and Boards. Well, the failure rate is probably the same, but these problems might never surface on your average 2% utilization server.
My experience with drives hasn't mirrored yours. Maybe 7200.11 are bad quality drives, or maybe you got a bad batch, but I've had just as many SCSI and Fiber Channel disks go bad as SATA disks.
For terms of PSU's, well, sure, many average desktops from Dell or whomever will probably use cheaper components. That doesn't mean all ATX power supplies are bad. The form factor doesn't mean anything, you know. There's some really excellent PSU's available, and Backblaze used two of them in their storage machine.
Same way you pronounce Noonian Soong.
2002 called. It wants its joke back.
With any modern dual-core CPU it takes so little time to compile things these days it's not even an issue anymore..
I'm certain you can find an internal Wide SCSI disk out there for pennies.
Yea, but the whole POINT of Windows ME was the ability to use VXD drivers, so people's old hardware would still work. If they didn't allow them, they might as well have just pushed Windows 2000, since WinME looked like Windows 2000, didn't allow you to drop to DOS, etc.
If they left WinME out in the wild for awhile and delayed Windows XP, it would have been more popular, but before it gained any traction at all, they released Windows XP and the rest is history.
In the end, they should have just not released ME, and let Windows 98 stick around until XP. Windows 2000 was a much better OS than WinME, but it lacked in a lot of consumer hardware support. It's amazing what two more years of vendor support can do to a system, and by the time Windows XP was released most vendors finally had NT drivers available so the switch to NT was finally viable. I was actually pretty excited at the prospect of *everyone* using NT finally. And worried at the same time =)
I used Windows 98 for a little while but I mostly ran NT4 and then 2000. I never thought 9x would die..
Totally. I still hate the "dancing" top bar. The idea is that you just smash the mouse up to the top of the screen to click menus, so you don't have to actually make a coordinated effort to click a specific menu item. Maybe that was good for 1985 when the Mouse was still a new thing, but people are pretty good at using them now.
OSX would have been the opportunity to finally use traditional menus (like every other system in the entire world) but they stuck with old MacOS menu.
It still annoys me there's no click through.
No, it was never completely fixed. They put in all sorts of work-arounds but memory was still basically unprotected and multitasking was still very Windows 3.1ish.
I hated MacOS with a passion - all of the old MacOS versions would bring me to tears, and not tears of joy. I still don't love it, but 10.4 was pretty good.
Certainly; it's a great idea to have redundant PSU's on critical hardware if available, but I wouldn't call your average beige box server unreliable because of that single commission.
But really, that's the biggest difference between "server" computers and "desktop" ones when it comes to fault tolerance. That, and the aforementioned RAID controller of some sort. Those aren't even present on low end servers..
Sure, a decent server from HP will probably be more reliable, but it's usually because they chose components conservatively. RAM is never clocked to the extreme, it's always ECC, and they use very well tested chipsets. However, you can make a good selection of a commodity board and get all of these things too; you just have to be choosy.
The #1 reason for server downtime in my experience has been RAM or a bad component on a main board. #2 is RAID cards, and a far distant #3 is disk failure.
There's some excellent servers out there that provide very good levels of reliability utilizing things like mirrored/hot plug RAM, hot plug PCI cards, etc. Of course, these systems are extremely expensive; that's why most SAN systems simply use two "servers" (Like a Clariion; it's just two linux boxes) and customized software/hardware to facilitate seamless failover.
So no, a Backblaze type of system isn't ideal for mission critical storage but it provides a similar level of redundancy to most servers while being orders of magnitude cheaper.
It irks the shit out of me when I see people over and over saying desktop hard drives are crap and servers are somehow better. These people don't realize that it's all the same stuff. It's like a Macintosh versus a normal PC. The only difference is a 512Kbit BIOS update.
Actually, it's probably not too bad.
PCI isn't that slow, and it seems as though their main concern is more about access time versus transfer rates. In fact, most enterprise computing looks at disks from an IOPS perspective and you don't even need to worry about transfer.
So, your average PCI bus won't be a bottleneck here, the disks will, and even if we were using PCIe x8 instead of PCI, it's unlikely the IOPS performance would be any different.
They put 15 disks in each array. Having that many spindles in the array will increase IOPS performance greatly. And, you underestimate Linux RAID performance.
Okay, there's many problems with your post.
What IS "Server-Grade" equipment, anyways? Because for the thousands of servers I've touched in the last decade, the only differences seem to be multiple power supplies and RAID cards. Many servers have dual-CPU sockets, sometimes four.
So really, the only real difference here is that they didn't put in *redundant* power supplies. Big whoop. Last time I had a PSU on a server go bad was 2001.
PSU A powers the mainboard. And? You do realize that these hard drives don't actually use much power and they've used 750W supplies?
And again with the unsubstantiated claim that desktop hard drives are not as reliable as "enterprise" hard drives, and you even used MTBF. MTBF is more about warranty than anything else. The manufacturer wants to sell you the much more expensive "server" drives. They're made on the same assembly lines with most of the same parts. They weigh reliability against performance - Desktop drives usually have more space and slower rotational speeds, server drives spin faster with less space.
Silent data corruption isn't a problem with Linux RAID 6. It will routinely "scrub" the disk data to make sure there's no parity errors. This effectively prevents this problem. All decent RAID cards do this, and so does Linux RAID. If parity errors are found the disk goes offline and you're alerted. Crisis averted.
Please, stop reading from a text book and get some experience in the field.
Woah woah woah. Hold on there. What do you mean, frequent failures because of "consumer grade" hardware?
That's utter bullshit and you should know that. Desktop hard drives have not been proven to be any less or more reliable than "enterprise" disks. Ever. At my last job, we used a shit ton of desktop hard drives in our data center and the failure rate was no different. Sure, a few hundred disks isn't a great sampling but nobody else has proven it either.
Besides, the BackBlaze is set up into four sets of RAID 6 on ZFS. A lot of disks would have to fail, here. And the redundancy is NO different than any server you'd find from ANY manufacturer of comodity PC's.
He's not asking for mainframe class hot-site backups here (although we're not exactly sure what he's asking for, but that would have been something to mention) he's apparently looking for a cheap place to store a bunch of terrabytes in case the building burns down.
No, you're not wrong. And I don't even think the bearings are any different. Enterprise class hard drives generally spin faster, but they are almost always lower capacity. So, that's the trade-off - space or speed. It's always been that way. Because manufacturing hard drives has improved so much in recent years, there's less of a gap, but you still can't get a 10K 1.5TB disk. I think Hitachi is the only company making a 10K 1TB disk?
The MTBF is more about warranty than quality of the product. I guess some people think desktop hard drives come from a "less clean" room assembly plant?
The only difference between a four and six digit UID was about 8 months.
Hmm, hopefully the "undisclosed" company isn't Chevron or Mobile Oil.
Once the GPU can do general purpose computing (read: Everything) then we can talk about a paltry 3x improvement. The GPU is a specialized processor and therefore can do that one thing very efficiently.
They could make cars extremely fuel efficient if they they only didn't have to turn, accelerate and brake, and carry passengers. But they do, and the CPU also has to be able to process all kinds of math.
Just like how some Oil exec last year said we'd see $1 gas at the pump this year, we don't see a nearly six hundred fold increase in GPU peformance in the next six.
The computing world is different these days. Any system - including any Linux system and any MacOS system - require reboots on occasion for system updates.
There's a security flaw in the kernel and you need to patch it, and that means reboots.
This has nothing to do with reliability. The system will keep on running if you don't apply the security update, so you can choose to run the update or not. You can design your IT environment so reboots won't cause downtime.
Sure, you have to reboot your Linux system LESS but these days even Solaris is susceptible to the new age of security conscious, immediate updates.
Yea, in the good 'ol days of Netware 4.11, you could leave that system online forever. But now, everything is connected to the Internet and people are trying a lot harder to break in to your systems. A six-month or yearly update schedule is no longer viable for today's IT systems.
I remember running NT4 boxes running Exchange 5.0 or 5.5 that would run for a year with no problems. But they weren't connected to the Internet.
Things are different now.
I do see your point, and I've also seen disadvantages of big IT as you have.
The problem, as I see it, isn't that the IT department becomes closed off or the people arrogant about it. It's that many companies don't scale the IT staff as much as they should. Small groups of IT engineers are expected to handle an ever-growing number of users, and you simply can't accommodate everyone's personal desires. You have to do everything you can to centrally manage as much as possible, so you can provide the best user experience that you can.
Even with a properly staffed IT group though, if you allow users to have everything they want you will be left with an unimaginable mess and it can happen very rapidly.
I'm in the middle, really. Some IT people assume all users are dipshits. I don't. I give people more credit, and I've been successful in this approach then the rigid, "lock everything down to shit" approach. I have happier users and therefore an easier time getting the big things done.
It really doesn't matter if Windows 7 is a lot like Vista. You can't assume anything, so you have to re-test, re-configure and get new approvals all they down the line. New images, new deployment strategy.. everything from ground zero, except perhaps hardware compatibility. But, there's no guarantee that just because a hardware item works on Vista it will work on W7 so you have to test that again too.
Windows 7 is a lot like Vista, I agree, but the interface (taskbar) is quite a bit different so there will be even more re-training issues. Vista looks and behaves mostly like XP from an end-user perspective, so it's easier on the non-computer savvy users. (Don't get me wrong, it's not an excuse or anything; people will learn to live with the W7 UI changes eventually.)
Microsoft doesn't say skip Vista. What Gartner says is if you haven't even started planning a Vista upgrade, it might make sense at this point to move forward with Windows 7. It makes NO sense do so if you're already in the process of moving on Vista.