But it's not a single failure mode. Indeed, any time the X server crashes, every open X app is lost.

Right, and every time Microsoft Windows crashes you lose all your applications as well.

Sorry, but Vista does not solve that problem, I don't have any experience with windows 7 so I can't comment on that.

All I know is that even if I do something simple on a windows box it requires a reboot. Which pretty much happens every couple of days when you really work a lot with windows, which will reset the counter but will not count as a 'crash'.

Yeah, but I didn't say when Linux crashes, I said when the X server crashes. When the equivalent happens in Windows, well, if you read the article, you know what happens.

For those rare cases when a windows display driver crash will not take the whole machine with it (usually by freezing or bluescreening) you might get out of it that way.

But for the majority of those cases you'll have very little choice but to reach for the power button and keep it pressed for 10 very long seconds.

Again, this is anecdotal, but I have never seen windows recover from any kind of crash in a graceful way.

There are several layers to this stuff, hardware, driver, display server, window manager + applications.

If any layer 'below' your application crashes you're lucky to keep your data (whatever the operating system). Now, I don't do the things the author refers to (switching screen modes to full screen and back and stuff like that), which might easily be why I never see these issues and for him they are common.

But to suggest that in windows (where the lines between the layers are much more blurry than they should be) is 'crash proof' in any way shape or measure has me chuckling a bit.

My linux machines (servers & desktops) go down when:

- the power fails

- I have to move

- I upgrade kernels

- hardware failures (mostly disks, occasionally a power supply or a memory bank)

Other than that they're on, up & running 24x7 year after year.

My graphics driver crashed on Vista just yesterday. The screen flashed and I got a nice pop-up in the bottom right corner telling me that the display driver just had to be restarted.

How many machines are you running Vista on, anyway?

Now, I have lots of Linux crash stories. But I run a lot more Linux servers than you do, so of course I would.

I'm happy to hear you run lots more linux servers than I do, as I said before, it is all anecdotal evidence, and so should be taken with a lump of salt.

We have exactly one machine running vista here, it's fairly new and it has crashed more than all our linux machines combined.

So, just out of curiosity, how many linux machines do you have ?

A few dozen.

I won't tell you guys how to troubleshoot your crashing Vista machine, but the OS isn't at all the first place I'd look to find the problem.

Most Vista boxes don't ever crash. Mine certainly don't. On the server side of things (Windows Server versus RHEL, say) my impression is that Microsoft outcompetes Redhat on stability. Just not on cost.

You should have checked my profile I guess before you made your claim ;)

Anyway, a few dozen is pretty good, more than most people have.

I figured you were owner of a large outfit that runs thousands of machines.

That's when you start to get interesting statistics.

Hmm. I just have a lot of trouble believing that you run lots and lots of machines and yet have never seen a kernel crash.

Well, maybe we're just lucky but I don't think that is it.

We strip the kernels down to the minimum we need on machine installation and we always try to find out what caused a problem.

On top of that a buddy of mine wrote a piece of software he calls 'flame', which is used to manage 300 machines.

Sure, occasionally one of those will go down. But post-mortem analysis will invariably point to some piece of hardware that broke. Either a cpu, a memory chip or a drive hanging on by a hair. I think the trick to managing large numbers of servers and not being in reboot hell all the time is to find a recipe and to stick to it, and good quality monitoring tools to give you an early warning.

And if you are running on cheap non-ecc motherboards for servers that really is asking for it.

I like to sleep well :)

The last crash that I remember was in February this year, a faulty memory chip that had been registering complaints for a while in the mcelog (but due to a human error we missed it). We had just upgraded a machine that was already in production and we did not burn in the machine out of service. Stupid mistake...

We burn in our new machines religiously and we only use top quality hardware. Still, during the burn-in phase we have a good 10% hardware failures.

But once the machines pass burn in - and we really give them hell - they perform pretty good.

Since we're in anecdotal territory anyway, once during the burn in testing of an ethernet switch over the weekend we came back to the office on monday morning to find the test frozen.

The switch was no longer passing data in or out. A reboot would not revive it.

We opened it up to find a crack running straight down the middle of one of the chips... so much for that particular brand of router :)

Where the heck would you look for troubleshooting a Vista machine if not the OS? Are you saying it's probably a hardware problem, or are you one of these people that thinks when an application takes down the whole system it's not a sign there's anything wrong with the OS?

I'd look at Hardware -> Drivers -> Anti-Virus -> Malware.

In that order. Then I'd start poking around for OS issues.

Other than hardware, none of those problem sources is any excuse for the whole system to crash -- but people still behave like they are on MS Windows.

I hear your penis is huge also?

Your wife exaggerates.

i don't get what the eff windows has to do with X server crashes taking down X apps? "Windows does it too" IS NOT an acceptable solution.

True. But the point was that X doesn't do it in my experience, maybe I'm unique, maybe not you'll need more numbers. But the ONE windows box we have here has the same problem. So I could make the exact opposite case ("windows crashes, X does not, windows should learn from 'X'") and it would still just be anecdotal evidence.

The author suggests that windows has 'the solution' for these kinds of crashes, that is why saying but windows does it too is relevant, it somewhat negates his assertion.

If windows would really be rock solid I think I would have a different experience, and - sadly - I'm not alone in that.

What the underlying reasons are is anybodies guess, but by bitching about it it is not going to get solved.

His parting shot is:

"I'm sure the blame-game will be played thoroughly in the comments, but it will only detract from the real problem here. The Linux desktop needs a modern, robust, and advanced graphics stack, which makes sure that crashes and bugs remain isolated, without them affecting the users' work. Microsoft has shown us how it's done, now all the X world needs to do is follow."

Whereas if he would simply contact the maintainers of the code with a reproducible error and a stacktrace I'm sure they'd be happy to help.

A comment—complaining about an article that complains about a specific X in Y, and compares it to a better X in Z—suggesting that the poster use Z instead of Y. (I have a theory that this template covers about 10% of IT-related blog post replies by weight.) I have to wonder, haven't you ever used something that's better in almost every way, but worse in one important one? You don't spend time writing a blog post about Y's failures unless you love Y, and want it to get better (or are getting paid by Z, but that's only applicable to very well-known bloggers.)

That's only relevant to the discussion at hand if the person in question actually seems to believe that Y is better than Z in all ways except one. Instead, he appears to believe that Z is better than Y in all ways, period, and uses Y just to have reasons to complain.