Linux Kernel Boot Statistics: 2.6.24 To 2.6.39

[cyan]

200€ is the price of windows ;']
10 000 computers running linux ? are you sure there are so many around the world ?

Well we have around 12000 which we power down overnight and wake up with WOL packets precisely to save power, and it's not a trivial amount.

I did consider renting out the compute power instead during off-peak times, which would probably result in more money, but it's not really core business for a bank

I believe boot times are really important, but not for any silly reasons of productivity (the difference is tiny), but for reasons of user attitudes, especially new users. When the computer is booting, users are typically sitting watching it, not able to do anything. It's the watched-pot effect. Particularly for new users, who are fully focussed on the system and are consciously forming their impressions of the software, it's a big thing. We've found that it is critically important to manage user's impressions of the system when doing a large change like a desktop operating system replacement - a single hostile user in a branch (easy to have - users hate change) can cause the entire branch to actively look for reasons to fault the system. Winning them back from that state is hard

[dnebdal]

One thing that is slightly odd:
If you look at the boot time (seconds) and the bandwidth (MB/sec), you should be able to multiply them and get the amount of data read during the boot process.

Plot that, and you get this:

It's fairly obvious that the time correlates well with the amount of data read, as opposed to with the version number in itself.

That can be shown more clearly:


I'm not sure what's going on on the slower end of the Core2 graph there, but basically the boot time has been the same with varying amounts of data - maybe it's stuck waiting for something else?

[dnebdal]

Oh, and thinking about it, the amount of data read through the different kernel versions is also kind of interesting:


(Damn you, 1 min edit limit.)

[Jimmy]

In this day and age having to do a full boot should become a rarity not common practice.

I agree full-heartedly. However, until suspend/hibernate becomes reliable, it's just not an option. It's nice to have a backup option (fast booting) even if it's not as good.

Look at Windows... boot times are horrible in comparison but suspend and hibernate work nearly flawlessly on everything. If suspend doesn't work on windows you really take a hit. Also Windows has a long history of being reboot happy when int comes to updating software.

Even when you can avoid frequent rebooting, it still happens. I'd rather it not be painful like Windows. That's why we should care about boot times. So even when you're facing the necessity of a reboot it's a slight inconvenience rather than a full on go-brew-coffee break.

[dnebdal]

Win7 actually boots very quickly on my laptop. I'm not sure if it beats the Ubuntu install, but both are completely undramatic to reboot.

(The desktop takes ages, though - I suspect it might be because it's got a slow HD, instead of the SSD in the laptop.)