Mostly just out of curiosity I've been meaning to do this very thing and now I won't have to! Thanks for putting that together.

Thanks for the numbers! It's nice to have some kind of reference.

All the results are horrible. I have a Windows XP here booting in 11 seconds flat (to logon screen, hdd stopped working completely, services up) on 2.8G P4 with one Raptor as HDD and 1 gigabyte of DDR400. And it's 3 years old OS installation. (Btw, it has been on and networked without any firewalls and anti-virus applications on a solid copper connection for the whole time. Not one single problem there either. It's rock solid and insanely secure.)

A second insight: There is no one single technical reason why the booting of Linux distribution should take more than 3 seconds from the very beginning to the GDM screen and being stable. It's all just lack of integration, bad architectures, starting unnecessary (at bootup time) services, and pain retardness of the developers. All it would take is a little bit of that thinking outside of the box, but I understand it is not very much practiced in the floss world.

Actually I once made a detailed technical plan about it just to evaluate how plausible the super fast boot process would be to implement. It was very plausible, something roughly 5 good developers could achieve within month.

Then find and persuade a group of people to join you and make it happen. That's one reason why FOSS is good, if you're not happy with something you can fix it yourself.

First, test bootup time on your own system. Secondly, lightweight distros tend to boot quickly. You can't call Linux slow when there is a port (ELKS/linux-8086) to the 8086! And go and suggest your plan somewhere. Not all developers are dismissive.

If you are going to make up numbers, atleast try to keep them 'somewhat' reasonable. A Brand new install of Windows XP on a quad-core system with 6 gigs of RAM will be 'lucky' to boot in under a minute.

Clearly you have only used crufty broken systems, or you run antivirus crap or something. XP Boot times well under a minute are perfectly reasonable on hardware *of the day*. On *current* hardware, you have to *actively* try to fail.

Starting without services is not really practical, but there is an alternative.

I usually delay starting all network based services until after GDM is started (or a user logs in, if only using network manager). This saves a considerable amount of time.

I believe Windows does something similar. By prioritizing the services needed to get to the desktop, and starting the remainder a few seconds later, the startup time impacts the time to a usable desktop less (even though the total time may be the same).

Yeah,
I can see a difference between linux boot and windows.
windows boot all necessary thing to get to the login screen then...load everything else.
linux boot everything to login screen.

When you log on windows, it might take like 1 or 2 min to get a usable desktop. On linux,its like 2-5 seconds.

One thing that could be done on linux, is too load everything needed for gdm/kdm/xdm to be usable (network or not if you have a thin client) then the boot sequence continue while you're in the login screen.
Seriously you don't really have to wait for mysql and apache to start or most of the service to get to gdm. Most device detection could be done while or after login.

Bah, yet another person does a nicer job than me on this : ). More bootcharts can be seen at the bottom of https://wiki.ubuntu.com/BootCharting . One thing you might want to add is the time it take to get to the applet clock when autologin is enabled and you are booting from cold. Why the clock? Because quite often until it appears the system will not respond to mouse clicks (it always seems to appear just before responsive occurs). Another thing that is interesting is just how quickly you can shutdown from a started system.

The one thing that sped up the Ubuntu boot speed the most was doing a profile boot. You will never get the time that it takes to do a profile back again (unless you are booting 5 times a day every day for a year or something) but it caused the biggest reduction in boot up time. If only there was a way to do profiling that wasn't so slow... (perhaps http://www.mimuw.edu.pl/[…]/application.html will happen).

It may be worth redoing the bootcharts with Fedora 9 final as my results certainly changed between the beta (which was using a debug kernel) and the final release: https://wiki.ubuntu.com/Boo[…]90a9717ad0c72c9b400c236a581

Harald, if you still have the same hardware, it would be interesting to see the same results for F10, F10 + updates, and F11beta2, to show how reducing boot time is progressing.

You should really test Mandriva - especially 2009 Spring with speedboot enabled - as it's rather ahead of the curve when it comes to boot optimization in general-purpose distributions. In configurations where speedboot works it will be substantially faster than anything else.

http://blog.crozat.net/2009/02/speedboot-explained.html