| Author |
Message |
|
|
Post subject: Faster bootup
 Posted: Oct 03, 2008 - 08:42 PM
|
|
Joined: Aug 06, 2007
Posts: 11
Status: Offline
|
|
Yes, I know sidux boots up quite fast already  But after reading the article below, it would be nice to see some ideas/stuff brought over to future sidux releases.
LPC: Booting Linux in five seconds
http://lwn.net/Articles/299483/
Quote:
At the Linux Plumbers Conference Thursday, Arjan van de Ven, Linux developer at Intel and author of PowerTOP, and Auke Kok, another Linux developer at Intel's Open Source Technology Center, demonstrated a Linux system booting in five seconds. The hardware was an Asus EEE PC, which has solid-state storage, and the two developers beat the five second mark with two software loads: one modified Fedora and one modified Moblin. They had to hold up the EEE PC for the audience, since the time required to finish booting was less than the time needed for the projector to sync.
|
|
|
| |
|
|
|
 |
|
|
|
Post subject: RE: Faster bootup
Posted: Oct 03, 2008 - 09:33 PM
|
|
Joined: Dec 19, 2006
Posts: 1029
Status: Offline
|
|
|
|
|
|
|
|
|
Post subject: Re: RE: Faster bootup
Posted: Oct 04, 2008 - 07:24 AM
|
|
Joined: Aug 06, 2007
Posts: 11
Status: Offline
|
|
|
kelmo wrote:
Ah, just 1 hour difference 
But this post makes it a feature request then  |
|
|
| |
|
|
|
|
|
|
|
Post subject: RE: Re: RE: Faster bootup
Posted: Oct 05, 2008 - 09:52 AM
|
|
Joined: Dec 19, 2006
Posts: 1029
Status: Offline
|
|
Okay, it is a good feature to request too, and one I am very interested in. So maybe we can discuss a few things here.
We have already delivered on one change to enhance system boot speed, using dash as /bin/sh:
http://sidux.com/PNphpBB2-viewtopic-t-10547.html
The soc2006 project dedicated to improving Debian boot speed highlighted this as a hotspot:
http://initscripts-ng.alioth.debian.org ... harts.html
The Linux Plumber conference inspired nice discussion from a variety of sources, one of the best summaries of how a distribution can deliver some boot time improvements was made by the Mandriva guys:
http://blog.crozat.net/2008/09/improvin ... linux.html
All of the mentioned sources (soc2006, LPC presentation, Mandriva) use a readahead tool of some description. There are quite a few readahead solutions existing in the wild today, all of them with different pros and cons. This is an area of opportunity for our distribution, but some more in depth discussion and calculations need to take place to decide if a readahead solution is worth the effort, and does it deliver really deliver an advantage. I have some further information on this specific topic, but more on that later.
The soc2006 project also discussed the issue of dynamic initscript sequencing to optimise boot/shutdown, we plan to deliver this in pontos release with insserv activated per default:
http://sidux.com/PNphpBB2-viewtopic-t-12724.html
One of the most important factors for maintaining excellent boot speed is management of installed initscripts. Removing software that installs a boot script and is not used is common sense, but quite often not done for a variety of reasons. One reason which comes immediately to mind, is that the end user may not know if a piece of software is critical to the system or not.
The LSB information in almost all initscripts goes a little way to improving this situation:
http://wiki.debian.org/LSBInitScripts
Scripts with the LSB information often contain a description about the purpose of the script/service (however some have really lousy descriptions, or none at all). A really good service manager is lacking on Debian though, in my opinion.
There is also a lack of API for disabling initscripts, which makes it very difficult to manage scripts installed by packages you would like to use, but do not want the package to do anything at boot time. We have made some work at trying to deliver on this area of opportunity:
http://lists.debian.org/debian-devel/20 ... 00623.html
Unfortunately, after proposing that solution to the Debian Developers, there is no clear answer if it is the correct solution, though everyone agrees that a solution is needed (and has been needed for years already). |
|
|
| |
|
|
|
|
|
|
|
Post subject: RE: Re: RE: Faster bootup
Posted: Oct 05, 2008 - 09:53 AM
|
|
Joined: Dec 19, 2006
Posts: 1029
Status: Offline
|
|
| Can anyone help identify other areas of opportunity in the boot process that are realistically deliverable? |
|
|
| |
|
|
|
|
|
|
|
Post subject:
Posted: Oct 05, 2008 - 11:02 AM
|
|
Joined: Aug 06, 2007
Posts: 11
Status: Offline
|
|
| If I understand it correctly, the use of an initrd or initramfs also slows down the booting of the kernel, while not adding a lot of additional "compatibility". Would it not be an option to drop initramfs or make it optional at install? |
|
|
| |
|
|
|
|
|
|
|
Post subject:
Posted: Oct 05, 2008 - 05:34 PM
|
|
Joined: Dec 19, 2006
Posts: 1029
Status: Offline
|
|
You really _must_ use an initramfs if your Linux kernel configuration is modular, which it is unless you hand craft your own image for your specific hardware set (and build in all required support).
The size of the initramfs can be reduced dramatically, and reducing its size might shave up to 1 second from your boot time (roughly), but you lose a lot of benefits. For example, a universal initramfs allows you to make wholesale changes to local hardware configuration without rendering the system unbootable. |
|
|
| |
|
|
|
|
|
|
