Post a Comment
It is an interesting challenge, but meaningless since it is a challenge to individuals and there are too many variables which will impact boot speed.
For example: Fedora takes about two minutes to boot on my machine, but 40 seconds of that is waiting for atalkd to start up. Now I suppose that I could toss atalkd and the daemons which require it into the background, but that would break other parts of the system (albeit, in a very minor way).
While the services which people run will have an impact, so will the amount of tweaking they will do. While I'm sure that some people would love to tweak their "hotrod" computers, individual tweaks for individual systems doesn't help the community at large.
If this results in something like init-ng being nicely packaged for Fedora, that would be a big advantage. On the other hand, it would probably lead to an "init" war, just like we have a DE war. Which side would I fall on: probably the init-ng side. But just because I agree with the change doesn't mean that I think the results would be beneficial to Linux. It already has enough trouble maintaining support for multiple and diverging standards.
Who are those people???? They are using term windows and text-mode browser, emacs or vi as email client and user perception of e.g. syntem performance is none-relevant as a user would just adapt to the particular system (any other way would be "un-mentionable"
In my 7+ years running Linux, I have never met them, seen them, or know anyone who has.
Nevertheless they keep popping up places like this!!!!
Where TH are you and what planet are you coming from??
I've been trying to get initng to work in a way that I could start using it as an all out replacement to SysVinit, but it isn't there yet. But it seems to be close. I was reading on the http://fedoraproject.org/wiki/FCNewInit">redhat for the upcoming FC5 and it looks like a new program called servicemanager is going to make all this happen? There doesn't look like any info regarding this however, so I have no idea when I can try to test something, and if it's actually going to improve boot time, as opposed to simply replace SysVinit with a "better" init system. Unfortunately I'm not at the level to really know whats best for linux, but I guess the discourse regarding this issue is good, and probably long overdue.
Would it be cheating to use InitNG? (http://initng.thinktux.net/index.php/Main_Page) I've played with it a bit, and honestly, it seems like one of the fastest growing projects I've seen in a while... it works *really* well... my home system boots in about 8 seconds, 4 of which is the kernel. It's the easiest way to reduce boot time, and still keep all the services you have.
You see, comparing windows and linux boot times isn't qa good idea, since it's apples and oranges... however your average user will still do that, since boot time is important to them. Now, windows has been innovating over the last few versions, making changes... but linux hasn't. (My latest gentoo install still uses the same version of SysVInit that Redhat 7.3 shipped with... with only a few minor revisions). Enter InitNG. It's really pretty nice, since it loads everything in parallel (more than one at a time), instead of in serial (one at a time)... which makes booting pretty quick. Now, something to note is that what InitNG maxamizes is the HardDisk access time, and (to a lesser extent) the CPU usage. Niether windows, nor SysVInit does this. So, want to make thing boot quicker? Use InitNG!
**For those who don't know, SysVInit is the program the kernel starts that loads all your services... InitNG is a replacement for that (which can be installed along side the orriginal SysVInit.)**
RE: InitNG.... would this be cheating?
[i]If your on a laptop or a desktop or a server, good boot times are great.[i]
I would prefer my server to boot up right rather than fast. After all it shouldn't go down unless you have planned downtime and if it wasn't planned you need to investigate, which probably means further inavailability to users.
On a laptop I think it's better to go for a good working sleep/resume function than a fast boot. Ofcourse sometimes a fast boot would be nice.
When at my desktop I don't much care I'll probably be there a while anyway :-)
Back in the day when I worked on Alpha Micros, I was at a show.
During the demo, the system crashed (this happened reasonably often on Alpha Micros), and without skipping a beat the vendor quipped "But look how fast it boots!" because they had special software to speed up the entire init process.
This is my favorite.
I challenged my systems at home.
PII 266 Mhz laptop took only 30 seconds to boot (18seconds for BIOS) BeOS 5 pro. The same hardware took 4 minutes and 30-35 seconds to boot (autologin enabled) Redhat Enterprise Linux WS 4.0.
PC-BSD 0.7.5 Took 2 minutes and 40 seconds.
Novell Linux Desktop 9.0: took 4 minutes:52 seconds
Mandriva 2005 limited Edition: 4minutes:00 seconds
Ubuntu 5.03 : 3minutes:05 seconds
Xandros 3 SP2: 3minutes: 45 seconds
Windows 95 : 35 seconds
Windows XP Pro + SP2: 4minutes:15seconds (with NIS 2005:because it is a necessity)
It takes my gentoo 2005.0 (amd64) box 20 sec to reach the gdm-login and another 3 sec for kde3.4.1 to be operational.Personally i think this is reasonable satisfactory without initng.I hope gentoo can give the green plus soon to initng,will be nice to see what can be further tweaked.
My Gentoo Box:
From BIOS to Lilo - 8 sec
a) From Lilo to Gdm Login (no init-ng) 25 sec
And another 10 sec to load Gnome
b) From Lilo to Gdm Login (init-ng) 12 sec [!]
And another 10 sec to load Gnome
So... from a cold start to the desktop
without init-ng 8+25+10=43 sec
with init-ng 8+12+10=30 sec
So... the only thing faster on this machine ( amd 800mhz 640ram) would be BeOS ( 20(+-) sec from a coldstart to the desktop)
and yeah. forgot to mention - WinXp From Lilo to the Login ( Themes desabled, minimum services running ) 30 sec and another 30 seconds to a fully functional desktop(AV,Firewall) so WinXP 8+30+30=68sec= 1 min 8 sec.


