On Fri, 6 Aug 2010 13:12:59 -0700 Greg KH <gregkh@suse.de> wrote:
On Fri, Aug 06, 2010 at 09:48:18PM +0200, Stefan Seyfried wrote:
It works very well on servers etc., on everything that does not need (or even want) a fully dynamic and asynchronous boot.
Um, servers want a dynamic and async boot, do they somehow not require very large uptimes which require a quick boot cycle?
Greg, as you certainly know, servers usually take minute(s) from "reboot -f" until the VGA screen (or whatever equivalent) comes on again[1]. Then some more minutes until the memory is found, drives are initialized etc. On such a machine, the 20 seconds you can gain from grub to desktop are not relevant. And oh - there is no desktop anyway. AFAICT most of the "fastboot" efforts are centered around "start X as early as possible" which is totally irrelevant for any serious server machine. All the commercial software installed on those machines usually has some "sleep xxx" inside their init scripts anyway ;-) Those machine simply do not get rebooted. Period.
No wonder the first thing people try to disable as much as possible on servers is udev & friends...
Really? Have you tried to disable udev on a recent system these days? For what good reason?
Not me. I just see what's being done here at the customer (which is probably one of the worlds largest 5 SLES deployments). Of course udev is not disabled, but everyone tries to make sure udev and HAL are not interfering with the system after initial bootup.
It is only the crazy desktop people where it is not going to work. Exactly those people that don't bring any money into the business...
Um, our preload group is profitable, and growing.
I'm glad to hear that (really. It shows that not everything we did was a waste of time and that there is finally some gain from all the pain ;). But I'm still sure that getting rid of all the server subscriptions would be not a wise move for Novell/SUSE ;-)
And note, one of the first users of udev was very large systems with hundreds and thousands of disk drives, way back in SLE9.
But in SLES9 /usr on separate partition was still a supported setup (I know, because I did run the testcases for this setup ;) I can totally understand why Kay wants to get rid of all those "legacy features". It's just that I think it won't work out well at the customers I'm visiting. Maybe having a "not-so-dynamic" boot mode for servers, which can live with /usr being on a separate partition and a "fully dynamic mode" for desktops might be an option. Crazy requirements like "/usr on NFS which is mounted via 3G" would simply be rejected, so that in "server mode" we could say * everything that is needed to mount filesystems must be in / * only old ifconfig method is supported, no NetworkManager Then dbus and all that stuff would be out of question, making /usr not so important anymore. Just braindumping here of course... :) [1] e.g. SUN Fire X4170 or HP ProLiant BL460c G6 machines. Timed with a stopwatch. 1 Minute between "echo b > /proc/sysrq-trigger" and VGA BIOS message. SUN with 24GB RAM, HP with 144GB RAM. Both machines are definitely not crap. -- Stefan Seyfried "Theory and practice sometimes clash. And when that happens, theory loses. Every single time." -- Linus Torvalds -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-factory+help@opensuse.org