[node-devel] about boot time of ovirt-node

Thu Aug 22 10:03:39 UTC 2013

Am Mittwoch, den 21.08.2013, 08:45 +0000 schrieb Bohai (ricky):
> Hi, Fabian
> 
> Thanks for your answer and your time.
> 
> I try to get the log as you said.
> The below is the summary of journalctl's log .
> Maybe the NIC device and NIC device's process are the reason why initrd's slow.
> About the detail, please reference the attachments.

Hey Ricky,

the journctl.log gave some insight. It looks like there is much time
spend doing all sorts of stuff with udev.

> -------------------------------summary about journalctl-----------------------------------
> Aug 21 08:14:47 localhost systemd-udevd[344]: '/sbin/initqueue --settled --onetime --unique /sbin/dmsquash-live-root /dev/dm-3'(err) '/sbin/initqueue at 51(): echo '/sbin/dmsquash-live-root /dev/dm-3''
> Aug 21 08:14:59 localhost systemd-journal[196]: Suppressed 402 messages from /system/systemd-udevd.service
> Aug 21 08:14:59 localhost systemd-udevd[302]: validate module index
> Aug 21 08:15:02 localhost systemd-udevd[302]: validate module index
> Aug 21 08:15:05 localhost systemd-udevd[302]: validate module index
> Aug 21 08:15:08 localhost systemd-udevd[302]: worker [342] /devices/pci0000:00/0000:00:03.0/0000:02:00.1/net/eth0 timeout; kill it
> Aug 21 08:15:08 localhost systemd-udevd[302]: seq 2637 '/devices/pci0000:00/0000:00:03.0/0000:02:00.1/net/eth0' killed
> ------------------------------------------------------------------------------------------------------

And yes, this snippet really shows that systemd/udev is waiting for
outstanding stuff from devices.
I don't fully understand the cmdline, but the presence of --settled
indicates that systemd is blocking until all outstandings bits poured
in. So them is actually spend on initializing hardware.

The udev part should be much quicker in the VM logs.

Greetings
fabian

> 
> 
> 
> > -----Original Message-----
> > From: Fabian Deutsch [mailto:fabiand at redhat.com]
> > Sent: Friday, August 16, 2013 5:18 PM
> > To: Bohai (ricky)
> > Cc: Luohao (A); boh.ricky at gmail.com; Haofeng; node-devel at ovirt.org
> > Subject: Re: [node-devel] about boot time of ovirt-node
> > 
> > Am Freitag, den 16.08.2013, 11:16 +0200 schrieb Fabian Deutsch:
> > > Am Freitag, den 16.08.2013, 07:58 +0000 schrieb Bohai (ricky):
> > > > Hi Fabian,
> > > >
> > > > As your hint , I use the systemd-analyze to get the machine boot time as
> > below:
> > > >
> > > > real machine:
> > > > Startup finished in 1.657s (kernel) + 34.576s (initrd) + 38.054s (userspace) =
> > 1min 14.288s
> > > > virtual machine:
> > > > Startup finished in 1.382s (kernel) + 2.624s (initrd) + 37.547s (userspace) =
> > 41.554s
> > > >
> > > > The main difference is about initrd's load time.
> > >
> > > You can add rd.debug and/or rd.udev.debug to the cmdline when booting.
> > > In that case dracut will log to either journald (if systemd is used
> > > within dracut) or to /run/initramfs/init.log.
> > 
> > journalctl -ab can be used to retrieve the logged data
> > 
> > - fabian
> > 
> > > I'm curious what dracut is taking so long ..
> > >
> > > - fabian
> > >
> > > > And i found a similar problem's discuss post in the opensuse community.
> > > > Link:
> > > > http://lists.opensuse.org/opensuse-factory/2013-03/msg00177.html
> > > >
> > > > Summary from the post:
> > > > "
> > > > With grub2, it loads in 2-3 seconds on an HP dv2815 laptop with a modest
> > > > AMD Turion CPU at 2 GHz.
> > > >
> > > > The difference might be grub2, or it might be due to poor performance
> > > > for an i7 running in 8686-compatibility mode. I doubt that Intel spent
> > > > much time optimizing those operations.
> > > > "
> > > >
> > > >
> > > > > -----Original Message-----
> > > > > From: Fabian Deutsch [mailto:fabiand at redhat.com]
> > > > > Sent: Thursday, August 15, 2013 5:23 PM
> > > > > To: Bohai (ricky)
> > > > > Cc: node-devel at ovirt.org; Luohao (A); Haofeng; boh.ricky at gmail.com
> > > > > Subject: Re: [node-devel] about boot time of ovirt-node
> > > > >
> > > > > Am Donnerstag, den 15.08.2013, 07:13 +0000 schrieb Bohai (ricky):
> > > > > > Hi all,
> > > > > >
> > > > > > In the use of ovirt-node , I found a strange thing.
> > > > > > Ovirt-node system boots more faster in a virtual machine than a real
> > > > > machine.
> > > > > > (Because of no enough real machine, sometimes I use virtual machine
> > > > > > to test the setup and install function. And the real machine is absolutely
> > more
> > > > > power
> > > > > >  than the real machine.)
> > > > > >
> > > > > > I don't to know why and how to find the reason about it.
> > > > > > Or whether it's a problem of ovirt-node?
> > > > >
> > > > > I suppose you mean the OS part of the boot process? So not including
> > > > > POST and other bios related stuff.
> > > > > On the Fedora based node you could run a systemd-analyze to find out
> > > > > what service is taking how much time.
> > > > >
> > > > > Greetings
> > > > > fabian
> > > > >
> > > > > > Any help or hint is welcome.
> > > > > >
> > > > > > Best regards to you.
> > > > > > Ricky
> > > > > >
> > > > > > _______________________________________________
> > > > > > node-devel mailing list
> > > > > > node-devel at ovirt.org
> > > > > > http://lists.ovirt.org/mailman/listinfo/node-devel
> > > > >
> > > >
> > > > _______________________________________________
> > > > node-devel mailing list
> > > > node-devel at ovirt.org
> > > > http://lists.ovirt.org/mailman/listinfo/node-devel
> > >
> > >
> > > _______________________________________________
> > > node-devel mailing list
> > > node-devel at ovirt.org
> > > http://lists.ovirt.org/mailman/listinfo/node-devel
> > 
> 