Greetings testers,
This week's Test Day will focus on another upcoming Fedora 11 feature: 20 Second Startup [1]. This is a follow-on to a Fedora 10 effort to reduce boot delay [2].
Harald Hoyer, David Kovalsky and Ondrej Hudlicky have created a test procedure and looking to gather and analyze bootchart data. Come prepared with your favorite old-skool hardware setup.
Join #fedora-qa this Thursday, February 19, 2009 to help collect and analyze boot speed data. Follow the action at https://fedoraproject.org/wiki/QA/Test_Days/2009-02-19.
Thanks, James
[1] https://fedoraproject.org/wiki/Features/20SecondStartup [2] https://fedoraproject.org/wiki/Features/30SecondStartup
James Laska wrote:
Greetings testers,
This week's Test Day will focus on another upcoming Fedora 11 feature: 20 Second Startup [1]. This is a follow-on to a Fedora 10 effort to reduce boot delay [2].
Harald Hoyer, David Kovalsky and Ondrej Hudlicky have created a test procedure and looking to gather and analyze bootchart data. Come prepared with your favorite old-skool hardware setup.
Join #fedora-qa this Thursday, February 19, 2009 to help collect and analyze boot speed data. Follow the action at https://fedoraproject.org/wiki/QA/Test_Days/2009-02-19.
Thanks, James
[1] https://fedoraproject.org/wiki/Features/20SecondStartup [2] https://fedoraproject.org/wiki/Features/30SecondStartup
How to generate a bootchart:
- install bootchart # yum install bootchart
note: this modifies your /etc/grub.conf and adds "init=/sbin/bootchartd"
- reboot - login - start a terminal
$ su - # bootchart -f png
produces bootchart.png
On Mon, Feb 16, 2009 at 8:33 AM, Harald Hoyer harald@redhat.com wrote:
How to generate a bootchart:
- install bootchart
# yum install bootchart
note: this modifies your /etc/grub.conf and adds "init=/sbin/bootchartd"
- reboot
- login
- start a terminal
$ su - # bootchart -f png
produces bootchart.png
Does this work if your root is encrypted?
Doesn't seem to work for me.
Appears it is looking for '/sbin/bootchartd' before luksOpen/mount/... I never get prompted for passphrase.
Anyway, end up with "won't boot" kernels in /boot/grub/grub.conf: I have to remove 'init=/sbin/bootchartd' from boot parameters.
BZ'ed here: https://bugzilla.redhat.com/show_bug.cgi?id=485847
tom
On Mon, Feb 16, 2009 at 4:16 PM, Tom London selinux@gmail.com wrote:
On Mon, Feb 16, 2009 at 8:33 AM, Harald Hoyer harald@redhat.com wrote:
How to generate a bootchart:
- install bootchart
# yum install bootchart
note: this modifies your /etc/grub.conf and adds "init=/sbin/bootchartd"
- reboot
- login
- start a terminal
$ su - # bootchart -f png
produces bootchart.png
Does this work if your root is encrypted?
Doesn't seem to work for me.
Appears it is looking for '/sbin/bootchartd' before luksOpen/mount/... I never get prompted for passphrase.
Anyway, end up with "won't boot" kernels in /boot/grub/grub.conf: I have to remove 'init=/sbin/bootchartd' from boot parameters.
BZ'ed here: https://bugzilla.redhat.com/show_bug.cgi?id=485847
This now works!
Looking at the bootchart, appears that system is spending about 5 seconds elapsed time doing a "find".
I think I tracked this down to the following snippet from /etc/rc.d/rc.sysinit:
# Clean up /var. rm -rf /var/lock/cvs/* /var/run/screen/* find /var/lock /var/run ! -type d -exec rm -f {} ; rm -f /var/lib/rpm/__db* &> /dev/null rm -f /var/gdm/.gdmfifo &> /dev/null
The problem is, I seem to have over 325 entries in /var/run/gdm of the form:
[root@tlondon gdm]# ls |wc 326 326 6520 [root@tlondon gdm]#
[root@tlondon gdm]# ls -lt | head total 1304 drwxrwxrwt 2 tbl tbl 4096 2009-02-17 09:02 auth-for-tbl-VJxBYM drwxrwxrwt 2 gdm gdm 4096 2009-02-17 09:01 auth-for-gdm-Lpopiu drwxrwxrwt 2 tbl tbl 4096 2009-02-16 07:39 auth-for-tbl-CAt0DD drwxrwxrwt 2 gdm gdm 4096 2009-02-16 07:39 auth-for-gdm-zEKwWb drwxrwxrwt 2 tbl tbl 4096 2009-02-13 17:25 auth-for-tbl-Hhw23F drwxrwxrwt 2 gdm gdm 4096 2009-02-13 17:25 auth-for-gdm-CsXMMG drwxrwxrwt 2 gdm gdm 4096 2009-02-08 15:27 auth-for-gdm-KVALOw drwxrwxrwt 2 tbl tbl 4096 2009-02-08 15:27 auth-for-tbl-2JXTRC drwxrwxrwt 2 gdm gdm 4096 2009-02-06 12:56 auth-for-gdm-EOu7CD [root@tlondon gdm]#
Oldest entries:
[root@tlondon gdm]# ls -lrt | head total 1304 drwxrwxrwt 2 gdm gdm 4096 2008-10-17 14:11 auth-for-gdm-sA9Si4 drwxrwxrwt 2 tbl tbl 4096 2008-10-20 23:12 auth-for-tbl-0mcMmJ drwxrwxrwt 2 gdm gdm 4096 2008-10-20 23:12 auth-for-gdm-K9AaX6 drwxrwxrwt 2 tbl tbl 4096 2008-10-28 10:11 auth-for-tbl-9DQPFo drwxrwxrwt 2 gdm gdm 4096 2008-10-28 10:11 auth-for-gdm-qW9ehq drwxrwxrwt 2 tbl tbl 4096 2008-10-28 10:30 auth-for-tbl-tyoE78 drwxrwxrwt 2 gdm gdm 4096 2008-10-28 10:30 auth-for-gdm-rAR0ac drwxrwxrwt 2 tbl tbl 4096 2008-10-28 14:15 auth-for-tbl-O6TyNk drwxrwxrwt 2 gdm gdm 4096 2008-10-28 14:15 auth-for-gdm-KscdOd [root@tlondon gdm]#
What should be cleaning these up?
tom
On Tue, Feb 17, 2009 at 10:05 AM, Tom London selinux@gmail.com wrote:
Looking at the bootchart, appears that system is spending about 5 seconds elapsed time doing a "find".
I think I tracked this down to the following snippet from /etc/rc.d/rc.sysinit:
# Clean up /var. rm -rf /var/lock/cvs/* /var/run/screen/* find /var/lock /var/run ! -type d -exec rm -f {} ; rm -f /var/lib/rpm/__db* &> /dev/null rm -f /var/gdm/.gdmfifo &> /dev/null
The problem is, I seem to have over 325 entries in /var/run/gdm of the form:
[root@tlondon gdm]# ls |wc 326 326 6520 [root@tlondon gdm]#
[root@tlondon gdm]# ls -lt | head total 1304 drwxrwxrwt 2 tbl tbl 4096 2009-02-17 09:02 auth-for-tbl-VJxBYM drwxrwxrwt 2 gdm gdm 4096 2009-02-17 09:01 auth-for-gdm-Lpopiu drwxrwxrwt 2 tbl tbl 4096 2009-02-16 07:39 auth-for-tbl-CAt0DD drwxrwxrwt 2 gdm gdm 4096 2009-02-16 07:39 auth-for-gdm-zEKwWb drwxrwxrwt 2 tbl tbl 4096 2009-02-13 17:25 auth-for-tbl-Hhw23F drwxrwxrwt 2 gdm gdm 4096 2009-02-13 17:25 auth-for-gdm-CsXMMG drwxrwxrwt 2 gdm gdm 4096 2009-02-08 15:27 auth-for-gdm-KVALOw drwxrwxrwt 2 tbl tbl 4096 2009-02-08 15:27 auth-for-tbl-2JXTRC drwxrwxrwt 2 gdm gdm 4096 2009-02-06 12:56 auth-for-gdm-EOu7CD [root@tlondon gdm]#
Oldest entries:
[root@tlondon gdm]# ls -lrt | head total 1304 drwxrwxrwt 2 gdm gdm 4096 2008-10-17 14:11 auth-for-gdm-sA9Si4 drwxrwxrwt 2 tbl tbl 4096 2008-10-20 23:12 auth-for-tbl-0mcMmJ drwxrwxrwt 2 gdm gdm 4096 2008-10-20 23:12 auth-for-gdm-K9AaX6 drwxrwxrwt 2 tbl tbl 4096 2008-10-28 10:11 auth-for-tbl-9DQPFo drwxrwxrwt 2 gdm gdm 4096 2008-10-28 10:11 auth-for-gdm-qW9ehq drwxrwxrwt 2 tbl tbl 4096 2008-10-28 10:30 auth-for-tbl-tyoE78 drwxrwxrwt 2 gdm gdm 4096 2008-10-28 10:30 auth-for-gdm-rAR0ac drwxrwxrwt 2 tbl tbl 4096 2008-10-28 14:15 auth-for-tbl-O6TyNk drwxrwxrwt 2 gdm gdm 4096 2008-10-28 14:15 auth-for-gdm-KscdOd [root@tlondon gdm]#
What should be cleaning these up?
Manually cleaning out /var/run/gdm shaves ~4 seconds off my boot.
BZ filed here: https://bugzilla.redhat.com/show_bug.cgi?id=485974
tom
On Mon, 2009-02-16 at 17:33 +0100, Harald Hoyer wrote:
James Laska wrote:
Greetings testers,
This week's Test Day will focus on another upcoming Fedora 11 feature: 20 Second Startup [1]. This is a follow-on to a Fedora 10 effort to reduce boot delay [2].
Harald Hoyer, David Kovalsky and Ondrej Hudlicky have created a test procedure and looking to gather and analyze bootchart data. Come prepared with your favorite old-skool hardware setup.
Join #fedora-qa this Thursday, February 19, 2009 to help collect and analyze boot speed data. Follow the action at https://fedoraproject.org/wiki/QA/Test_Days/2009-02-19.
Thanks, James
[1] https://fedoraproject.org/wiki/Features/20SecondStartup [2] https://fedoraproject.org/wiki/Features/30SecondStartup
Sorry for the double quote - somehow I didn't get the initial mail.
I have just substantially revised the wiki space for this event:
https://fedoraproject.org/wiki/QA/Test_Days/2009-02-19
I converted all the 'how to test' bits into a set of proper test cases, expanded some stuff, replaced some duplicate instructions with links to the appropriate pages, and generally (I hope) made the process a lot clearer. It should be quite easy now to follow the cases and generate data for your system. I've filled in a full set of data from my main system (and comparison data from Mandriva), so you can follow that as a template for filling in the results charts.
Thanks guys! The more data we have the better it is for Harald ;)
On Tue, 2009-02-17 at 15:49 -0800, Adam Williamson wrote:
Sorry for the double quote - somehow I didn't get the initial mail.
I have just substantially revised the wiki space for this event:
https://fedoraproject.org/wiki/QA/Test_Days/2009-02-19
I converted all the 'how to test' bits into a set of proper test cases, expanded some stuff, replaced some duplicate instructions with links to the appropriate pages, and generally (I hope) made the process a lot clearer. It should be quite easy now to follow the cases and generate data for your system. I've filled in a full set of data from my main system (and comparison data from Mandriva), so you can follow that as a template for filling in the results charts.
Thanks guys! The more data we have the better it is for Harald ;)
Looks pretty good. I have a dumb question though, for which I haven't found an answer - how can I enable/disable readahead? I haven't checked in Rawhide yet, but in F10 it is not among services.
Thanks, Martin
On Wed, 2009-02-18 at 09:09 +0100, Martin Sourada wrote:
Looks pretty good. I have a dumb question though, for which I haven't found an answer - how can I enable/disable readahead? I haven't checked in Rawhide yet, but in F10 it is not among services.
It's explained in the test cases, I think? I'm pretty sure I remember writing it there.
You edit /etc/sysconfig/readahead , the first line is the key. The advice to enable / disable it as a service, in the old version of the test day page, was presumably outdated.
On Wed, 2009-02-18 at 01:28 -0800, Adam Williamson wrote:
On Wed, 2009-02-18 at 09:09 +0100, Martin Sourada wrote:
Looks pretty good. I have a dumb question though, for which I haven't found an answer - how can I enable/disable readahead? I haven't checked in Rawhide yet, but in F10 it is not among services.
It's explained in the test cases, I think? I'm pretty sure I remember writing it there.
Ah, thanks, somehow it slipped my attention :)
You edit /etc/sysconfig/readahead , the first line is the key. The advice to enable / disable it as a service, in the old version of the test day page, was presumably outdated.
Martin
Adam Williamson wrote:
On Wed, 2009-02-18 at 09:09 +0100, Martin Sourada wrote:
Looks pretty good. I have a dumb question though, for which I haven't found an answer - how can I enable/disable readahead? I haven't checked in Rawhide yet, but in F10 it is not among services.
It's explained in the test cases, I think? I'm pretty sure I remember writing it there.
You edit /etc/sysconfig/readahead , the first line is the key. The advice to enable / disable it as a service, in the old version of the test day page, was presumably outdated.
don't forget to: # touch /.readahead_collect
after you active readahead to trigger the data collection. Reboot for the data collector and reboot again after a full boot.
On Wed, 2009-02-18 at 18:20 +0100, Harald Hoyer wrote:
Adam Williamson wrote:
On Wed, 2009-02-18 at 09:09 +0100, Martin Sourada wrote:
Looks pretty good. I have a dumb question though, for which I haven't found an answer - how can I enable/disable readahead? I haven't checked in Rawhide yet, but in F10 it is not among services.
It's explained in the test cases, I think? I'm pretty sure I remember writing it there.
You edit /etc/sysconfig/readahead , the first line is the key. The advice to enable / disable it as a service, in the old version of the test day page, was presumably outdated.
don't forget to: # touch /.readahead_collect
after you active readahead to trigger the data collection. Reboot for the data collector and reboot again after a full boot.
...well, that was completely obvious!
so my results which I did with two reboots but *not* with /.readahead_collect would be invalid?
Adam Williamson wrote:
On Wed, 2009-02-18 at 18:20 +0100, Harald Hoyer wrote:
Adam Williamson wrote:
On Wed, 2009-02-18 at 09:09 +0100, Martin Sourada wrote:
Looks pretty good. I have a dumb question though, for which I haven't found an answer - how can I enable/disable readahead? I haven't checked in Rawhide yet, but in F10 it is not among services.
It's explained in the test cases, I think? I'm pretty sure I remember writing it there.
You edit /etc/sysconfig/readahead , the first line is the key. The advice to enable / disable it as a service, in the old version of the test day page, was presumably outdated.
don't forget to: # touch /.readahead_collect
after you active readahead to trigger the data collection. Reboot for the data collector and reboot again after a full boot.
...well, that was completely obvious!
so my results which I did with two reboots but *not* with /.readahead_collect would be invalid?
Well, yes, if you didn't reboot after installing readahead without turning it off.
I added the command to the wiki pages, yesterday.
On Thu, 2009-02-19 at 08:54 +0100, Harald Hoyer wrote:
after you active readahead to trigger the data collection. Reboot for the data collector and reboot again after a full boot.
...well, that was completely obvious!
so my results which I did with two reboots but *not* with /.readahead_collect would be invalid?
Well, yes, if you didn't reboot after installing readahead without turning it off.
Ah. That's OK, then. If you read the test cases, they ask you to reboot *twice* after enabling readahead. I figured it'd require one pass to sort itself out. So the test cases as I wrote them should be valid.
I added the command to the wiki pages, yesterday.
Thanks.
On Tue, 2009-02-17 at 15:49 -0800, Adam Williamson wrote:
On Mon, 2009-02-16 at 17:33 +0100, Harald Hoyer wrote:
<snip>
Sorry for the double quote - somehow I didn't get the initial mail.
I have just substantially revised the wiki space for this event:
https://fedoraproject.org/wiki/QA/Test_Days/2009-02-19
I converted all the 'how to test' bits into a set of proper test cases, expanded some stuff, replaced some duplicate instructions with links to the appropriate pages, and generally (I hope) made the process a lot clearer. It should be quite easy now to follow the cases and generate data for your system. I've filled in a full set of data from my main system (and comparison data from Mandriva), so you can follow that as a template for filling in the results charts.
Nice improvement! That cleans up the page dramatically.
Thanks, James
James Laska wrote:
Greetings testers,
This week's Test Day will focus on another upcoming Fedora 11 feature: 20 Second Startup [1]. This is a follow-on to a Fedora 10 effort to reduce boot delay [2].
Harald Hoyer, David Kovalsky and Ondrej Hudlicky have created a test procedure and looking to gather and analyze bootchart data. Come prepared with your favorite old-skool hardware setup.
Join #fedora-qa this Thursday, February 19, 2009 to help collect and analyze boot speed data. Follow the action at https://fedoraproject.org/wiki/QA/Test_Days/2009-02-19.
Thanks, James
[1] https://fedoraproject.org/wiki/Features/20SecondStartup [2] https://fedoraproject.org/wiki/Features/30SecondStartup
Has the kernel team been contacted?
JBG
On Mon, Feb 16, 2009 at 8:25 AM, James Laska wrote:
Greetings testers,
This week's Test Day will focus on another upcoming Fedora 11 feature: 20 Second Startup [1]. This is a follow-on to a Fedora 10 effort to reduce boot delay [2].
Join #fedora-qa this Thursday, February 19, 2009 to help collect and analyze boot speed data. Follow the action at https://fedoraproject.org/wiki/QA/Test_Days/2009-02-19.
I just tried to follow that procedure and it seems there's a requirement to provide a phone number and address. Really? Are you going to swing by my house and work on my boot times? ;) I don't like giving out my name and address... too dangerous.
Sarah Connor
David L wrote:
On Mon, Feb 16, 2009 at 8:25 AM, James Laska wrote:
Greetings testers,
This week's Test Day will focus on another upcoming Fedora 11 feature: 20 Second Startup [1]. This is a follow-on to a Fedora 10 effort to reduce boot delay [2].
Join #fedora-qa this Thursday, February 19, 2009 to help collect and analyze boot speed data. Follow the action at https://fedoraproject.org/wiki/QA/Test_Days/2009-02-19.
I just tried to follow that procedure and it seems there's a requirement to provide a phone number and address. Really? Are you going to swing by my house and work on my boot times? ;) I don't like giving out my name and address... too dangerous.
You can mark that information as private in the account system. Refer
http://fedoraproject.org/wiki/Legal/PrivacyPolicy
Rahul
James Laska wrote:
Greetings testers,
This week's Test Day will focus on another upcoming Fedora 11 feature: 20 Second Startup [1]. This is a follow-on to a Fedora 10 effort to reduce boot delay [2].
Harald Hoyer, David Kovalsky and Ondrej Hudlicky have created a test procedure and looking to gather and analyze bootchart data. Come prepared with your favorite old-skool hardware setup.
Join #fedora-qa this Thursday, February 19, 2009 to help collect and analyze boot speed data. Follow the action at https://fedoraproject.org/wiki/QA/Test_Days/2009-02-19.
Thanks, James
[1] https://fedoraproject.org/wiki/Features/20SecondStartup [2] https://fedoraproject.org/wiki/Features/30SecondStartup
Thanks to all who participated!
As a first outcome a new and improved readahead-1.4.8 hit rawhide today.
readahead-1.4.8 - set low io priority (no more regression in boot time with readahead) - moved data files from /etc/readahead.d to /var/lib/readahead - don't start readahead from readahead.event if system has less than 384MB
Harald Hoyer wrote:
James Laska wrote:
Greetings testers,
This week's Test Day will focus on another upcoming Fedora 11 feature: 20 Second Startup [1]. This is a follow-on to a Fedora 10 effort to reduce boot delay [2]. Harald Hoyer, David Kovalsky and Ondrej Hudlicky have created a test procedure and looking to gather and analyze bootchart data. Come prepared with your favorite old-skool hardware setup.
Join #fedora-qa this Thursday, February 19, 2009 to help collect and analyze boot speed data. Follow the action at https://fedoraproject.org/wiki/QA/Test_Days/2009-02-19.
Thanks, James
[1] https://fedoraproject.org/wiki/Features/20SecondStartup [2] https://fedoraproject.org/wiki/Features/30SecondStartup
Thanks to all who participated!
As a first outcome a new and improved readahead-1.4.8 hit rawhide today.
readahead-1.4.8
- set low io priority (no more regression in boot time with readahead)
- moved data files from /etc/readahead.d to /var/lib/readahead
- don't start readahead from readahead.event if system has less than
384MB
Here is a blog post about the results:
http://www.harald-hoyer.de/personal/blog/20_Seconds_Boot_Feature_Test_Day
On Mon, Mar 2, 2009 at 10:56 AM, Harald Hoyer harald@redhat.com wrote:
Harald Hoyer wrote:
Here is a blog post about the results:
http://www.harald-hoyer.de/personal/blog/20_Seconds_Boot_Feature_Test_Day
How much time could be saved by building a custom kernel with the always-loaded modules compiled in? I'm planning on putting F11 on my netbook (not arrived yet), and would like to do an apple-for-apple comparison with Moblin.
Thanks,
Michel Salim wrote:
On Mon, Mar 2, 2009 at 10:56 AM, Harald Hoyer harald@redhat.com wrote:
Harald Hoyer wrote: Here is a blog post about the results:
http://www.harald-hoyer.de/personal/blog/20_Seconds_Boot_Feature_Test_Day
How much time could be saved by building a custom kernel with the always-loaded modules compiled in? I'm planning on putting F11 on my netbook (not arrived yet), and would like to do an apple-for-apple comparison with Moblin.
Thanks,
Have a look at your bootchart.. I would say more than 90% of your modprobe time could be saved.