Bug 28044

Summary: Mageia Faster Boot Would be Great!
Product: Mageia Reporter: Ezequiel Partida <ezequiel_partida>
Component: Release (media or process)Assignee: Base system maintainers <basesystem>
Status: RESOLVED FIXED QA Contact:
Severity: enhancement    
Priority: Normal CC: ouaurelien, sysadmin-bugs
Version: Cauldron   
Target Milestone: Mageia 9   
Hardware: All   
OS: Linux   
Whiteboard:
Source RPM: CVE:
Status comment:

Description Ezequiel Partida 2021-01-09 09:26:27 CET
Hello,

Today I tested lxle from lxle.net and it took around 14 seconds to boot on my kid´s Samsung Netbook with SSD and 2gb ram.

LXLE uses LXDE.

Mageia with LXDE booted around twice the time..

I understand that probably Mageia is more complete and robust linux distro, but it would be great to have a release for netbooks or older PCs, I know that Mageia supports a lot of old hardware and it is great, but speeding things up at boot would be wonderful.

On the mean time, dualboot works wonderful on my new HP OMEN, I home Intel supports Optan on linux soon.

Love you guys!!
Comment 1 Aurelien Oudelet 2021-01-09 10:00:03 CET
Yeah. Faster boot time will be here when we ditch all init.d rc script for starting services and when network will be set by Network Manager, not the old initscripts.

In the meantime, you could see which service delays the boot process by do:

$ systemd-analyze blame

This will tell you what make boot longer.
If you can, disable such services.
Don't disable network.service as long as you rely on Drakx-net applet to set network.

CC: (none) => ouaurelien

Comment 2 Aurelien Oudelet 2021-01-12 14:58:20 CET
Not sure really where to assign this.
By the way, I do think this improvement can't be done for M8 now.
Set milestone to M9.
Put it to release, not really a RPM request.

Therefore, I advice users to put here output of "$ systemd-analyze blame"
to see where we should go to improve boot time.

Component: New RPM package request => Release (media or process)
Target Milestone: --- => Mageia 9
CC: (none) => sysadmin-bugs

Comment 3 Ezequiel Partida 2021-01-19 23:56:05 CET
(In reply to Aurelien Oudelet from comment #1)
> Yeah. Faster boot time will be here when we ditch all init.d rc script for
> starting services and when network will be set by Network Manager, not the
> old initscripts.

Now that you mentioned.. I did see that a lot of services where executed on LXLE and boom!! It was booted!  Now I know!.. ;-)
 
 
> In the meantime, you could see which service delays the boot process by do:
> 
> $ systemd-analyze blame
> 
> This will tell you what make boot longer.
> If you can, disable such services.
> Don't disable network.service as long as you rely on Drakx-net applet to set
> network.
Comment 4 Ezequiel Partida 2021-01-19 23:58:14 CET
(In reply to Aurelien Oudelet from comment #2)
> Not sure really where to assign this.
> By the way, I do think this improvement can't be done for M8 now.
> Set milestone to M9.
> Put it to release, not really a RPM request.
> 
> Therefore, I advice users to put here output of "$ systemd-analyze blame"
> to see where we should go to improve boot time.

Mageia 8 Rocks!  I thing M9 would be amazing!

Regards
Comment 5 Aurelien Oudelet 2021-01-26 10:47:29 CET
@ Ezequiel,

Have you tried this:
$ systemd-analyze blame

to let know us what delays the boot process on your machine?
Comment 6 Aurelien Oudelet 2021-01-26 11:03:29 CET
Example:
$ systemd-analyze blame
14.230s systemd-journal-flush.service                                                            
 5.425s network-up.service                                                                       
 4.792s systemd-udev-settle.service                                                              
 3.080s packagekit.service                                                                       
 2.928s dkms-autorebuild.service                                                                 
 1.545s udisks2.service                                                                          
 1.543s home.mount                                                                               
 1.260s lvm2-monitor.service                                                                     
  856ms systemd-random-seed.service                                                              
  740ms shorewall.service                                                                        
  727ms mandriva-everytime.service                                                               
  718ms systemd-fsck@dev-disk-by\x2duuid-73b3802f\x2df451\x2d4a9a\x2d8cb0\x2dc12621c8e45c.service
  711ms systemd-fsck@dev-disk-by\x2duuid-03894e8a\x2d50d0\x2d4026\x2da829\x2d5c2fe72657f9.service
  640ms network.service                                                                          
  555ms fwupd.service                                                                            
  453ms systemd-tmpfiles-setup.service

On my system, very fast boot, but it is delayed by systemd-journal-flush.service
because it is loaded on the same time systemd wants to mount /var and as long as my /var partition is on a rust-spinned drive... I know it is a side effect... having a NVMe for / and /var somewhere else... but it is my choice...

In facts, the only other things that delays the boot process are:
 5.425s network-up.service                                                                       
 4.792s systemd-udev-settle.service 

and:
 2.928s dkms-autorebuild.service     

dkms-autorebuild.service is loaded early and blocks boot for about 3 sec. Too long. But, it is needed here as I use a dkms enabled graphic driver (nvidia) but, it should note take so looong because there nothing to do this time: no rebuilding (no new kernel, no new driver)

systemd-udev-settle.service is deprecated and should be removed... not for M8...
This is responsible to permit all coldplug devices to be recognized by udev... but it not necessary on modern hardware (no ISA...) and it is needed by all dmraid / lvm stuff that I don't have on my systems. Also, many average Desktop users will not have, because by default, our installer does not autosetup LVM partitions for newer installation, it uses basic partitions. (Contrary to Fedora)
I could gain 4,7 sec here.

and
network-up.service is waiting to be online so it blocks boot process for 5.5 sec.
But here it is fast for an old-initscript... I really wonder if NetworkManager will be as fast as it on my system. But, under many circumstances, NetworkManager will be faster than old initscripts... so... maybe.

Regarding this, I could gain nearly 13 seconds for booting...
Note that these services are launched each after one, not in parallel like other .service in systemd because they are needed by other one.

So, I am patient, I do other stuff while this is booting. And it is not so loooong...
But, I wonder what is going on your system.
Comment 7 Aurelien Oudelet 2021-03-08 12:59:43 CET
Assigning.

This should no longer seen in Bugsquad.

Assignee: bugsquad => basesystem

Comment 8 Ezequiel Partida 2021-06-09 23:45:03 CEST
(In reply to Aurelien Oudelet from comment #6)
> Example:
> $ systemd-analyze blame
> 14.230s systemd-journal-flush.service                                       
> 
>  5.425s network-up.service                                                  
> 
>  4.792s systemd-udev-settle.service                                         
> 
>  3.080s packagekit.service                                                  
> 
>  2.928s dkms-autorebuild.service                                            
> 
>  1.545s udisks2.service                                                     
> 
>  1.543s home.mount                                                          
> 
>  1.260s lvm2-monitor.service                                                
> 
>   856ms systemd-random-seed.service                                         
> 
>   740ms shorewall.service                                                   
> 
>   727ms mandriva-everytime.service                                          
> 
>   718ms
> systemd-fsck@dev-disk-by\x2duuid-
> 73b3802f\x2df451\x2d4a9a\x2d8cb0\x2dc12621c8e45c.service
>   711ms
> systemd-fsck@dev-disk-by\x2duuid-
> 03894e8a\x2d50d0\x2d4026\x2da829\x2d5c2fe72657f9.service
>   640ms network.service                                                     
> 
>   555ms fwupd.service                                                       
> 
>   453ms systemd-tmpfiles-setup.service
> 
> On my system, very fast boot, but it is delayed by
> systemd-journal-flush.service
> because it is loaded on the same time systemd wants to mount /var and as
> long as my /var partition is on a rust-spinned drive... I know it is a side
> effect... having a NVMe for / and /var somewhere else... but it is my
> choice...
> 
> In facts, the only other things that delays the boot process are:
>  5.425s network-up.service                                                  
> 
>  4.792s systemd-udev-settle.service 
> 
> and:
>  2.928s dkms-autorebuild.service     
> 
> dkms-autorebuild.service is loaded early and blocks boot for about 3 sec.
> Too long. But, it is needed here as I use a dkms enabled graphic driver
> (nvidia) but, it should note take so looong because there nothing to do this
> time: no rebuilding (no new kernel, no new driver)
> 
> systemd-udev-settle.service is deprecated and should be removed... not for
> M8...
> This is responsible to permit all coldplug devices to be recognized by
> udev... but it not necessary on modern hardware (no ISA...) and it is needed
> by all dmraid / lvm stuff that I don't have on my systems. Also, many
> average Desktop users will not have, because by default, our installer does
> not autosetup LVM partitions for newer installation, it uses basic
> partitions. (Contrary to Fedora)
> I could gain 4,7 sec here.
> 
> and
> network-up.service is waiting to be online so it blocks boot process for 5.5
> sec.
> But here it is fast for an old-initscript... I really wonder if
> NetworkManager will be as fast as it on my system. But, under many
> circumstances, NetworkManager will be faster than old initscripts... so...
> maybe.
> 
> Regarding this, I could gain nearly 13 seconds for booting...
> Note that these services are launched each after one, not in parallel like
> other .service in systemd because they are needed by other one.
> 
> So, I am patient, I do other stuff while this is booting. And it is not so
> loooong...
> But, I wonder what is going on your system.

Thank you for the info Aurelien.

Mageia 8 is getting even faster to boot specially with SSD.

Would it be to dificult to make them work in parallel like Kubuntu?

I did some test and Kubuntu boots in around 5 seconds vs 13 mageia 8.

Just asking! ;-)
Comment 9 Ezequiel Partida 2022-12-13 18:34:38 CET
Hello Everyone,

The new mageia 9 boot system is great.. Thanks for listening.

Regards

Status: NEW => RESOLVED
Resolution: (none) => FIXED