Scheduled Maintenance: We are aware of an issue with Google, AOL, and Yahoo services as email providers which are blocking new registrations. We are trying to fix the issue and we have several internal and external support tickets in process to resolve the issue. Please see: viewtopic.php?t=158230

 

 

 

Boot time/memory comparison: sysvinit vs systemd

Here you can discuss every aspect of Debian. Note: not for support requests!
Message
Author
User avatar
Flemur
Posts: 25
Joined: 2015-01-22 16:52

Boot time/memory comparison: sysvinit vs systemd

#1 Post by Flemur »

Debian jessie 7.8 with sysvinit and systemd installed.

Conclusion: same boot time to within wrist-watch accuracy (18sec), systemd used 3M more mem.

Memory measured with ps_mem.py (called 'mem'); I forgot to do "free -m"...

With systemd:

Code: Select all

Boot time = 18sec (grub prompt to visible desktop)
$ systemd-analyze
Startup finished in 4.599s (kernel) + 6.528s (userspace) = 11.127s
(note the difference between 11s reported and 18s real boot time)
# mem
 Private  +   Shared  =  RAM used   Program
...snip...
  1.3 MiB +   1.9 MiB =   3.2 MiB   systemd (2)
  3.5 MiB +   1.5 MiB =   5.0 MiB   bash (3)
  3.1 MiB +   2.1 MiB =   5.2 MiB   fluxbox
  7.3 MiB +  49.0 KiB =   7.3 MiB   dhclient
 11.3 MiB +   3.8 MiB =  15.1 MiB   lxterminal
 54.1 MiB + 821.0 KiB =  54.9 MiB   Xorg
---------------------------------
                        112.9 MiB
#
with sysvinit

Code: Select all

Boot time 18 sec.  (grub prompt to visible desktop)
# mem
 Private  +   Shared  =  RAM used   Program
...snip...
  1.2 MiB +   1.7 MiB =   2.9 MiB   polkitd
  3.6 MiB +   1.5 MiB =   5.0 MiB   bash (3)
  3.2 MiB +   2.1 MiB =   5.3 MiB   fluxbox
  7.3 MiB +  48.0 KiB =   7.4 MiB   dhclient
 11.5 MiB +   3.7 MiB =  15.3 MiB   lxterminal
 54.4 MiB + 836.0 KiB =  55.2 MiB   Xorg
---------------------------------   
                        109.3 MiB
#
Xorg could be made 20-30M smaller by using nouveau drivers rather than nvidia.
Debian 8+fluxbox-pulseaudio

mean_dean
Posts: 132
Joined: 2014-12-29 18:21

Re: Boot time/memory comparison: sysvinit vs systemd

#2 Post by mean_dean »

for minimal systems I have never seen that systemd improved boot time

for minimal installs systemd would just complicate various things in my opinon

Randicus
Posts: 2663
Joined: 2011-05-08 09:11

Re: Boot time/memory comparison: sysvinit vs systemd

#3 Post by Randicus »

Flemur wrote:Debian jessie 7.8 with sysvinit and systemd installed.
I thought Debian 7 is Wheezy and Jessie is what will become Debian 8.

mean_dean
Posts: 132
Joined: 2014-12-29 18:21

Re: Boot time/memory comparison: sysvinit vs systemd

#4 Post by mean_dean »

at least he didnt call it jessie jackalope or whatever the j release of buntu was...

User avatar
Head_on_a_Stick
Posts: 14114
Joined: 2014-06-01 17:46
Location: London, England
Has thanked: 81 times
Been thanked: 132 times

Re: Boot time/memory comparison: sysvinit vs systemd

#5 Post by Head_on_a_Stick »

Please post the output of:

Code: Select all

systemd-analyze critical-chain
Also, can you provide a break down of the timing of the boot components under SysVinit plz.
deadbang

User avatar
Flemur
Posts: 25
Joined: 2015-01-22 16:52

Re: Boot time/memory comparison: sysvinit vs systemd

#6 Post by Flemur »

Randicus:
I thought Debian 7 is Wheezy and Jessie is what will become Debian 8.


I installed 7.8 stable then changed repositories to testing. I don't know "what it will become", just what it is.
See below about the names.

mean_dean » 2015-01-28 18:39
at least he didnt call it jessie jackalope or whatever the j release of buntu was...


Giving stupid names to software distributions is lame and pointless: that's why you care.

Code: Select all

$ systemd-analyze critical-chain
The time after the unit is active or started is printed after the "@" character.
The time the unit takes to start is printed after the "+" character.

graphical.target @6.501s
└─multi-user.target @6.501s
  └─virtualbox.service @5.972s +528ms
    └─basic.target @5.963s
      └─paths.target @5.961s
        └─acpid.path @5.961s
          └─sysinit.target @5.954s
            └─nfs-common.service @5.640s +313ms
              └─rpcbind.target @5.639s
                └─rpcbind.service @5.379s +259ms
                  └─network-online.target @5.378s
                    └─network.target @5.377s
                      └─ifup@eth0.service @5.377s
                        └─networking.service @4.015s +1.361s
                          └─local-fs.target @4.012s
                            └─run-rpc_pipefs.mount @5.904s
                              └─local-fs-pre.target @3.475s
                                └─systemd-remount-fs.service @3.436s +36ms
                                  └─keyboard-setup.service @1.397s +2.038s
                                    └─systemd-udevd.service @1.301s +93ms
                                      └─systemd-tmpfiles-setup-dev.service @870ms +429ms
                                        └─kmod-static-nodes.service @799ms +69ms
                                          └─system.slice @789ms
                                            └─-.slice @785ms
$ 
Debian 8+fluxbox-pulseaudio

User avatar
Head_on_a_Stick
Posts: 14114
Joined: 2014-06-01 17:46
Location: London, England
Has thanked: 81 times
Been thanked: 132 times

Re: Boot time/memory comparison: sysvinit vs systemd

#7 Post by Head_on_a_Stick »

Flemur wrote:

Code: Select all

(note the difference between 11s reported and 18s real boot time)
This difference will be the time taken for the actual desktop to load.

Interesting info, thank you!
deadbang

User avatar
Flemur
Posts: 25
Joined: 2015-01-22 16:52

Re: Boot time/memory comparison: sysvinit vs systemd

#8 Post by Flemur »

Head_on_a_Stick » 2015-01-29 07:53
Interesting info, thank you!


Numbers are good!

This difference will be the time taken for the actual desktop to load.

I just mentioned that since I've seen people use systemd-analyze time as their "boot time".
I'm doing autologin/startx without a display/login manager, so that 6-7 or so seconds difference seems like a lot since fluxbox starts so fast ... ?

Here's a bootchart (izzat what you wanted? I'm more used to systemd from Arch, tho not anything like an expert at any of 'em)
Nothing at all has been optimized beyond what the Deb installer might do.

Image
Debian 8+fluxbox-pulseaudio

User avatar
Head_on_a_Stick
Posts: 14114
Joined: 2014-06-01 17:46
Location: London, England
Has thanked: 81 times
Been thanked: 132 times

Re: Boot time/memory comparison: sysvinit vs systemd

#9 Post by Head_on_a_Stick »

Flemur wrote:I'm doing autologin/startx without a display/login manager, so that 6-7 or so seconds difference seems like a lot since fluxbox starts so fast ... ?
Seems about right to me -- I have a PekWM/lxpanel desktop that takes about that time to start once logged in (no autostart).

Have you tried disabling the autostart so you can get an idea of the time it takes to start the desktop?

Thanks for the graphs -- it's that kind of functionality that's missing from SysVinit, rather than any deficiencies in boot time.

I'll try my Debian sid system later with SysVinit & a stopwatch to see if it can match the ~4 second startup time under systemd...

Debian sid, systemd:

Code: Select all

Startup finished in 2.314s (kernel) + 1.875s (userspace) = 4.190s
On the stopwatch from gummiboot menu to GNOME desktop (with autostart): ~8 seconds.

Somewhat embarrassingly, SysVinit won't boot to the display manager and I can't disable the display manager under systemd... :oops:
Under SysVinit, to the (console) log in is ~5.4 seconds so it seems very slightly slower, but I can't be absolutely sure.
Last edited by Head_on_a_Stick on 2015-01-29 17:31, edited 3 times in total.
deadbang

thenewguy
Posts: 33
Joined: 2011-07-09 19:35

Re: Boot time/memory comparison: sysvinit vs systemd

#10 Post by thenewguy »

DistroWatch did a side-by side test of Debian Testing where boot times between SysV and systemd were compared. The boot times were identical.
http://distrowatch.com/weekly.php?issue=20141027#qa

Seems weird when projects like openSUSE and Fedora claimed they switched init technology in large part because of the difference in boot times. Looking over the systemd website, the claims about systemd being faster seem to have been removed. I think everyone is now aware (or at least they should be) systemd is not faster.

User avatar
Flemur
Posts: 25
Joined: 2015-01-22 16:52

Re: Boot time/memory comparison: sysvinit vs systemd

#11 Post by Flemur »

Head_on_a_Stick : Have you tried disabling the autostart so you can get an idea of the time it takes to start the desktop?

A little less than 3 seconds for startx -> desktop.

thenewguy: I think everyone is now aware (or at least they should be) systemd is not faster.

I'd seen that claim a lot, hence this thread. Upstart's about the same, too.

I like systemd's syntax and tools, and don't know enough to care about the "taking over the system" aspect, and very rarely mess with any of them anyway; rather than shiny new boot managers I'd prefer a shiny new ALSA.
Debian 8+fluxbox-pulseaudio

User avatar
NkfzGx3ok
Posts: 20
Joined: 2014-10-30 12:08

Re: Boot time/memory comparison: sysvinit vs systemd

#12 Post by NkfzGx3ok »

thenewguy wrote:Seems weird when projects like openSUSE and Fedora claimed they switched init technology in large part because of the difference in boot times.
Have any official links to any of that? I've been around systemd and the discussion in Fedora and SUSE since 2010 and the speed has always been a welcome side effect and never a tier 1 reason.
thenewguy wrote: Looking over the systemd website, the claims about systemd being faster seem to have been removed. I think everyone is now aware (or at least they should be) systemd is not faster.
Again, citation, please? You can use archive services to go back and view the pages and get links. AFIAK in the past 5 years systemd has never been marketed it's developers or distros devs because of speed (again, just a nice side effect), that is an end user argument flung around the net because visually that is the only noticeable difference.


Oh, and to be on topic, one of our debian test servers:

Code: Select all

# Startup finished in 1.489s (kernel) + 2.015s (initrd) + 1.838s (userspace) = 5.342s

User avatar
Flemur
Posts: 25
Joined: 2015-01-22 16:52

Re: Boot time/memory comparison: sysvinit vs systemd

#13 Post by Flemur »

https://news.opensuse.org/2011/12/22/sy ... suse-12-1/
"systemd –[sic] boot faster and cleaner with openSUSE 12.1"
Debian 8+fluxbox-pulseaudio

User avatar
edbarx
Posts: 5401
Joined: 2007-07-18 06:19
Location: 35° 50 N, 14 º 35 E
Been thanked: 2 times

Re: Boot time/memory comparison: sysvinit vs systemd

#14 Post by edbarx »

Boot times under a few minutes do not count as a problem for anyone who does not boot his computer often enough. At least, this is how I see it. Boot times became a nuisance, when the giant of operating systems, that is, Microsoft Windows, started to make the life of computer users miserable when boot times took several minutes.

For a few seconds more of boot time to use my computer, I see no reason to use a potentially broken and immature OS-initialisation system. It is far better for me to write my own as the source for OS-initialisation systems is open and easy to access.

However, right now, I am having other more serious problems, regarding my health, and regarding other more urgent issues.
Debian == { > 30, 000 packages }; Debian != systemd
The worst infection of all, is a false sense of security!
It is hard to get away from CLI tools.

confuseling
Posts: 2121
Joined: 2009-10-21 01:03

Re: Boot time/memory comparison: sysvinit vs systemd

#15 Post by confuseling »

wizard10000 wrote:...
On a server I really wouldn't care about a few seconds of boot time and to be honest I don't care much about it on my laptop either. If someone is bouncing a server often enough for a couple seconds boot time to even factor into an init system either their server's broken or they're a really poor sysadmin :)

Same with my personal machine, which only gets rebooted during kernel upgrades for the most part.
I've heard it argued that small improvements in boot times are very important to virtualised server farms, where you might be spinning hundreds (or thousands) of images up depending on demand.

But yeah, for my laptop and my Raspberry Pi, the difference between 10 seconds and 1 minute is basically meaningless...

Phones and in-vehicle entertainment etc., on the other hand, are another area where people care.
The Forum's search box is terrible. Use site specific search, e.g.
https://www.google.com/search?q=site%3A ... terms+here

User avatar
mardybear
Posts: 994
Joined: 2014-01-19 03:30

Re: Boot time/memory comparison: sysvinit vs systemd

#16 Post by mardybear »

Boot time is such a silly sell or welcomed side effect...and systemd doesn't seem to make much of a difference anyway. Using the same hardware, my busybox-based installs have booted to desktop before sysvinit even gets to the display manager login. That's just me using a desktop though, i'm no system administrator. Just IMO, lean and simple is generally better, not adding extra layers of complexity.
800mhz, 512mb ram, dCore-jessie (Tiny Core with Debian Jessie packages) with BusyBox and Fluxbox.
Most don't have computer access, reuse or pay forward an old computer.

thenewguy
Posts: 33
Joined: 2011-07-09 19:35

Re: Boot time/memory comparison: sysvinit vs systemd

#17 Post by thenewguy »

NkfzGx3ok wrote:
thenewguy wrote:Seems weird when projects like openSUSE and Fedora claimed they switched init technology in large part because of the difference in boot times.
Have any official links to any of that? I've been around systemd and the discussion in Fedora and SUSE since 2010 and the speed has always been a welcome side effect and never a tier 1 reason.
thenewguy wrote: Looking over the systemd website, the claims about systemd being faster seem to have been removed. I think everyone is now aware (or at least they should be) systemd is not faster.
Again, citation, please? You can use archive services to go back and view the pages and get links. AFIAK in the past 5 years systemd has never been marketed it's developers or distros devs because of speed (again, just a nice side effect), that is an end user argument flung around the net because visually that is the only noticeable difference.


Oh, and to be on topic, one of our debian test servers:

Code: Select all

# Startup finished in 1.489s (kernel) + 2.015s (initrd) + 1.838s (userspace) = 5.342s
Someone else already replied with the link to openSUSE's notes on systemd supposedly being faster. Here is a link to Debian's debate on systemd where the same claim is made: https://wiki.debian.org/Debate/initsyst ... to_systemd
And here is the link to where Fedora makes the same claim: https://fedoraproject.org/wiki/Features/systemd

User avatar
NkfzGx3ok
Posts: 20
Joined: 2014-10-30 12:08

Re: Boot time/memory comparison: sysvinit vs systemd

#18 Post by NkfzGx3ok »

thenewguy wrote:Someone else already replied with the link to openSUSE's notes on systemd supposedly being faster. Here is a link to Debian's debate on systemd where the same claim is made: https://wiki.debian.org/Debate/initsyst ... to_systemd
And here is the link to where Fedora makes the same claim: https://fedoraproject.org/wiki/Features/systemd
Let's keep the following which you said in mind:
thenewguy wrote:Seems weird when projects like openSUSE and Fedora claimed they switched init technology in large part because of the difference in boot times.
https://wiki.debian.org/Debate/initsyst ... to_systemd
"Systemd is incredibly fast (1 second to boot). It was not designed with speed in mind, but doing things correctly avoids all the delays currently incurred by the boot process."
5th bullet point on the list, doesn't seem very 'switching init technology for speed'-y to me.

https://fedoraproject.org/wiki/Features/systemd
"Note that in the long run systemd will provide quicker boot up, but completing that will take more work, since some additional changes to our early boot-up need to take place such as splitting up rc.sysinit. However the focus of systemd is primarily doing things right, and not exclusively on speed."
Final note half way down the page, even goes in to add clarification. Doesn't seem very 'switching init technology for speed'-y to me.

https://news.opensuse.org/2011/12/22/sy ... suse-12-1/
"systemd – boot faster and cleaner with openSUSE 12.1"
Attention grabbing headline, mentions other things than just speed but I can see it, kinda.

"It comes with many new features like socket and dbus-activation, use of cgroups (control groups) and aggressive parallelization capabilities which leads to a faster boot-up of the system."
First actual mention in the article, goes in to technical details and a lists speed as a benefit users may notice, doesn't seem very 'switching init technology for speed'-y to me.


====
Just to add: I'm not sure what any of these user facing articles/wiki entries have to do with the mass of internal talks/debates that were had across irc and mailing lists between developers and contributors [that lead to systemd being made default in all those distros years ago].

Almost forgot, I am still waiting on:
thenewguy wrote: Looking over the systemd website, the claims about systemd being faster seem to have been removed. I think everyone is now aware (or at least they should be) systemd is not faster.
Do you have the archive links so we can see when and what was removed, please?
====

Deb-fan
Posts: 1047
Joined: 2012-08-14 12:27
Been thanked: 4 times

Re: Boot time/memory comparison: sysvinit vs systemd

#19 Post by Deb-fan »

Judging by what many people have said, mainly and mostly Arch linux users, have to conclude it is faster and parallelizing bootup, one of the advantages claimed. However ... brings me to the next point/question. Which version was used here ? The one in testing and unstable is like 3.15, don't quote me on this, think last time checked the one in experimental was 3.18 and in Arch it's no doubt or almost no doubt, whatever is the lastest version of systemd.

Think common sense dictates that bugs were found and squashed and/or features added between 3.15 and .18 ? Again that would just be a reasonable assumption imo. When a newer version is released, makes sense something would be different and hopefully better in it.

One of Arch linux's trademark, the whole our distro is cutting edge latest thing. Back when did try Arch, yeah I liked it and had no problem admitting it can be great, though readily admit Debian is great imo/experience too. Had a minimal Sid install and an Arch install side-by-side and from what I saw 1st hand, sometimes I'd check the unstable repos and compare what was avail in the Arch community repos. Sometimes the versions avail would be the same, sometimes Arch would have newer, sometimes Debian would.

(Edit: nope self-correction on the above, had the Deb experimental repo enabled in that case, am almost sure anyway and don't/didn't like the idea of using it often.)

Also both have additional repo's someone who wants to can enable, Arch has a few, one option being the AUR and really Debian does already have an equiv to it. That being Debian-based Sid distro's, who are 100% Debian compatible that maintain and patch software that may not be avail in Debs software repos, shrugs. For me personally in Arch the community repo was good enough and in Deb/Sid, didn't really want to mess w experimental often either.

Not to mention the tons of other ways someone can get access to newer or even exotic packages not readily avail in a Deb repo. Just saying, judging by the feedback of quite a few Arch users, too many keep harping about a very noticeable boot time improvement for me to believe they're all fulla-chit, more shrugs.

Another thing that occurs to me, there are several other outside factors that could improve or degrade init whether someone is talking about sysv or systemd( for that matter OS function overall PERIOD, including all sys resources o course), not going to bother listing all of the chit-ton of them. If ya don't know, then ya don't, google, experiment and learn all the ways to tweak and enhance a gnu/Linux OS = any distro. With some know how a person can get an outdated system, booting faster than a much higher spec system that's in the hands of an incompetent user. Within reason of course, not talking about an SSD w a dual-quad core vs a 512mb ddr2 w a P4 cpu. ;)

For me though, definitely have more important things to do, than sit in front of a pc w a stop watch anyway. By this pt ... couldn't care less what init system anonymous nixers prefer, for whichever reasons. :D

Jessie hasn't even gone stable, so which version of systemd it'll come w default isn't set in stone. Nor is the gazillion other easy options people have availabe to them, to use something else. For sure 100% it's all about choice, application and skillset. Have absolutely nothing against people using whatever they prefer.
Edit2: Oops cr&p, another thanks to the OP, is interesting and to other community members who shared opinions/insights too.
Most powerful FREE tech-support tool on the planet * HERE. *

Deb-fan
Posts: 1047
Joined: 2012-08-14 12:27
Been thanked: 4 times

Re: Boot time/memory comparison: sysvinit vs systemd

#20 Post by Deb-fan »

Dang it ... just when I thought was safe to go back in the water ! Some more comments about this bs popped into head. *hears everyone in the community cheering, lmao ... Stop ya bytching it'll only take a min peeps. :D

There are several systemd specific tweaks I'm aware of and some for sysvinit too for that matter. So systemd out-of-box, even latest version vs latest and tweaked, yanno ? As for which bottomline would prove faster, on desktop usecase or xyz other application ... Am not interested enough to bother finding out, shrugs. Also not curious enough to ever do a custom minimal Deb w latest systemd vs an Arch w same comparison either. If someone linked to a credible one ... will be happy 2 click on it though. ;)

Also hardware specs do play a role, think it's reasonable to assume also that many Arch users aren't just concerned with latest shiny software versions, don't doubt quite a % of them are also obsessed w the latest shiny hardware to run their latest shiny software on. :)

GPT+UEFI and a gnu/nix OS installed to fully use them vs outdated bios/mbr hardware thingy. Personally also think it's reasonable to assume all these Arch users trumpeting about how much faster systemd is for them, are using an older Arch install w sysvinit vs an Arch install w systemd default on the same hardware. As tweaked and latest obsessed as many Archers seem to be, I don't doubt @ all, there's 100's of Arch users sitting @ their pc w a stop watch, while am typing this. :P

Last stupe comment, also as several people have already said, boottime isn't all that friggin important me thinks. Many nixers NEVER turn off their personal computers and system admins w a production server(s), unless there's some good reason or they're retarded, shoot for months, even years between reboots, shrugs. So it'd poss be of interest to some gnu/nix desktoppers I guess.
Most powerful FREE tech-support tool on the planet * HERE. *

Post Reply