[gentoo-user] initng or runit?

[gentoo-user] initng or runit?

[Jerry McBride]

Would some kind soul save me a bit of research time? Which of the two 
alternative init schemes are faster, initng or runit?


Thank you in advance , Jerry
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] initng or runit?

[Neil Bothwick]

[-- Attachment #1: Type: text/plain, Size: 508 bytes --]

On Sun, 20 Aug 2006 22:01:24 -0400, Jerry McBride wrote:

> Would some kind soul save me a bit of research time? Which of the two 
> alternative init schemes are faster, initng or runit?

There's a third option that doesn't involve switching init systems, set 
RC_PARALLEL_STARTUP="no" in /etc/conf.d/rc. I saw a significant speedup
when using this, and it is fully integrated into baselayout.


-- 
Neil Bothwick

If bankers can count, how come they have eight windows and only four
tellers?

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

[gentoo-user] Re: initng or runit?

[Alexander Skwar]

· Neil Bothwick <neil@digimed.co.uk>:

> On Sun, 20 Aug 2006 22:01:24 -0400, Jerry McBride wrote:
> 
>> Would some kind soul save me a bit of research time? Which of the two 
>> alternative init schemes are faster, initng or runit?
> 
> There's a third option that doesn't involve switching init systems, set 
> RC_PARALLEL_STARTUP="no" in /etc/conf.d/rc. I saw a significant speedup
> when using this, and it is fully integrated into baselayout.

You saw a *speedup* when *disabling* parallel startup by setting 
RC_PARALLEL_STARTUP="no"?

I thought boot would be a bit faster, when parallel startup is 
*enabled*.

Alexander Skwar
-- 
Was eine Nation groß macht, sind nicht in erster Linie die großen
Männer. Es ist das Format der Mittelmäßigen.
                -- José Ortega y Gasset


-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Re: initng or runit?

[Neil Bothwick]

[-- Attachment #1: Type: text/plain, Size: 815 bytes --]

On Mon, 21 Aug 2006 10:51:27 +0200, Alexander Skwar wrote:

> > There's a third option that doesn't involve switching init systems,
> > set RC_PARALLEL_STARTUP="no" in /etc/conf.d/rc. I saw a significant
> > speedup when using this, and it is fully integrated into baselayout.  
> 
> You saw a *speedup* when *disabling* parallel startup by setting 
> RC_PARALLEL_STARTUP="no"?
> 
> I thought boot would be a bit faster, when parallel startup is 
> *enabled*.

Doh! I pasted from this box, on which I haven't enabled parallel startup
(yet), but I'm using it on my other one, which is faster with it set to
"yes". I meant to change it but it's still too early in the week for me
to function properly :(


-- 
Neil Bothwick

Welcome to the world of Windows 95. Stay a while -- stay foooreveeer.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] initng or runit?

[Hans-Werner Hilse]

Hi,

On Sun, 20 Aug 2006 22:01:24 -0400 Jerry McBride
<mcbrides9@comcast.net> wrote:

> Would some kind soul save me a bit of research time? Which of the two 
> alternative init schemes are faster, initng or runit?

That most likely doesn't depend on the init process. Most time is
consumed by the numerous (re-)starts of /bin/sh, i.e. bash in most
cases, through all the init scripts.

If you're about to play with an embedded device like machine and boot
time really matters, I'd suggest writing the system setup tasks (rcS)
in pure C. If you want to save a few shell startups, you might as well
use /etc/inittab and sysvinit. Sysvinit, initng, runit or minit (which
I like best) doesn't really matter for timing. That time is wasted in
other places.

For a VDR (digital PVR) machine, I'm using busybox' reduced sysvinit
clone. Works like a charm, from boot till VDR running it's about 30sec.
You might get a few more seconds for reimplementing system setup in
pure C, as suggested. AFAIK, e.g. the Linksys Linux firmware does that.
You might consider using their program as a template.

-hwh
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] initng or runit?

[Jerry McBride]

On Monday 21 August 2006 09:35, Hans-Werner Hilse wrote:
> Hi,
>
> On Sun, 20 Aug 2006 22:01:24 -0400 Jerry McBride
>
> <mcbrides9@comcast.net> wrote:
> > Would some kind soul save me a bit of research time? Which of the two
> > alternative init schemes are faster, initng or runit?
>
> That most likely doesn't depend on the init process. Most time is
> consumed by the numerous (re-)starts of /bin/sh, i.e. bash in most
> cases, through all the init scripts.
>
> If you're about to play with an embedded device like machine and boot
> time really matters, I'd suggest writing the system setup tasks (rcS)
> in pure C. If you want to save a few shell startups, you might as well
> use /etc/inittab and sysvinit. Sysvinit, initng, runit or minit (which
> I like best) doesn't really matter for timing. That time is wasted in
> other places.
>
> For a VDR (digital PVR) machine, I'm using busybox' reduced sysvinit
> clone. Works like a charm, from boot till VDR running it's about 30sec.
> You might get a few more seconds for reimplementing system setup in
> pure C, as suggested. AFAIK, e.g. the Linksys Linux firmware does that.
> You might consider using their program as a template..
>

We've settled upon initng and fcache. Between the two, I can boot a "fully 
loaded" laptop in under 13 seconds. That's pretty impressive.

Thanks for the tips and info.

Cheers... Jerry.


-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] initng or runit?

[Thomas Kear]

I've never tried runit, but being an initng dev, I'm expected to be
somewhat biased.

Having just had a quick look at runit, it appears we have the larger
collection of init scripts, however being simple bash scripts it could
be expected that runit's are slightly easier to make.

The real advantage that initng posesses are, unless runit's website is
simply failing to mention it, auto-restarting of failed daemons,
virtual dependencies, and clever multi-distro compatible scripts that
use a preprocessor at install time (see the #ifd / #elsed / #endd
statements in a file like
http://svn.initng.org/initng-ifiles/trunk/initfiles/system/modules.ii).

I've never tried RC_PARALLEL_STARTUP, haven't used sysvinit at all in 9 months.

-- 
Thomas Kear
thomas.kear@gmail.com
+6421996486
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] initng or runit?

[Dale]

Thomas Kear wrote:
>
>
> I've never tried RC_PARALLEL_STARTUP, haven't used sysvinit at all in
> 9 months.
>

I recently started using this "feature".  It works OK.  Nothing fails to
start.  It does try to start my networks too early but it seems to just
wait until it is ready then starts them up just fine.  It seems to boot
a little bit faster.  The messages are a bit more confusing at first. 
The little "[OK]" thing on the right is gone too.

It didn't seem to hurt anything.  You may want to give it a try.  Still
trying to figure out this other stuff though.  < scratches head > 
Where's Holly when you need her?

Dale

:-)


-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] initng or runit?

[Mrugesh Karnik]

[-- Attachment #1: Type: text/plain, Size: 860 bytes --]

On Wednesday 23 August 2006 07:28, Thomas Kear wrote:

> The real advantage that initng posesses are, unless runit's website
> is simply failing to mention it, auto-restarting of failed daemons,
> virtual dependencies, and clever multi-distro compatible scripts that
> use a preprocessor at install time (see the #ifd / #elsed / #endd
> statements in a file like
> http://svn.initng.org/initng-ifiles/trunk/initfiles/system/modules.ii
>).

I've used runit for a little while. It uses daemontools from qmail to 
monitor services. Hence it also has auto-restarting of failed daemons.
I haven't used it long enough to know about dependencies though. I 
thought runit was pretty good.

-- 
----------------------------------------
Mrugesh Karnik
GPG Key 0xBA6F1DA8
Public key on http://wwwkeys.pgp.net
----------------------------------------

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] initng or runit?

[Hans-Werner Hilse]

Hi,

On Wed, 23 Aug 2006 13:58:01 +1200 "Thomas Kear"
<thomas.kear@gmail.com> wrote:

> Having just had a quick look at runit, it appears we have the larger
> collection of init scripts, however being simple bash scripts it could
> be expected that runit's are slightly easier to make.

Runit does *not* depend on bash scripts but whatever you want. The
runsv binary just starts ./run in the service directory.

I like fefe's minit better, because you can even set arguments to
that ./run binary. That way, you can just make a symlink to the daemon's
executable and store the params in that mentioned file. This means you
don't have to start some kind of shell or script in order to execute
the daemon.

> The real advantage that initng posesses are, unless runit's website is
> simply failing to mention it, auto-restarting of failed daemons,

heh, runit's main purpose is "service supervision". If it wouldn't
restart failed daemons, the easiest approach would have bin
a /bin/sh-skript as init.

> virtual dependencies,

Yep, that's right. Minit has better dependency management, though, but
I think this would be overkill in most embedded systems anyway.

> and clever multi-distro compatible scripts that
> use a preprocessor at install time (see the #ifd / #elsed / #endd
> statements in a file like
> http://svn.initng.org/initng-ifiles/trunk/initfiles/system/modules.ii).

I don't think that start scripts for various daemons are absolutely
needed. They're more part of the init's documentation. And hey: For
best speed, multi-distro compatibility is a no-no.

> I've never tried RC_PARALLEL_STARTUP, haven't used sysvinit at all in
> 9 months.

I'd like to add that gentoo's own /sbin/rc is not a feature of
sysvinit. sysvinit is not *that* bad, after all. What has gone wild is
the /etc/init.d style of doing things.

-hwh

PS: the mentioned 30secs boot time in my previous mail to this thread
are for a Pentium MMX 166. Using Gentoo's init scheme, that would have
been forever.
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] initng or runit?

[Jerry McBride]

On Wednesday 23 August 2006 10:17, Hans-Werner Hilse wrote:
> Hi,

--snip--

>
> I'd like to add that gentoo's own /sbin/rc is not a feature of
> sysvinit. sysvinit is not *that* bad, after all. What has gone wild is
> the /etc/init.d style of doing things.
>

Amen! And a bottle neck of epic proportions.

-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] initng or runit?

[Daniel Iliev]

Jerry McBride wrote:
> Would some kind soul save me a bit of research time? Which of the two 
> alternative init schemes are faster, initng or runit?
>
>
> Thank you in advance , Jerry
>   
I have tried initng several months ago. It rocks. It's several times
faster then the "normal" init. The problem at the time was there were no
scripts for everything I wanted to start automatically. So one day I
figured out that writing scripts and using faster init takes me more
time then using slower init which works with almost no maintenance. This
made me go back to the normal init. I have to say that while using
initng I noticed that many scripts were added for a relatively short
time. It is possible that now there are initng scripts for most of the
services one would ever use, but you have to check it out for yourself.

I can't say a word about "runit", because it's the first time I read
about it.

My next experiment for speeding the boot up will be fcache, but I'm
waiting for a proper mood to try it ( it means: "I'm too lazy" )


-- 
Best regards,
Daniel


-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] initng or runit?

[Jerry McBride]

On Wednesday 23 August 2006 18:42, Daniel Iliev wrote:
> Jerry McBride wrote:
> > Would some kind soul save me a bit of research time? Which of the two
> > alternative init schemes are faster, initng or runit?
> >
> >
> > Thank you in advance , Jerry
>
> I have tried initng several months ago. It rocks. It's several times
> faster then the "normal" init. The problem at the time was there were no
> scripts for everything I wanted to start automatically. So one day I
> figured out that writing scripts and using faster init takes me more
> time then using slower init which works with almost no maintenance. This
> made me go back to the normal init. I have to say that while using
> initng I noticed that many scripts were added for a relatively short
> time. It is possible that now there are initng scripts for most of the
> services one would ever use, but you have to check it out for yourself.
>
> I can't say a word about "runit", because it's the first time I read
> about it.
>
> My next experiment for speeding the boot up will be fcache, but I'm
> waiting for a proper mood to try it ( it means: "I'm too lazy" )
>

Hi Daniel,

Fcache works, but we didn't see the performance boost that going to initng 
gave. Since it requires it's very own ext3 partition to work, plus a kernel 
patch... we dropped it.  

Using initng is the ticket... Maybe the gentoo devs will directly support it 
or a variant someday...

Cheers, Jerry



-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] [OT] initng or runit?

[Daniel Iliev]

Jerry McBride wrote:
> On Wednesday 23 August 2006 18:42, Daniel Iliev wrote:
>   
>> Jerry McBride wrote:
>>     
>>> Would some kind soul save me a bit of research time? Which of the two
>>> alternative init schemes are faster, initng or runit?
>>>
>>>
>>> Thank you in advance , Jerry
>>>       
>> I have tried initng several months ago. It rocks. It's several times
>> faster then the "normal" init. The problem at the time was there were no
>> scripts for everything I wanted to start automatically. So one day I
>> figured out that writing scripts and using faster init takes me more
>> time then using slower init which works with almost no maintenance. This
>> made me go back to the normal init. I have to say that while using
>> initng I noticed that many scripts were added for a relatively short
>> time. It is possible that now there are initng scripts for most of the
>> services one would ever use, but you have to check it out for yourself.
>>
>> I can't say a word about "runit", because it's the first time I read
>> about it.
>>
>> My next experiment for speeding the boot up will be fcache, but I'm
>> waiting for a proper mood to try it ( it means: "I'm too lazy" )
>>
>>     
>
> Hi Daniel,
>
> Fcache works, but we didn't see the performance boost that going to initng 
> gave. Since it requires it's very own ext3 partition to work, plus a kernel 
> patch... we dropped it.  
>
> Using initng is the ticket... Maybe the gentoo devs will directly support it 
> or a variant someday...
>
> Cheers, Jerry
>
>
>
>   
Thanks, Jerry

I appreciate this info about fcache. Now I have a good idea of what I
have to expect.
I've got some unpartitioned space and nothing prevents me to do some
tests. The only question that I have is how much space does fcache need?

BTW, the "ck-sources", which can be found in the portage comes with
several performance related patches. Emerge automatically applies these
patches and fcache is among them.

-- 
Best regards,
Daniel


-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] [OT] initng or runit?

[Jerry McBride]

On Wednesday 23 August 2006 21:28, Daniel Iliev wrote:

>
> I appreciate this info about fcache. Now I have a good idea of what I
> have to expect.

It's the reason for this mailling list. Enjoy.

> I've got some unpartitioned space and nothing prevents me to do some
> tests. The only question that I have is how much space does fcache need?
>

The way it works is... you cache the files that are normally loaded up to a 
point that you choose. We tried just the time from boot to a console and then 
again from boot to a fully loaded KDE desktop... A rule of thumb we followed 
would be what ever you have for ram, plus whatever you use for swap. About 
1gig for us. The number you end up with will be overkill, but you are assured 
of enough cache space for all your needed files.

Cheers.

> BTW, the "ck-sources", which can be found in the portage comes with
> several performance related patches. Emerge automatically applies these
> patches and fcache is among them.
>
> --
> Best regards,
> Daniel
-- 
gentoo-user@gentoo.org mailing list