rc improvements (wanted?)

[Alexander Leidinger]

Quoting Peter Jeremy <peterjeremy at optushome.com.au> (from Fri, 18 Jul  
2008 17:18:07 +1000):

> On 2008-Jul-18 08:37:25 +0200, Alexander Leidinger  
> <Alexander at Leidinger.net> wrote:
>> Are you aware that the parallel starting in Solaris 10 reduced the
>> booting time by a nice percentage?
>
> Given that Solaris boots in geologic time, this probably wouldn't
> be difficult.

How do you define "booting Solaris"? Do you include the extensive  
tests prior to loading the kernel into this? I'm not talking about the  
time a 25k needs (even when you reducing the amount of testing on the  
system controller, it takes a long while until it reaches a state  
which I would call the start of the boot of the OS). We are talking  
about the pure userland part of booting. What is done during the  
startup of important programs in Solaris is not unreasonable (and  
similar/comparable between Solaris versions), and still, there's a  
nice difference between Solaris 9 and 10 if you count the time until  
you can start to do useful stuff.

>> If yes, do you expect that FreeBSD
>> behaves significantly different or do you "just" want to see numbers?
>
> Parallel starting is not guaranteed to be an improvement.  Starting a
> whole pile of processes that are I/O bound during initialisation
> (think squid or some databases) may be worse than starting them one
> at a time.  Likewise, a whole pile of processes that are CPU bound

It depends, think about independent disks and or keeping the squid  
data in RAM (e.g. tmpfs).

But this doesn't matter, we will always be able to come up with  
situations where the parallel start is not a good idea. We don't come  
by default with such a situation and I'm sure a lot of configs out  
there that don't fall into this class. Based upon your argument we  
could say we can not enable parallel starting even if we see it is an  
improvement for the reboot after the installation.

What I wanted to know is if there's an substantial argument (it can  
not behave similar to Solaris, because of A and B), or if he "just"  
wants to know what the difference on FreeBSD is.

> will just thrash the scheduler.  (Though parallel starting of I/O and
> CPU bound processes should be a win).

You forgot about round-trip-time bound processes (basically processes  
which wait for an event to occur before they say they are successfully  
started), and we have several of them.

Bye,
Alexander.

-- 
Those who hate and fight must stop themselves -- otherwise it is not
stopped.
		-- Spock, "Day of the Dove", stardate unknown

http://www.Leidinger.net    Alexander @ Leidinger.net: PGP ID = B0063FE7
http://www.FreeBSD.org       netchild @ FreeBSD.org  : PGP ID = 72077137