Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 3 Jun 2003 17:48:43 +0100
From:      Paul Robinson <paul@iconoplex.co.uk>
To:        Sam Leffler <sam@errno.com>
Cc:        Andrew Gallatin <gallatin@cs.duke.edu>
Subject:   Re: Making a dynamically-linked root
Message-ID:  <20030603164843.GB29331@iconoplex.co.uk>
In-Reply-To: <093601c329ee$24fe0b90$52557f42@errno.com>
References:  <20030603122226.BGPM11703.pop018.verizon.net@kokeb.ambesa.net> <093601c329ee$24fe0b90$52557f42@errno.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Jun 03, 2003 at 09:34:59AM -0700, Sam Leffler wrote:

> Gordon posted boot-time numbers because I prodded him about not committing
> the changes until he had a handle on the performance implications.  The time
> for a system to reach the "login prompt" was one criteria for some companies
> I watched go through the same exercise (I also suggested some other tests
> for which I haven't seen results).  Mind you they were not talking about a
> diskless boot to "login:" but rather booting into a GUI environment where a
> lot of applications run during startup.

And they clocked performance from boot time? Weeeeeee!

OK, since my last post (3 minutes ago) I've thought about this. Is there not
a way of just adding sendmail_enable_background as well as sendmail_enable,
and if set to YES the command is run with &, otherwise without? With the
default behaviour being with? Do this for everything else that rc.conf has
an enable that executes an external program/script? That's like, an hours
work across the tree, possible backing out of some with strange dependancies
once people complain and raise a few PRs, and then we can go on about how
much FreeBSD r0x0rs when it's been DoS'ed? I'm being serious, I think, but I 
think on a box with sendmail, dns, apache, mysql, inetd, sshd, etc... you 
could shave 20 seconds off boot time on a slower box.

> The point, regardless, was that blindly making these changes while we are
> still trying to resolve basic system performance issues is not a great idea.

Ahhh, we're back into dynamic-linking again. I think you're going to find 
two clear opinions on this:

1. Those people who cut their teeth in the old days with the guiding hand of 
a bearded man in sandals telling you what is "good" and what is "bad", 
insisting that root's shell should always be a statically-linked /bin/sh and 
everything you need to get the box up to single-user should be statically 
linked

2. Those who didn't who are prepared to "compete" with other Unixes, even if
the decisions they've made may be wrong, incoherent, or even (shock!) based
on inexperience and a naieve belief that they will be considered "better"  
because they're "faster" rather than being "better" because they're more
"stable" or "secure" or "bring-back-from-the-dead-able".

Which one FBSD goes for is open vote, and has a ring of bikeshed about it, 
but it's important. It could have major consequences.

Of course, there is a flag to make buildworld already that makes everything 
dynamic...

> netbsd recently switched to a dynamically-linked root and before committing
> to the change they devoted a bunch of effort into improving the performance
> of their dll runtime.

NetBSD are also interested in lots of embedded systems where bootstrap time 
isn't just nice to keep low, it's essential. 

I don't even have a commit bit, so don't listen to me. I can't even make the 
changes I'm suggesting as a stop-gap, but I thought I'd whack my 2p in.

-- 
Paul Robinson



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030603164843.GB29331>