[CONSPIRE] Installing software from source on Linux?

Rick Moen rick at linuxmafia.com
Fri Oct 28 12:34:47 PDT 2005

Quoting Dan Bikle (dan.bikle at gmail.com):

> Question: if you had a choice between FreeBSD 5.3 and 5.4
> what would you pick?

Well, you might hate me a bit for saying this, but I'd go with "Neither
one:  Use Dragonfly BSD, instead."  (http://www.dragonflybsd.org/)

And, yes, that means you're about to get subjected to another treatise
from yr. humble servant about *ix lore -- sorry about that.  Also,
obligatory disclaimer:  If accused of being an ignorant bystander of BSD
affairs, I will readily confess to guilt on that score:  I'm just a
sympathetic guy who knows a bit about the subject and talks to a bunch
of people.  Don't take anything I say on the subject as true for no
better reason than my saying it.

Back in FreeBSD 4.x days (around 2003), Jordan Hubbard and a lot of
other core FreeBSD committers (developers with general CVS commit
access) were proposing a massive restructuring of the kernel, intending
to improve the handling of threading and SMP.  This new, and very
complex new codebase became the core feature innovation in FreeBSD 5.x.
Unfortunately, there was extremely strong disagreement over the merits
of the new architecture: A large contingent of core committers lead by
Matt Dillon considered it a strategic mistake:  They felt that the newer
threading and SMP models introduced in 5.x was buggy and performed
poorly.  (The new threading model is the one you'll see referred to as
"KSE" = Kernel Scheduling Entities.)

Reading between the lines (though I might be mistaken), I'm gathering
that they felt that the overly baroque 5.x thread model was a
misbegotten sop to the huge mountain of badly written and thread-crazy
Java code that had suddenly become significant to Jordan and some other
5.x-proponent committers, because they had jumped ship from places like
sinking FreeBSD corporate sponsor Wind River, and had taken on
employment at Apple Computer, which for various reasons had gone heavily
into Java for Enterprise Object Framework / WebObjects and other major
developer products.

Basically, I agree with Matt Dillon, and reports from BSD-land, although
infused with the controversy and perpetual flamewars that have been *BSD's 
hallmark from the very beginning, suggest that Dragonfly BSD, which
forked from the rock-solid FreeBSD _4.x_ codebase and continues its
development, has absolutely proved his point.

I believe you mentioned that your ex-Whitebox server is an Internet
host:  That would be exactly the sort of machine I would expect to
benefit _least_ from FreeBSD 5.x's alleged advances over 4.x (and thus
over Dragonfly BSD) in SMP and handling of Java-scale massive threading.
But you would know your usage case better than I would, so you should be
the judge.

(If you're a Java fan, you might consider me a bit harsh on the subject.
Oh well.  ;->  )

Between FreeBSD 5.3 and 5.4, to answer your question directly, I would
definitely go with 5.4.  (The nice thing though is, you can install 5.3,
see if you like it, and then make up your mind if you'd like to upgrade
to 5.4.  If you do, just refresh the ports skeleton and type "make
world" at the top of the ports tree.  Wait a couple of days for the
compiling to stop, and you're at 5.4.)

Why 5.4?  Quite simply because they declared it the "STABLE" branch,
replacing 5.3, which was the first 5.x release to be granted that tag.
Given that you're buying into the 5.x architecture (which I personally
wouldn't), then it seems fair to take the release managers' word that
5.4 is ready.

> Then I entered a ticket to wipe my whitebox clean and
> put FreeBSD on there.

Not intending to sound critical, but getting used to the administrative
and maintenance regimes of _any_ *ix takes a little experimentation and
patience.  You can't judge a BSD or Linux distribution fairly on the
basis of a day or two of playing around, because you will not know
enough to use it properly.  

My own preference, Debian GNU/Linux, was a real bear to get used to,
because of the maintainers' maddening habit of creating cool bits of
architecture and then not adequately (in my view) documenting them.  I
stuck with it because I intuited that there was long-term satisfaction
to be found there once I grasped the "Debian way", and in my view I
turned out to be right.

For other people, CentOS, FreeBSD, Gentoo Linux, Ubuntu Linux, or any
number of others will work for them.

I don't know if that's useful to you, but it's my honest take on things.

More information about the conspire mailing list