Am I way off base here?

Garance A Drosihn drosih at rpi.edu
Tue Nov 25 17:16:35 PST 2003


At 6:49 PM -0500 11/25/03, Skip Ford wrote:

 I have my
 own good ideas which I'd like to see implemented which
 would address the performance issues.
Good luck.

If my opinion above is correct, you'll be given a really
hard time if you try to undo dynamic root regardless of
the validity of your argument.  If you're not, I'm wrong.
My goal is to explore speed improvements and caching for
dynamically-loaded binaries.  We will never want to have
all binaries statically-loaded, so why should we get so
hung up on statically-loaded executables for performance?
*If* my ideas work out, then we will see much less of a
performance penalty for dynamically-loaded binaries.  So
if my argument is valid, then we'll have much less of a
need for statically-linked anything.  We will also have
the welcome ability to dramatically speed up *any*
program which is frequently run.  Not just statically-
linked programs.
The trick is finding the time to implement any of this,
given that I'm just a "weekend warrior" when it comes to
writing/debugging code for *BSD.
Another path we (FreeBSD) need to follow up on is that
NetBSD has made this same change, and ran into these same
performance issues.  Instead of falling back to a statically-
linked /bin, I am told that they optimized the loader so
that *it* performs faster.  Again, this is a change that
benefits *every* program (*), not just the statically-linked
ones.  I prefer that kind of change.
(* - well, hopefully it does.  I don't know what specific
changes they made, but I think we should look into them
instead of just throwing up our hands and saying "We must
not ever consider dynamically-loading a popular program").
These are just my opinions.  They could be all wet, in which
case I do not mind if /bin/sh is statically linked for
5.3-release.  Note: 5.3-release, not 5.2-release.  For the
past few years I have lived through all kinds of pain while
tracking freebsd-current, and I'm told to live with it
because this is the "bleeding-edge" branch.  This change
is a very mild experiment to try, compared to all the
other things we've gone through.  It's annoying that we
can't even try some simple experiment like this without
getting all kinds of hostility.
I want a dynamically-linked /bin/sh simply to see what
operational problems might come up with that.  All I
want to do is gather more information.  Once we have
that additional information (from all of the above
experiments), we can make a smarter decision.
 > It should also be noted that Matt nearly got voted to
 > a position on FreeBSD's core team, so it is pretty
 obnoxiously unfair to paint all FreeBSD developers
 with any particular coat of paint.
Only Wes Peters has stepped forward to say that DragonFly
ideas are welcome on the FreeBSD lists.  Everyone else
either says to take it elsewhere or says nothing in
response to those posts which is just as bad.
Well, Matt tends to flare up at times, and I'm not sure I
want to encourage him too much...   :-)
That said, I again resent your remarks, as I have certainly
have no problem with any of Matt's ideas.  There are several
freebsd committers on this mailing list, and we can be
enthusiastic about what Matt is doing here without having
to get into the middle of fights on FreeBSD lists.
I had fought many times on the freebsd lists to back Matt
while he was a committer.  I have done that for years now.
Now that there is the split, I just feel my time is better
spent encouraging *everyone* to pursue their own good ideas,
instead of trying to win little personality wars that little
minds get wrapped up in.  If you don't like what some other
group is doing, then just ignore it.  Or "do it right" in
your own project.  There's no need in getting pissed off
at them for what they want to do.  The amount of time and
energy that you *will* waste is worth much more than anything
you can possibly gain from these pissing matches.
For what it's worth, of all the freebsd developers that I
have worked with, there are only three who I like more than
Matt.  And all three of them are guys who rarely get into
any of these blazing flame-wars with Matt (or with much of
anyone else, for that matter).
--
Garance Alistair Drosehn            =   gad at xxxxxxxxxxxxxxxxxxxx
Senior Systems Programmer           or  gad at xxxxxxxxxxx
Rensselaer Polytechnic Institute    or  drosih at xxxxxxx




More information about the Kernel mailing list