Re: revivesa status 2008/07/09

[buhrow%lothlorien.nfbcal.org@localhost (Brian Buhrow)]

        As a NetBSD-user since 0.8 days, and as one who still runs 0.9
binaries, I think merging the SA code back into 5.x is a good thing,
especially if it can be done in a modular fashion, as Bill has been doing.
I know that, in theory, you can just copy libthread.so from the 5.x branch
into the 4.x binary tree that you're running at the time you do the
upgrade, but NetBSD has never required that level of file manipulation to
do an upgrade in the past, and I think it would be a bad precedent to
change.  Putting the SA code back into the 5.x branch, continues NetBSD's
long tradition of backward compatibility, and keeps the behavior of least
astonishment that long time NetBSD users have come to know and appreciate.
Given that we're the smallest of all the BSD camps, and given that Bill has
done most of the work necessary to make it all "just work", I think it best
to not introduce a potentially user-losing change. New users won't use SA,
and that's fine.  Those of us who are using production critical SA code,
should be able to upgrade without having to worry overly much about whether
or not the SA code is going to function.  Sometimes, in production
environments, being able to run back reved binaries is absolutely the right
thing to do, even if it's just for a while.
-Brian
On Jul 13,  1:12pm, Bill Stouder-Studenmund wrote:
} Subject: Re: revivesa status 2008/07/09
} 
} --Kj7319i9nmIyA2yE
} Content-Type: text/plain; charset=us-ascii
} Content-Disposition: inline
} Content-Transfer-Encoding: quoted-printable
} 
} On Sat, Jul 12, 2008 at 11:41:43AM +0100, Mindaugas Rasiukevicius wrote:
} > Thor Lancelot Simon <tls%rek.tjls.com@localhost> wrote:
} > Perhaps you can give us a link to that benchmark which would show how SA
} > performs better than 1:1? There were many benchmarks which proven 1:1
} > performance, but unfortunately I have not seen any benchmark where SA
} > performs better than 1:1 on NetBSD.
} 
} Why?
} 
} Would you really rip out 1:1 threading if someone handed you said=20
} benchmark? I doubt it, and if you were about to I'd try to talk you out of=
} =20
} it.
} 
} Benchmarking is great for selecting ONE solution. But I'm unaware of=20
} anyone suggesting _replacing_ 1:1 threading with SA. The suggestion is to=
} =20
} have both.
} 
} Actually, the suggestion is to restore the level of system call backwards=
} =20
} compatability that we have had with most every other system call since=20
} NetBSD 1.0.
} 
} > It is easy to say "well-understood", "obvious", "saw". But perhaps we can
} > be more constructive by providing numbers, test-applications, or something
} > like that?
} 
} But to what end? The only thing that I can see that would benefit would
} either be a discussion about if we should remove 1:1 threading in favor of
} SA (a discussion no one is having) or a discussion as to which one should
} be the default threading. Given that ad is kicking butt with his
} benchmarks that are using our 1:1 threading, I see no reason to not ship
} the OS so that, out of the box, it reproduces hist test cases reasonably
} well.
} 
} The question is should we restore support for system calls we supported in=
} =20
} NetBSD 4.0 and NetBSD 3.0. We removed support because, well, because the=20
} code needed more love than we had people interested in giving it. That was=
} =20
} the right thing to do.
} 
} The question now is NetBSD better off merging the revived code in or not.
} 
} Take care,
} 
} Bill
} 
} --Kj7319i9nmIyA2yE
} Content-Type: application/pgp-signature
} Content-Disposition: inline
} 
} -----BEGIN PGP SIGNATURE-----
} Version: GnuPG v1.4.7 (NetBSD)
} 
} iD8DBQFIemGTWz+3JHUci9cRArjkAJ9RN17IwbjXFWCcLEnWvduzC6HX7gCfXsNt
} 6FVrkcjPc/DMk+lr3vZMNJU=
} =M0gy
} -----END PGP SIGNATURE-----
} 
} --Kj7319i9nmIyA2yE--
>-- End of excerpt from Bill Stouder-Studenmund