Subject: Re: Please Revert newlock2
To: Bucky Katz <firstname.lastname@example.org>
From: Lars Heidieker <email@example.com>
Date: 02/21/2007 21:04:46
-----BEGIN PGP SIGNED MESSAGE-----
On 21 Feb 2007, at 19:23, Bucky Katz wrote:
> Jason Thorpe <firstname.lastname@example.org> writes:
>> On Feb 20, 2007, at 7:42 PM, Bucky Katz wrote:
>>> Yes. SA does not equal M:N threads, and as I've said elsewhere, if
>>> it were my system, I'd burn down SA but salvage M:N through a
>>> different mechanism.
>> ...and I think there have been discussions about doing so in NetBSD
>> in the future. I think Matt Thomas even spoke to this earlier in
>> this thread.
> Sure, but we've got "works now" followed by "long uncertain period"
> followed by "might work again".
> 4 users out of 5 when surveyed prefer that existing functionality not
> be broken until such time as its replacement is available. (The 5th
> didn't return his survey.)
Maybe I am missing the point, but from a user perspective you would =20
want descent threading, scalable over multiply cpus,
this is what the 1:1 threading approach can deliver but here the m:n =20
failed and it's unlikely to be fixed because off the complexity.
So from a user perspective the 1:1 approach delivers definitely more.
So why not implement (and that should be done by someone who needs =20
it) a m:n libpthread that does not need kernel support,
for the very remote applications that perform better with m:n as in =20
most cases 1:1 (especially with an O(1) kernel scheduler)
performs simply better (priority inversion with m:n) and yes the =20
argument of kernel resources does not bring you anywhere today
as main memory sizes have increased in a way that the kernel stack =20
overhead is in most cases not really important and
by using interrupt stacks, the stack size can be reduced further.
This is not to forget that sa and the other m:n stuff was mainly =20
inspired by reducing the kernel resource consumption.
Die mystischen Erkl=E4rungen gelten f=FCr tief;
die Wahrheit ist, dass sie noch nicht einmal oberfl=E4chlich sind.
-- Friedrich Nietzsche
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)
-----END PGP SIGNATURE-----