8 Apr
2005
8 Apr
'05
6:16 p.m.
The kernel developers have decided not to merge the
realtime-lsm,
after all. Instead, they propose an rlimits extension for granting
per-user realtime scheduling privileges. This does (barely) meet our
minimum needs.
It is inferior to the realtime-lsm solution for several reasons I feel
too tired and discouraged to repeat again here. (Those who care
should look the discussion up in the LKML archives.) It all smells
like NIH syndrome to me (Not Invented Here). Since their solution
won't be available for end-users until all the shells and PAM modules
have been updated and everyone has upgraded to new distributions, I'll
continue supporting the SourceForge realtime-lsm package for another
year or two, as long as we still need it.
While I agree that realtime-lsm may be best for short-term (actually
Con's patch works with nothing more than recompiling the kernel), I
wouldn't like it as a long-term solution. The main thing I don't like is
that whenever you screw during development, you have to reboot your
machine. At least the rlimits patch and Con's SCHED_ISO allow you to put
a cap on the amount of CPU. Unfortunately, it seems like Con's patch
can't made to work easily on a per-user basis, so it may be that Ingo's
rlimits is the best long-term solution. I have nothing against
realtime-lsm being merged, but I don't think it should stop there.
The main advantage I see to an rlimit approach is that it's a lot less
risky for distributions to use out of the box (as in you check "this is
a desktop and it becomes enabled). That combined with what looks like
complete opposition (rightly or wrongly) to realtime-lsm, I think it
would be more practical to push for rlimits, while also pushing on
distros to include it.
I came away dissatisfied with the whole experience.
There are a
number of very good Linux kernel developers, but they tend to get
outshouted by a large crowd of arrogant fools. Trying to communicate
user requirements to these people is a waste of time. They are much
too "intelligent" to listen to lesser mortals.
As far as I see it, we'll at least get listened to by Con, Ingo and
Andrew Morton. I've had a long discussion with Con recently and from his
point of view, the problem is that not enough people ask (loud enough)
for such features. For instance, I'm still the first and only user of
his real-time patch.
Jean-Marc
--
Jean-Marc Valin <Jean-Marc.Valin(a)USherbrooke.ca>
Université de Sherbrooke