[Chicken-users] starvation in i/o intensive situations

[F. Wittenberger]

Hi Felix & All,

I observed a starvation of some threads, when there's concurrent i/o
going on.

The attached patch improves fairness in most cases by eating up the
(new) current-ready-queue before refilling it.  But it does obviously
not fix 100% of possible cases of the starvation problem.

I have no environment to quantify the improvement.  But the patch avoids
some needless select(2) handling and improves responsiveness to an
extend that I can "feel it in the mouse".

Unfortunately I'm unable to reproduce the starvation without handling
bad fd's too.  Therefor I must regret to include that change in the
attached patch.

best regards

/Jörg

sched.patch
Description: Text Data