Re: >256 fd patch...

Darren Senn (
Sat, 22 Mar 1997 13:27:11 -0800 (PST)

Ted said:
> This is still false optimization. If someone is polling for data, the
> amount of time taken to do the context switch is *so* *large* that the
> time to do the extra two comparisons is lost in the noise.

Oh, granted, on both counts. I was merely arguing against the assumption
that select() wouldn't be used in that context.

> It has nothing to do with select()'s interface as much as it has to do
> with generic system call overhead. As far as your claim that there's a
> quite a body of code that does it, what programs do this? They really
> are quite broken.....

Real time applications mostly. Two reasons: it is more important that a
real time app take the same amount of time in every pass through its innermost
code than it is for the average time to be small, and many real time os's
have much less system call overhead than Linux does (read: no context switch).
Now I know that the whole real time os thing is an old argument that's been
beat to death in the Linux community, and is a sore point with a lot of
people ("Linux isn't real time!") but my point stands: select() gets used
in inner loops.