> OK. It seems that a prominant companies field rep has read some of the
> linux FUD and believed it:
>
> > My limited understanding of Linux is:
> >
> > A. The kernel doesn't implement kernel-mode threads.
That has been a lie since 1.3 (and possibly even 1.2?).
> > B. With Linux, the scheduler can't preempt the kernel,
Yes, but
> > which is
> > needed for the scheduler to divide CPU time among the threads.
This is staggeringly untrue. And not even Linux related.
> > C. Lastly, with Linux, the kernel is not reentrant. In other
> > words, no matter how many processors are in a system, only one processor
> > can execute code at a time.
I'll forget that he probably meant "can execute kernel code" at a time.
Even that's not true.
A lot of kernel calls are still serialised (the read and write paths, for
example), but work to reduce this is proceeding and, as a lot of people
forget, Linux has such low latencies for syscalls (typically around 1/10th
of those for (early?) Solaris/86, I believe) that for many common cases,
the amount of time spent holding the kernel lock is tiny.
Of course, things will have to improve before 64 CPU Linux boxes can scale
properly on io-bound stuff..
Matthew.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/