Re: [Caml-list] Thread behaviour

[Gabriel Kerneis <gabriel@kerneis.info>]

On Mon, Sep 30, 2013 at 04:12:23PM +0100, Tom Ridge wrote:
> > It all depends on the whim of the OS scheduler.  OCaml has no control
> > over it.  And you shoudn't expect any kind of fairness from the OS
> > scheduler, esp. Linux's, which gladly jettisons any pretense of
> > fairness in the hope of getting better throughput.
> 
> Ah! You are saying that the problem (maybe) lies with the Linux scheduler.
> This had never occurred to me. Probably because I assumed the Linux
> phrase "completely fair scheduler" meant something (although
> admittedly, I never tried to find out what).

A bit of off-topic Linux history for those interested:

The behaviour of sched_yield() changed in Linux 2.6.23 (with the
introduction of the CFS), to adopt a semantics where it does not yield,
but only reduces the priority of the calling thread.  This change has
been debated a lot, in particular because it broke many Java
applications.  As a work-around, a kernel variable (sched_compat_yield)
was introduced to optionnally recover the original behaviour; the
rationale for the change was that sched_yield is arguably ill-specified
and programs should not rely on it but use (non-portable and tricky)
futexes, or (portable) mutexes instead.

The variable sched_compat_yield has been remove in Linux 2.6.39, with no
clear rationale, during one of the many refactoring of the CFS.

As a matter of fact, this change also broke OCaml scheduling.  Nowadays,
sched_yield() is disabled in the ocaml runtime on Linux.  Have a look at
http://caml.inria.fr/mantis/view.php?id=2663 for more details on the
caml-side of the story (in French).

-- 
Gabriel