[erlang-questions] Changing scheduler behaviour to make tests less deterministic
Fri Dec 29 17:55:18 CET 2006
Understood. Thanks for considering it.
In addition to a valuable testing tool, I suspect it could be an
interesting system-tuning parameter. The current value of 2000
reductions per quantum was presumably chosen after experiments with a
fixed set of applications, some time ago. (Didn't I read that it was
originally 1,000, years ago?)
So different applications, running on newer hardware and newer VMs
might benefit from a different value. (SMP might make a difference
too.) It will be interesting to try, at least.
On 12/29/06, Rickard Green <rickard.s.green@REDACTED> wrote:
> We'll consider it, but I'll give no promise. If we add it, it will most
> likely not make in into R11B-3.
> Rickard Green, Erlang/OTP
> Chris Newcombe wrote:
> > Cool. Would you consider adding maybe a '+Q nnnn' option to allow
> > setting CONTEXT_REDS dyamically? (The 'Q' would be mnemonic for
> > 'quantum' as in scheduler quantum of course.)
> > Many thanks,
> > Chris
> > On 12/27/06, Rickard Green <rickard.s.green@REDACTED> wrote:
> >> As of R11B-3 there will be a "modified timing" command line argument
> >> to erl (+T). It will be possible to modify the timing more or less.
> >> It will (most likely) do the following:
> >> * spawn/spawn_link/spawn_opt will schedule out the parent process
> >> and with "heavier modified timing" force the parent to sleep
> >> for a while. (Among the few SMP bugs we have found in our Erlang
> >> code, most of them have been bugs where the code depended on the
> >> parent doing some initialization before the child began
> >> executing.)
> >> * A fixed amount of reductions will be bumped as soon as a process
> >> is scheduled in.
> >> BR,
> >> Rickard Green, Erlang/OTP
More information about the erlang-questions