[Ecls-list] Status of MP code

Matthew Mondor mm_lists at pulsar-zone.net
Wed Apr 18 05:04:52 UTC 2012


On Sun, 15 Apr 2012 23:33:11 +0200
Juan Jose Garcia-Ripoll <juanjose.garciaripoll at googlemail.com> wrote:

> The code as it appears in the tree is considered feature complete. This
> means I do not want to introduce any further properties or arguments to the
> functions in the MP package and now the priority is shared among these
> goals:

Although we also communicated off-list, I wanted to confirm on-list
that I think great progress was made.  There appears to be potential
issues if using process-kill, but the POSIX API itself has to account
for cancellation points in its cancel interface...

If I configure the HTTPd to use a pool of threads that doesn't scale
down, it's more stable than the POSIX-based ECL locks ever were,
including my simplified implementation (which also had other issues
related to interrupts).

There are various possible solutions to avoid process-kill, including a
suggestion from Juan, which I intend to eventually try, such that the
pool could scale down safely.

Thanks,
-- 
Matt




More information about the ecl-devel mailing list