On Jun 2, 2014, at 8:50 PM, YAMAMOTO Takashi <yamam...@valinux.co.jp> wrote:
>> >> On Jun 2, 2014, at 12:51 PM, Ben Pfaff <b...@nicira.com> wrote: >> >>> On Mon, Jun 02, 2014 at 12:50:45PM -0700, Jarno Rajahalme wrote: >>>> >>>> On Jun 2, 2014, at 11:45 AM, Ben Pfaff <b...@nicira.com> wrote: >>>> >>>>> I'd really hope that any serious OVS implementation would be able to >>>>> use C11 or GCC or Clang or other compiler-specific techniques to get a >>>>> "real" implementation of atomics. Really the pthreads version is just >>>>> to make porting easier. >>>> >>>> Are we already documenting this somewhere? >>> >>> I don't think so. It wasn't the attitude I started out with; >>> originally I hoped that the pthreads implementation would be viable >>> for real use, but I'm really not sure about that any longer. >> >> I just ran \x93time ./ovstest test-cmap check 1\x94, with >> ovs-atomics-pthreads it takes 16 times longer than on master. So it is >> likely unusable in practice, so this would be the time to document it, I >> guess. > > while i guess test-cmap is far from the real usage, > documenting expectations is a good idea. > > "ovs-atomic-pthreads implementation is provided for portability. > It might be too slow for real use because Open vSwitch is > optimized for platforms where real atomic ops are available.” Thanks, I just sent a patch incorporating this into a comment and adding a more generic note in INSTALL. Jarno _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
