On Sat, Sep 29, 2012 at 12:17 AM, Giuseppe Lettieri <g.letti...@iet.unipi.it> wrote: > > Il giorno 29/set/2012, alle ore 01:44, Jesse Gross <je...@nicira.com> ha > scritto: > >> Does this actually require changing the main loop? If it really is >> the poll related changes that you mention it seems like they could be >> contained to the netdevs and userspace datapath. You could handle a >> batch of packets each time through the main loop. > > > Well, I see what you mean. Indeed, it was me that raised the issue on the > list. But the poll related changes I mentioned where just an hypothesis. Take > the batching issue you are referring to, for example: if I disable batching > in the threaded datapath (just put batch=1), performance drops by ~20% only. > From this I get that the big problem with the existing loop lies elsewhere. I > am not saying that it really needs drastic changes, only that it might, and > we had no time to investigate. This is only in reply to your question of why > we continued down this road. > > I understand that our lack of time is no good reason to accept a patch. On > the other end, I tried to push it anyway since it solves a real problem for > FreeBSD people, maybe also PlanetLab users, and possibly others.
Thanks, I definitely understand trying to get something in now even if there are potentially more optimizations that can be made in the future. As I mentioned in the other message, processes seem a lot simpler to me as an immediate solution (with hopefully known performance characteristics) and maybe we can get some of the other optimizations to the main loop in the future. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
