On Wed, May 11, 2011 at 01:52:21PM -0700, Ethan Jackson wrote: > Curious about the history of it, is there any reason we used to report > there were two tables?
Yes. Before the latest rewrite of the classifier, performance of interesting OpenFlow rulesets was basically O(n) in the number of flows. The one kind of rule that could be optimized in an obvious way was exact-match rules, so we kept them in a separate table. This lack of a good classification algorithm even made its way into the OpenFlow 1.0 standard, which required exact-match rules to have a priority higher than any rule with wildcards exactly to allow software switches to avoid expensive linear searches through a classifier. I'm glad we have a better algorithm now. _______________________________________________ dev mailing list [email protected] http://openvswitch.org/mailman/listinfo/dev
