This series pulls (applied on top of xlate_v2) miss handling into it's own file with clean state semantics necessary for future multithreading. Both patches add some additional overhead which I expect will have a minor negative impact on performance in the short term. The hope is that enabling multiple threads will make up for the impact. Of course, once the full multithreaded architecture is in place, we will be able to due further optimizations further reducing the impact.
Ethan Jackson (2): ofproto-dpif: Remove pointers between rules and facets. ofproto-dpif: Factor out flow miss handling. ofproto/automake.mk | 2 + ofproto/ofproto-dpif-upcall.c | 612 ++++++++++++++++++++++++++++++++++ ofproto/ofproto-dpif-upcall.h | 110 ++++++ ofproto/ofproto-dpif-xlate.h | 4 +- ofproto/ofproto-dpif.c | 740 +++++++++-------------------------------- ofproto/ofproto-dpif.h | 12 +- 6 files changed, 901 insertions(+), 579 deletions(-) create mode 100644 ofproto/ofproto-dpif-upcall.c create mode 100644 ofproto/ofproto-dpif-upcall.h -- 1.7.9.5 _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
