Thanks for the comments. We will move the evaluation results to commit message.
It also makes sense to move CD into another file, we will try the new implementation in V3. Meanwhile, once DPDK17.11 supported, we could use DPDK's new "membership library" to implement CD, which we considered as part of the future version to include. Thanks > -----Original Message----- > From: Ben Pfaff [mailto:[email protected]] > Sent: Friday, November 3, 2017 11:00 AM > To: Wang, Yipeng1 <[email protected]> > Cc: [email protected] > Subject: Re: [ovs-dev] [PATCH v2 1/5] dpif-netdev: Basic CD feature with > scalar lookup. > > On Tue, Oct 31, 2017 at 04:39:33PM -0700, Yipeng Wang wrote: > > Cuckoo distributor (CD) is a double-hash function hash table, that helps > > redirect packets to their corresponding subtables to avoid the sequential > > search of megaflow subtables. This is another layer of cache to cache flows > > and their corresponding subtable indexes. Different from a hash table, CD > > can have certain false positive rate (since the full key is not stored for > > space efficiency). Our CD design was partially inspired by earlier concepts > > proposed in "simTable"[1] and "Cuckoo Filter"[2], and DPDK's cuckoo hash > > implementation. > > > > For the current implementation, the design does not allow displacing items > > when a bucket is full, which is different from the behavior of a cuckoo hash > > table. The advantage is that we do not need to store two signatures so > that > > the struct is more compact. We use 16 entries per bucket for the > > convenience of vector lookup. > > > > Each classifier has its own cuckoo distributor. > > > > [1] H. Lee and B. Lee, Approaches for improving tuple space search-based > > table lookup, ICTC '15 > > [2] B. Fan, D. G. Andersen, M. Kaminsky, and M. D. Mitzenmacher, > > Cuckoo Filter: Practically Better Than Bloom, CoNEXT '14 > > > > CC: Darrell Ball <dball at vmware.com> > > CC: Jan Scheurich <jan.scheurich at ericsson.com> > > Signed-off-by: Yipeng Wang <yipeng1.wang at intel.com> > > Signed-off-by: Antonio Fischetti <antonio.fischetti at intel.com> > > Co-authored-by: Antonio Fischetti <antonio.fischetti at intel.com> > > --- > > Evaluation: > > We create set of rules with various src IP. We feed traffic containing 1M > > flows with various src IP and dst IP. All the flows hit 10/20/30 > > rules creating 10/20/30 subtables. > > > > The table below shows the preliminary continuous testing results (full line > > speed test) we collected with a uni-directional phy-to-phy setup. OvS > > runs with 1 PMD. We use Spirent as the hardware traffic generator. > > > > Scalar CD results: > > 1M flows: > > no.subtable: 10 20 30 > > cd-ovs 3658328 3028111 2863329 > > orig_ovs 2683455 1646227 1240501 > > speedup 1.36x 1.84x 2.31x > > Thank you for the evaluation results. I think that they should be > included in the commit message. They are helpful in looking through the > commits later. > > Did you consider adding the cuckoo distributor as part of a new source > file? dpif-netdev is very big, which makes it harder to understand. _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
