> -----Original Message----- > From: discuss [mailto:discuss-boun...@openvswitch.org] On Behalf Of Benedikt > Rudolph > Sent: Wednesday, October 21, 2015 4:37 PM > To: disc...@openvswitch.com > Subject: [ovs-discuss] ovs dpdk low performance despite optimizations > > Hi all, > in our test scenario we observe comparably low packet forwarding > performance on a dpdk (2.0) enabled ovs (2.4) switch. > The configuration is a single bridge (br0) with two dpdk enabled > interfaces (10 Gbit/s each) as per the Install.DPDK.md example. > We followed *thoroughly* all performance related optimization steps from > the above document.
Did you catch the updates to that section on the mainline that were not in the OVS 2.4 package? https://github.com/openvswitch/ovs/blob/master/INSTALL.DPDK.md#performance-tuning > > Nonetheless the best rate we could get is ca. 2 M packets/s on each > interface, which equals to about ca. 1.3 Gbit/s for 64byte sized packets. > Our comparison is against the “l2fwd” example dpdk application. Using > “l2fwd“ we can easily achieve 13 M packets /s in the same test > configuration. The only indicator of something going wrong is a high > count of ”errs“ packets in the output of “ovs-ofctl dump-ports”. > > In our first runs (using 2 pmd threads) we noticed a high pmd thread > utilization as reported by the $ovs-appctl dpif-netdev/pmd-stats-show > command. We made sure to clear the statistics after every observation to > get an accurate report for the next interval. To mitigate this we > allowed dpdk to use up to 8 cores on a 12 core (24 ht threads) cpu. We > also set “other_config:n-dpdk-rxqs=20”. Now we observed “idle“ pmd > threads but still the packet forwarding performance is limited to the > above values. > We also experimented with the amount of huge pages memory. > > Do you know of any *recent* reports on OVS dpdk performance? I found > some papers on the old dpdk-vswitch versiont which claim much higher > pkt. counts than I observed. > > What further debugging steps can I make? Below I include the output of > “ovs-ofctl dump-ports br0" and “ovs-ofctl show br0" > Note that the output of says „Max Speed 1000” despite having 10 Gbit > Interfaces attached. I am not sure wether this is an error. I see the same, so not causing your poor performance. > > > $ovs-ofctl dump-ports br0 > > OFPST_PORT reply (xid=0x2): 3 ports > > port 2: rx pkts=5, bytes=925, drop=0, errs=0, frame=0, over=0, crc=0 > > tx pkts=47128219, bytes=3016206016, drop=0, errs=0, coll=0 > > port 1: rx pkts=47130239, bytes=3016335901, drop=0, errs=84435109, > > frame=0, over=0, crc=0 > > tx pkts=0, bytes=0, drop=0, errs=0, coll=0 > > port LOCAL: rx pkts=0, bytes=0, drop=0, errs=0, frame=0, over=0, crc=0 > > tx pkts=47128251, bytes=2827695060, drop=47128251, errs=0, > > coll=0 > > > $ovs-ofctl show br0 > > OFPT_FEATURES_REPLY (xid=0x2): dpid:00005cb9018f16e0 > > n_tables:254, n_buffers:256 > > capabilities: FLOW_STATS TABLE_STATS PORT_STATS QUEUE_STATS ARP_MATCH_IP > > actions: output enqueue set_vlan_vid set_vlan_pcp strip_vlan > > mod_dl_src mod_dl_dst mod_nw_src mod_nw_dst mod_nw_tos mod_tp_src > > mod_tp_dst > > 1(dpdk0): addr:5c:b9:01:xx:xx:xx > > config: 0 > > state: 0 > > current: 10GB-FD > > supported: 100MB-FD 1GB-HD 1GB-FD FIBER AUTO_NEG AUTO_PAUSE > > AUTO_PAUSE_ASYM > > speed: 10000 Mbps now, 1000 Mbps max > > 2(dpdk1): addr:5c:b9:01:xx:xx:xx > > config: 0 > > state: 0 > > current: 10GB-FD > > supported: 100MB-FD 1GB-HD 1GB-FD FIBER AUTO_NEG AUTO_PAUSE > > AUTO_PAUSE_ASYM > > speed: 10000 Mbps now, 1000 Mbps max > > LOCAL(br0): addr:5c:b9:01:xx:xx:xx > > config: PORT_DOWN > > state: LINK_DOWN > > current: 10MB-FD COPPER > > speed: 10 Mbps now, 0 Mbps max > > OFPT_GET_CONFIG_REPLY (xid=0x4): frags=normal miss_send_len=0 > > _______________________________________________ > discuss mailing list > discuss@openvswitch.org > http://openvswitch.org/mailman/listinfo/discuss _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
