None of the OpenFlow flows that you've shown have any hits, so it's hard to say anything meaningful about them. Do you have very precise flows that are matching L4 ports or using balance-tcp, which could make bad megaflows? Can you send the a hundred or so flows from "ovs-dpctl dump-flows"?
--Justin > On Dec 15, 2014, at 9:57 PM, 李伟 <[email protected]> wrote: > > Hi ,thank you for your prompt reply. > > [@sh5_44.1 ~]# ovs-dpctl show > system@ovs-system: > lookups: hit:1543483249 missed:63722785 lost:0 > flows: 56469 > masks: hit:6624457650 total:5 hit/pkt:4.12 > port 0: ovs-system (internal) > port 1: em1 > port 2: br0 (internal) > port 3: veth_44_124 > port 4: veth_44_14 > port 5: veth_44_182 > port 6: veth_44_189 > port 7: veth_44_19 > port 8: veth_44_190 > port 9: veth_44_192 > port 10: veth_44_197 > port 11: veth_44_198 > port 12: veth_44_199 > port 13: veth_44_20 > port 14: veth_44_200 > port 15: veth_44_201 > port 16: veth_44_207 > port 17: veth_44_205 > port 18: veth_44_208 > port 19: veth_44_209 > port 20: veth_44_21 > port 21: veth_44_210 > port 22: veth_44_211 > port 23: veth_44_212 > port 24: veth_44_213 > port 25: veth_44_214 > port 26: veth_44_216 > port 27: veth_44_217 > port 28: veth_44_218 > port 29: veth_44_219 > port 30: veth_44_22 > port 31: veth_44_220 > port 32: veth_44_221 > port 33: veth_44_222 > > [@sh5_44.1 ~]# ovs-ofctl dump-flows br0|grep '10.121.44.223' > cookie=0xa000007ef27f08, duration=13556.095s, table=0, n_packets=0, > n_bytes=0, idle_age=13556, priority=18000,ip,nw_src=10.121.44.223 > actions=enqueue:2:447 > cookie=0xa000004bdc6568, duration=13550.057s, table=0, n_packets=0, > n_bytes=0, idle_age=13550, > priority=28000,ip,nw_src=10.121.44.223,nw_dst=192.168.10.33 > actions=enqueue:2:447 > cookie=0xa000004bdc63c1, duration=13556.095s, table=0, n_packets=0, > n_bytes=0, idle_age=13556, > priority=28000,ip,nw_src=10.121.44.223,nw_dst=192.168.10.12 > actions=enqueue:2:447 > cookie=0xa000004bdc663c, duration=13553.072s, table=0, n_packets=0, > n_bytes=0, idle_age=13553, > priority=28000,ip,nw_src=10.121.44.223,nw_dst=192.168.10.44 > actions=enqueue:2:447 > cookie=0xa0000049f775a6, duration=13553.074s, table=0, n_packets=0, > n_bytes=0, idle_age=13553, > priority=28000,ip,nw_src=10.121.44.223,nw_dst=10.121.35.65 > actions=enqueue:2:447 > cookie=0xffffffffc144fcda, duration=13556.095s, table=0, n_packets=0, > n_bytes=0, idle_age=13556, > priority=28000,ip,nw_src=10.121.44.223,nw_dst=192.168.10.6 > actions=enqueue:2:447 > cookie=0xa000004bdc6497, duration=13550.058s, table=0, n_packets=0, > n_bytes=0, idle_age=13550, > priority=28000,ip,nw_src=10.121.44.223,nw_dst=192.168.10.25 > actions=enqueue:2:447 > cookie=0xa0000049f77400, duration=13550.058s, table=0, n_packets=0, > n_bytes=0, idle_age=13550, > priority=28000,ip,nw_src=10.121.44.223,nw_dst=10.121.35.45 > actions=enqueue:2:447 > cookie=0xa000004bdc6640, duration=13553.074s, table=0, n_packets=0, > n_bytes=0, idle_age=13553, > priority=28000,ip,nw_src=10.121.44.223,nw_dst=192.168.10.48 > actions=enqueue:2:447 > ...... > ...... > total: 53 flow for VM (10.121.44.223) > > > > At 2014-12-16 01:40:08, "Justin Pettit" <[email protected]> wrote: > >There's not enough information to understand the problem. In your > >"ovs-ofctl dump-flows" command, you have 7309 flows, but you've only shown > >us six. Also, those six flows are a little confusing, since the last one > >overlaps with all the other IP-related flows. It doesn't really matter, > >since they all have the same action, but if you change the actions, the > >behavior is undefined. > > > >What's the output of "ovs-dpctl show"? It would be interesting to know if > >lots of flows are missing in the datapath and being sent to userspace. > >Also, you may want to try 2.3.1, which is the latest version of OVS and is > >still maintained, unlike 2.1.x. > > > >--Justin > > > > > >> On Dec 14, 2014, at 9:02 PM, 李伟 <[email protected]> wrote: > >> > >> Hello : > >> There 10 VMs in openvswitch . these VMs access Internet through > >> NAT host.therefore There are a lot of internet ipaddr through OVS. > >> and add flows Following: > >> ovs-ofctl add-flow br0 "dl_type=0x0800,nw_proto=1 priority=900 > >> idle_timeout=0,actions=normal" > >> ovs-ofctl add-flow br0 "dl_type=0x0806 priority=900 > >> idle_timeout=0,actions=normal" > >> ovs-ofctl add-flow br0 "dl_type=0x8035 priority=900 > >> idle_timeout=0,actions=normal" > >> ovs-ofctl add-flow br0 "dl_type=0x0800,nw_proto=17 priority=900 > >> idle_timeout=0,actions=normal" > >> ovs-ofctl add-flow br0 "dl_type=0x0800,nw_proto=6 priority=900 > >> idle_timeout=0,actions=normal" > >> ovs-ofctl add-flow br0 "dl_type=0x0800 priority=900 > >> idle_timeout=0,actions=normal" > >> > >> but OVS high load still. How to reduce the OVS load ? thanks. > >> > >> > >> > >> ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- > >> PID USER PR NI VIRT RES SHR S %CPU MEM TIME+ COMMAND > >> > >> > >> 3145 root 20 0 1083m 29m 1608 S 565.7 0.0 21813:37 > >> /opt/apps/openvswitch/sbin/ovs-vswitchd --pidfile --detach > >> > >> [@sh5_44.1 ~]# ovs-dpctl dump-flows |wc -l > >> 53651 > >> > >> [@sh5_44.1 ~]# ovs-ofctl dump-flows br0|wc -l > >> 7309 > >> > >> [@sh5_44.1 ~]# ovs-vsctl get bridge br0 > >> other-config:flow-eviction-threshold > >> "4000" > >> > >> > >> [@sh5_44.1 ~]# ovs-vsctl --version > >> ovs-vsctl (Open vSwitch) 2.1.3 > >> Compiled Nov 3 2014 16:47:48 > >> > >> [@sh5_44.1 ~]# uptime > >> 12:15:04 up 2 days, 19:39, 1 user, load average: 702.12, 711.86, 711.05 > >> > >> > >> > >> > >> > >> > >> _______________________________________________ > >> discuss mailing list > >> [email protected] > >> http://openvswitch.org/mailman/listinfo/discuss > > > > > _______________________________________________ discuss mailing list [email protected] http://openvswitch.org/mailman/listinfo/discuss
