Hi, Looking at the RTT from the ping command it seems like it's not going through the controller (First packet latency 15ms, after that 0.2ms, so it seems flows are installed)
I tried with a POX example application (l2_learning), but same issue. This is the flowdump (fresh reboot, ping test between hosts): NXST_FLOW reply (xid=0x4): cookie=0x0, duration=16.258s, table=0, n_packets=16, n_bytes=1568, idle_timeout=10, hard_timeout=30, idle_age=1, priority=65535,icmp,in_port=2,vlan_tci=0x0000,dl_src=00:30:48:dd:7f:b9,dl_dst=00:30:48:de:63:39,nw_src=10.1.2.1,nw_dst=10.1.2.2,nw_tos=0,icmp_type=0,icmp_code=0 actions=output:1 cookie=0x0, duration=16.26s, table=0, n_packets=16, n_bytes=1568, idle_timeout=10, hard_timeout=30, idle_age=1, priority=65535,icmp,in_port=1,vlan_tci=0x0000,dl_src=00:30:48:de:63:39,dl_dst=00:30:48:dd:7f:b9,nw_src=10.1.2.2,nw_dst=10.1.2.1,nw_tos=0,icmp_type=8,icmp_code=0 actions=output:2 Seems to work fine, only throughput is problematic. As Gurucharan Shetty suggested, I tried running a regular linux bridge and that delivers good results (iperf reports 933Mbps). But with ovs iperf results are at most 30 Mbps. (200Mbps was measured generating udp packets with pktgen) I checked /sbin/lsmod and openvswitch is loaded but not used! Maybe that's an issue? Do you have to add a specific argument to use the kernel module for the bridge? ovs version: 1.7.3 (build from source) kernel: 2.6.38-11-generic-pae Kind regards, Tim 2013/3/6 Justin Pettit <jpet...@nicira.com> > Are you sure the traffic isn't being sent to the controller? Depending on > your controller, that could slows things down significantly. > > I don't think you mentioned what version of OVS you're using. What do you > see from "ovs-dpctl show", "ovs-dpctl dump-flows <br>", and "ovs-ofctl > dump-flows <br>"? Obviously, replace "<br>" with the name of your bridge. > > --Justin > > > On Mar 6, 2013, at 11:28 AM, Tmusic <tmusic...@gmail.com> wrote: > > > Hi, > > > > Thank you for the quick answer! > > I'm actually using it without VM's. > > It's an experiment with three "real" computers. One machine with ovs > acting as switch, the other machines are connected to the ovs-machine, each > on a separate nic. (controller runs at a fourth machine, but that's > probably not relevant) > > > > So it's all bare metal :) > > > > Kind regards, > > > > Tim > > > > > > 2013/3/6 Gurucharan Shetty <shet...@nicira.com> > > On Wed, Mar 6, 2013 at 7:14 AM, Tmusic <tmusic...@gmail.com> wrote: > > > Hi, > > > > > > I'm running ovs on Ubuntu 11.04 and noticed that it can process max. > 200Mbps > > > (NIC's are gigabit). > > That is quite low. I have seen 9 Gbps on non-tunnel mode (on 10G > > NICs). You are sending traffic > > through VM or through the host? If the former, use a high performance > > driver. You can always > > remove openvswitch and see the throughput with the exact same setup > > with just the linux bridge and compare. > > > > > > > > > > > > It's used as an openflow switch with POX as controller and about 10 > flows > > > are installed. ovs is running with kernel module. The CPU is a quadcore > > > Intel Xeon running at 2.2Ghz and one of the cores is fully used by > vswitchd. > > > So the CPU is limiting in this case. I'm pretty sure there are are no > loops > > > in the network. > > > > > > Is this normal level of performance that can be expected from ovs or > could > > > something wrong with the configuration? > > > > > > Kind regards and thank you in advance, > > > > > > Tim > > > > > > _______________________________________________ > > > discuss mailing list > > > discuss@openvswitch.org > > > http://openvswitch.org/mailman/listinfo/discuss > > > > > > > _______________________________________________ > > discuss mailing list > > discuss@openvswitch.org > > http://openvswitch.org/mailman/listinfo/discuss > >
_______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
