Hi Joe / Dumitru / Team,
        I am testing the process of ipv4 multicast in OVN/OVS, but the process
ovs-vswitchd encountered a segement fault, because the stack memory is 
exhausted.
My network is configured as below:

        1.1.1.2(vm1)              2.2.2.2(vm2)
           |                          |
        -----------------------------------------
        |                                       |
        |           logical_switch_01           |
        |                                       |
        -----------------------------------------
           |                          |     
           |                          |
        -----------------------------------------
        |1.1.1.1                    2.2.2.1     |
        |           logical_router              |
        |                 3.3.3.1               |
        -----------------------------------------
                            |
                            |
        -----------------------------------------
        |                                       |
        |           logical_switch_02           |
        |                                       |
        -----------------------------------------
                            |
                        3.3.3.3(vm3)

OVN-version:22.09
OVS-version:3.0.3

Configure:
[1]. On logical switch01/02, set mcast_snoop=true. 
[2]. On logical router, set mcast_relay=true.

1.1.1.2, 2.2.2.2, 3.3.3.3 is ip on virtual machines.
1.1.1.1, 2.2.2.1, 3.3.3.1 is gateway for vm1-3.

Test tool: iperf

Test case 01:
        vm1 use cmd :
        'iperf -c 224.0.55.55 -u -i 1 -l 1000 -b 100m --ttl 64 -t 1000'
        vm3 use cmd:
        'iperf -u -s -i 1 -B 224.0.55.55'
  In this case, vm3 will receive double flows than vm1 send. Beacuse when
flows send out from vm1 ,it flood to 1.1.1.1 and 2.2.2.1, then the two
gateway forword the flows to vm3.

Test case 02:
        vm1 use cmd :
        'iperf -c 224.0.55.55 -u -i 1 -l 1000 -b 100m --ttl 64 -t 1000'
        vm2 use cmd:
        'iperf -u -s -i 1 -B 224.0.55.55'
  In this case, ovs-vswitchd encountered a segement fault.Because when
flows send out from vm1, it flood to 1.1.1.1 and 2.2.2.1,, then flows 
are back to logical_switch_01 from 1.1.1.1 and 2.2.2.1, at this time,
when flows reach local_switch_01 from 1.1.1.1, the flows flood to 
router again. So , it becomes a loop, until the packet's ttl decrease
to 0, which leads the cpu usage to 1700% for ovs-vswitchd. As the
stack memory is 2M, the ovs-vswitchd encountered a segment fault.
If i configure the stack memory to 100M, the vsiwtchd do not case
segement fault at all, but the packet is dropped!

I think this is a problem, anyone else have ideas to solve this?

Thanks,
Wangchuanlei.
_______________________________________________
discuss mailing list
disc...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-discuss

Reply via email to