[ovs-discuss] Drops with high rate of packet_In in OVS

[Anton Matsiuk]

UPD:

Sorry, I forget to add dump from user space that shows that part of packets
are missed:
sudo ovs-ofctl dump-aggregate br0
NXST_AGGREGATE reply (xid=0x4): packet_count=380 byte_count=20736
flow_count=500



On 18 November 2013 15:52, Anton Matsiuk <anton.matsiuk at gmail.com> wrote:

> Hello all,
>
> I test the Open vSwitch in the following scheme: I use 2 hosts directly
> connected to OSV and external OpenFlow Controller. Host1 generates UDP
> datagrams with sequential ports towards Host2, Host 2 listens for  these
> UDP datagrams. In responce to every UDP datagram OVS generates Packet_In
> and Controller sends Flow_Mod with granularity up to L4 (so for every pair
> of UDP port numbers it installs separate rule). I send bunch of UDP
> datagrams from Host1 and calculate how many of them arrived to Host2.
>
> Starting  some number of packets ( > 300) packet drops are observed. For
> instance, if I generate 500 UDP packets in 120 ms only around 380 of them
> arrive to Host2.
> ovs-dpctl show -s command shows , that I have 500 flows installed 500
> packets on ingress interface (port 1) of OVS and 380 on egress (port 2).
> So, it seems like OVS drop packets starting with some load threshold:
>
> lookups: hit:0 missed:500 lost:0
> flows: 500
> port 1: eth2
> RX packets:500 errors:0 dropped:0 overruns:0 frame:0
>  TX packets:12 errors:0 dropped:0 aborted:0 carrier:0
> collisions:0
>         port 2: eth3
>  RX packets:0 errors:0 dropped:0 overruns:0 frame:0
> TX packets:380 errors:0 dropped:0 aborted:0 carrier:0
>  collisions:0
>
> ovs-dpctl dump-flows br0 shows all the flows installed but no  one with
> matched packets:
>
> in_port(1),eth(src=00:00:00:00:00:01,dst=00:00:00:00:00:02),eth_type(0x8100),vlan(vid=10,pcp=3),encap(eth_type(0x0800),ipv4(src=10.0.0.1,dst=10.0.0.2,proto=17,tos=0,ttl=64,frag=no),udp(src=52363,dst=50335)),
> packets:0, bytes:0, used:never,
> actions:userspace(pid=4294962855,slow_path(controller))
>
> I tried both with detached controller and running in the same machine as
> OVS (I use out-of-band option for controller and disable-in-band=true)
>
> Also I tried to replace Flow_Mods with Packet_Out messages for every
> ingress UDP datagram (identifying by buffer_id) but I experienced the same
> drops. I tested it on different machines (in Mininet and with separated
> real hosts)
>
> I suspect that such behavior is caused by
> actions:userspace(pid=4294962855,slow_path(controller)) but how can I
> verify where drops are occurred? is there possibility to debug internal
> queues or buffers?
>
> And how could I fix such drops? Should I implement some advances QoS
> policies in OVS in this case?
>
> --
>
> Best regards,
> Anton Matsiuk
>



-- 
Best regards,
Anton Matsiuk
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/discuss/attachments/20131119/3fc3477f/attachment.htm>