[ovs-discuss] packets getting dropped at egress

[Dave Waters]

Hi,

I am unable to increase the MTU of the bridge port (the internal port) to
be anything more than 1500. While i am able to reduce it, i am not able to
increase it.

The reason i want to increase the MTU of the internal port is because i am
encapsulating all packets that i recieve on eth1 inside VXLAN which
increase the packet size. If i recieve a 1500 byte packet on eth1, then i
am dropping the packets since the size goes beyond the MTU of the bridge
after i put in VXLAN encapsulation.

root at ubuntu:/home/dw# ovs-vsctl show
b69c9516-a7dd-445c-a82d-8fb704e587a3
    Bridge "br0"
        Port "gre0"
            Interface "vxlanIp"
                type: vxlan
                options: {key="500", remote_ip="192.168.0.1"}
        Port "eth1"
            Interface "eth1"
        Port "br0"
            Interface "br0"
                type: internal
        Port "eth0"
            Interface "eth0"
    ovs_version: "2.0.2"
root at ubuntu:/home/dw#

root at ubuntu:/home/dw# ifconfig br0 mtu 1600

root at ubuntu:/home/dw# ifconfig br0
br0     Link encap:Ethernet  HWaddr 00:90:0b:35:e4:36
          inet addr:192.16.10.147  Bcast:192.168.10.255  Mask:255.255.255.0
          inet6 addr: fe80::dc88:f5ff:feef:ef9b/64 Scope:Link
          UP BROADCAST RUNNING  MTU:1500  Metric:1
[clipped]

root at ubuntu:/home/dw#

Dave

On Thu, Jul 23, 2015 at 7:49 PM, gowrishankar <
gowrishankar.m at linux.vnet.ibm.com> wrote:

> On Thursday 23 July 2015 04:16 AM, Dave Waters wrote:
>
>> Hi,
>>
>> I have two ports in my OVS bridge. I receive packets from one port
>> (portA) and i send them out of the other (portB). When i send out the
>> packets, they are VXLAN tunneled to the other end.
>>
>> Now, the default MTU of all ports (including the bridge port) is 1500 in
>> my setup.
>>
>> The issue is that when i get a 1500 byte packet on portA, i try to push
>> it out on portB after slapping on the VXLAN headers. This results in a
>> packet size thats greater than what portB can handle, and hence the packets
>> are dropped.
>>
>>
> Is increasing MTU of portB to 1550 (to accommodate additional 50 bytes
> when no vlan taged) a constraint in your setup ?
>
>
>  I know that OVS does not handle IP fragmentation/reassembly, so how do we
>> deal with this situation? I dont think we can rely on path MTU discovery
>> since not all applications do PMTU before spewing out packets. Any ideas,
>> anybody?
>>
>> Warm regards,
>> Dave
>>
>>
> --
> Regards,
> Gowrishankar M
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20150724/1eb57abb/attachment-0002.html>