On 2015/7/1 22:57, Kavanagh, Mark B wrote:
As following topology, two VMs communicate via Vxlan tunnel.
TCP pkts may be droped for exceeding the mtu of host DPDK port in br2.
Now I can change the VMs' mtu from default to smaller to accommodate the
traffics, it works well.

The way I have not find to change mtu of DPDK phy port, is there a
blueprint for supporting this by ovs, like "ovs-appctl
netdev-dpdk/set-xxx" or another way?

         Host1                  |                     Host2
                     (l-ip)-   L3    -(r-ip)
-----   -----        -----     |     -----       -----  ----- -----
|VM1| --|br1|Vxlan --|br2|DPDK---DPDK|br2|--Vxlan|br2|--|br1|--|VM2|
-----   -----        -----     |     -----       -----  ----- -----

_______________________________________________
discuss mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/discuss
Hi,

There is currently no way to configure the MTU of a DPDK phy port at runtime.

Theoretically, enabling jumbo frames and adjusting the dev's max_rx_pkt_len in 
the rte_eth_conf, (and consequently, the size of mbufs in the mempool) could 
work; however, OVS sets the MTU for all ports to the minimum of the 
non-datapath ports, so you might need to play around with it. This is something 
that I plan on looking at in the next few weeks (compile-time option, not 
runtime).

Thanks,
Mark
Thanks, I will try it based on your suggestion.
_______________________________________________
discuss mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/discuss

Reply via email to