Thanks a lot for your help, but could you please offer more information about: 1. You said that The GRE tunnel ports have a 1500 byte MTU. If your packets are larger than this they will be dropped and it will be logged in the kernel message buffer. Is there any methods to modify it much bigger?
2.If there are more than one physical NICs on a host, does OVS GRE build its stack for each Linux-IP? The situation is like host1 owns a public IP and a private IP, the other host2 only hold a private IP, host2 build the tunnel via the only NIC to host1's public IP, but when host1 can only send back packets to host2 via its private IP NIC. When build tunnel I give two parameter remote IP and MAC, does OVS then judge which NIC to use or just build stack for each IP? On Sat, Jul 10, 2010 at 9:42 AM, Jesse Gross <je...@nicira.com> wrote: > On Thu, Jul 8, 2010 at 10:56 PM, Yan <yan.p.b...@gmail.com> wrote: > >> Hi, >> I found another problem about the tunnel, its about the gre-dev's mtu in >> OVS. >> Thanks a lot for your suggestions and GRE of OVS works fine , but when I >> tried to use it on a special path, it stopped working >> >> The special path is actually a physical tunnel network path so when I >> again use OVS's GRE over it, I think maybe the MTU is too big for OVS GRE >> port do deal with it properly.When I forwarded the packet to the GRE port in >> OVS, it couldn't get out from the destination side.(Between the two sides >> the physical path is a tunnel network already). >> > > There are several places that you could be losing packets due to MTU: > > 1. The GRE tunnel ports have a 1500 byte MTU. If your packets are larger > than this they will be dropped and it will be logged in the kernel message > buffer. > > 2. By default we do path MTU discovery inside of tunnel ports. If you are > sending packets that are larger than the actual MTU of the tunnel (the path > MTU of the physical path minus the tunnel overhead) with the "don't > fragment" bit set and ignore the returned ICMP message no packets will get > through. > > 3. We also do path MTU discovery on the outside of the tunnel packets. If > the tunnel packets are larger than the MTU of some link on the network and > you have a firewall that is filtering the returned ICMP messages, no packets > will get through. > > 4. GRE uses the Linux IP stack's view of the network. If the MTU of the > physical device is not set correctly packets can be dropped. > > It's hard to tell what is going on without knowing the specifics of the > network. > -- Yan Pu y...@nakao-lab.org +81-80-4164-9937 NAKAO LAB The university of Tokyo Graduate School of Interdisciplinary Information Studies
_______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss_openvswitch.org
