On Sat, Apr 17, 2010 at 05:51:20PM +0200, Trygve Sanne Hardersen wrote: > On Sat, Apr 17, 2010 at 1:09 AM, Ben Pfaff <b...@nicira.com> wrote: > > I don't think that changing ETH_ADDR_LEN will help. > > Yeah, that's my feeling as well. Do you have any idea how hard it will > be to support this? If someone can give me a few hint about the > requirements I'd like to take a shot at implementing support for this. > > My understanding is that an IPOIB device is no different from a normal > Ethernet device from the application perspective. If the only > functionality needed is 20-byte MAC addresses I believe it should be > achievable.
It would probably not be too hard to modify OVS to support 20-byte MAC addresses exclusively. Beyond changing ETH_ADDR_LEN, you would have to audit some of the code to make sure that it does not have some kind of hidden dependency on 6-byte addresses. You would probably have to drop OpenFlow support, since OpenFlow assumes that MAC addresses are 6 bytes long. Flow data structures would grow by 32 bytes, approximately doubling their size, which would be a shame. To support both sizes of MAC addresses would be more work. I think that it would amount to a more careful audit of each place a MAC address is used. It would probably require adding address-length fields some places. I'm not sure how it would interact with OpenFlow. You could drop OpenFlow support, or you could only expose the 6-byte MAC flows through OpenFlow, or you could add OpenFlow extensions. I have a question, though: do you really want to actually *switch* IPoIB? If you just want to use it on your host, without doing switching, then why not skip going through the bridge or OVS entirely? Just configure the network device directly. _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss_openvswitch.org
