On Wed, Apr 29, 2015 at 1:42 PM, Tom Herbert <[email protected]> wrote: > On Wed, Apr 29, 2015 at 11:30 AM, Behcet Sarikaya > <[email protected]> wrote: >> On Wed, Apr 29, 2015 at 12:06 PM, Tom Herbert <[email protected]> wrote: >>> On Wed, Apr 29, 2015 at 9:09 AM, Behcet Sarikaya <[email protected]> >>> wrote: >>>> Hi Tom, >>>> >>>> On Wed, Apr 29, 2015 at 11:00 AM, Tom Herbert <[email protected]> wrote: >>>>> On Wed, Apr 29, 2015 at 7:50 AM, Templin, Fred L >>>>> <[email protected]> wrote: >>>>>> Hi Lucy, >>>>>> >>>>>>> -----Original Message----- >>>>>>> From: Lucy yong [mailto:[email protected]] >>>>>>> Sent: Wednesday, April 29, 2015 7:48 AM >>>>>>> To: Templin, Fred L; [email protected]; [email protected] >>>>>>> Subject: RE: [Int-area] Why combine IP-in-UDP with GUE? >>>>>>> >>>>>>> >>>>>>> Getting back to our earlier discussion, IP-in-UDP and GUE are currently >>>>>>> two half-solutions. Put them together and you get a whole >>>>>>> solution. >>>>>>> Keep them apart, and someone else is going to have to write a whole >>>>>>> solution sometime down the line from now. >>>>>>> [Lucy] GUE can support IP payload. Don't know why you state that they >>>>>>> are two half-solutions. Is the compression a mandatory >>>>>>> requirement here? I think that IP-in-UDP proposal as a compression >>>>>>> version is better that use of first nibble. However we need clarify >>>>>>> what limitation and constraint the compression solution has. >>>>>> >>>>>> GUE is missing header compression, and IP-in-UDP is missing tunnel >>>>>> fragmentation. That is what I mean when I say that if combined you >>>>>> get a whole solution. >>>>>> >>>>> Adding this header compression just adds a whole bunch of complexity >>>>> to the protocol to save a grand total of four bytes for what is likely >>>>> a very narrow use case. >>>> >>>>>This is not applicable when GUE is used for >>>>> network virtualization, >>>> >>>> >>>> I don't think GUE is a replacement or even an improvement for VXLAN >>>> encapsulation. >>>> >>> All the arguments as to why VXLAN is insufficient in multi-tenant >>> deployments was made in nvo3. Please read those and the GUE drafts >>> (draft-hy-nvo3-gue-4-nvo-01, >>> draft-ietf-nvo3-gue-00, >> >> I read this draft, I could not see any such arguments. It just >> mentions VXLAN as a reference like other things. >> >> If true, it should explicitly address this issue. >> I am not sure if it can say more than what it is that is a generic >> encapsulation techniques that can be used in the data center to tunnel >> things. >> >> But VXLAN is designed to provide VM-to-VM communication. >> So the design criteria is completely different in these two techniques. >> > GUE has been adopted by nvo3 network virtualization data plane, along > with Geneve and VXLAN-GPE-- VXLAN has not. Again if you have comments > on that use of GUE please take them to nvo3. This thread is about a > generic modification to GUE without respect to the network > virtualization use case.
I reacted on this thread because you claimed network virtualization use case now you are saying it is not about that. I am fine then :-) Regards, Behcet > >>> and >>> draft-hy-gue-4-secure-transport-01). If you have any comments or >>> questions take them to the nvo3 list. >>> >>>> While VXLAN is 1-N type of tunneling, GUE is 1-1. >>>> >>> I don't understand what this means. >> >> The key is in VM-to-VM communication. The other VM could be under any >> VTEP or NVE. >> >> Regards, >> >> Behcet >>> >>>> Regards, >>>> >>>> Behcet >>>>> we are encapsulating something other than IP, >>>>> we need OAM, or using any other feature of GUE. In my deployment, I >>>>> don't have any use case for that since minimally I will be using >>>>> remote checksum offload option because that does give a material >>>>> performance advantage. >>>>> >>>>> The premise of GUE is simple, it has a simple header that encapsulates >>>>> any IP protocol expressed by IP protocol number and allows optional >>>>> extensions and control packets-- let's keep it simple! If saving those >>>>> four bytes is really important in some deployment and GUE is still >>>>> needed in certain case, then just use GUE and IP-in-UDP in tandem. >>>>> >>>>> Tom >>>>> >>>>>> Thanks - Fred >>>>>> [email protected] >>>>>> >>>>>>> Lucy >>>>>>> >>>>>>> Thanks - Fred >>>>>>> [email protected] >>>>>>> >>>>>>> > However, if GUE payload is >>>>>>> > IP, it is OK to inspect the first nibble of the payload to determine >>>>>>> > IPv4 or IPv6 because this aligns with IP protocol. >>>>>>> > >>>>>>> > Thanks, >>>>>>> > Lucy >>>>>>> > >>>>>>> > - Stewart >>>>>>> > >>>>>>> > >>>>>>> > >>>>>>> > _______________________________________________ >>>>>>> > Int-area mailing list >>>>>>> > [email protected] >>>>>>> > https://www.ietf.org/mailman/listinfo/int-area >>>>>>> > >>>>>>> > _______________________________________________ >>>>>>> > Int-area mailing list >>>>>>> > [email protected] >>>>>>> > https://www.ietf.org/mailman/listinfo/int-area >>>>>> >>>>>> _______________________________________________ >>>>>> Int-area mailing list >>>>>> [email protected] >>>>>> https://www.ietf.org/mailman/listinfo/int-area >>>>> >>>>> _______________________________________________ >>>>> Int-area mailing list >>>>> [email protected] >>>>> https://www.ietf.org/mailman/listinfo/int-area _______________________________________________ Int-area mailing list [email protected] https://www.ietf.org/mailman/listinfo/int-area
