> -----Original Message----- > From: Joe Touch [mailto:[email protected]] > Sent: Monday, May 04, 2015 9:06 AM > To: Xuxiaohu; Templin, Fred L; Lucy yong; Tom Herbert > Cc: [email protected] > Subject: Re: [Int-area] Why combine IP-in-UDP with GUE? > > > > On 5/3/2015 7:53 PM, Xuxiaohu wrote: > >>> As > >>> > > such, it's not recommended to perform fragmentation on the tunnel > >>> > > layer and the outer IP layer. > >> > > >> > It's provably required for IPv4 DF=1 and IPv6. > > > > If the IPv4 packets with DF=1 and IPv6 packets are transported over > > Ethernet, do you still want the Ethernet layer to do fragmentation? > > The proof is for IP in anything that eventually goes over IP again. > > If you put IP in Ethernet and then put that pack in IP (e.g., via PPP, > EtherIP, or GRE), then the tunnel will need to support fragmentation. > > The question is "what layer provides the tunnel". If you treat the > entire set of headers of encapsulation as a set, you can use any layer > in that set you want. If you treat UDP as the sole controllable > encapsulation layer, then you need to support fragmentation there. > > The ONLY reason IP doesn't require fragmentation over Ethernet is > because the IP layer directly above Ethernet has an MTU of 1280 (for > IPv6) or 68 (for IPv4) and Ethernet supports an MTU of 1500.
That, plus we are a long way off from seeing 8K everywhere. Go to most end user devices, and you will still see 1500. I suspect the same is true for most networking gear near the edges of the network. Thanks - Fred [email protected] > Joe _______________________________________________ Int-area mailing list [email protected] https://www.ietf.org/mailman/listinfo/int-area
