Please see inline. Thanks Saumya.
On 7/2/15, 2:24 AM, "Joe Touch" <[email protected]> wrote: > > >On 7/1/2015 1:38 PM, Saumya Dikshit (sadikshi) wrote: >> Hi Joe >> >> As far as I know, PLMTUD provides an extension to PMTUD and does not >> deprecates the legacy. > >There's not much to deprecate, per se. PLMTUD is certainly preferred. <Saumya> Certainly not much, but PLMTUD surely firms up the Classic PMTUD >> ICMP can be one of the potential upper-layer protocols which can be used >> for PLMTUD as well. > >It can be, but it's not needed at all. <Saumya>Yes, ICMP can be used as *ONE* of the instruments, and as long that is an option this draft may be helpful > >> Even though PLMTUD is ICMP agnostic, ICMP PTB¹s can be leveraged in >> determining probe success/failures, >> and also detecting ICMP black hole to further refine the probe. > >You cannot detect an ICMP black hole. You *infer* that event from a set >of conditions, and that set can have several causes. <Saumya> That’s right, black-hole is inferred, may be wrong choice of words at my end. Where ICMP PTB definitely plays a part. > >> Ensuring that icmp PTB generated from the underlay is percolated to end >> point device >> should only aid PLMTUD and reduce the scope of icmp black-holing. > >ICMP PTBs should be received from L2 only when L2 cannot transit a >message, i.e., when it cannot fragment and reassemble it. It should NOT >be the result of L2 preferring a native size. > >I.e., the basis of Sec 3.1.1 is false. A packet that is fragmented and >reassembled at the encapsulation layer (L2, to the E2E L3) is NOT "too >big", and should never generate an ICMP error anyway. <Saumya> The encapsulation and reassembly mention in 3.1.1 is with respect to the outer header encapsulation (for Vxlan tunnel encap). Frag/reassembly in underlay is performed in context of outer encap which is L3 based. That’s one of the reason Vxlan scores over other L2-tunneling schema. > >Joe > > > >> Thanks >> Saumya. >> >> On 7/2/15, 1:24 AM, "Joe Touch" <[email protected]> wrote: >> >>> >>> >>> On 7/1/2015 12:09 PM, Saumya Dikshit (sadikshi) wrote: >>>> <Saumya> This solution can be viewed as an enabler for PLMTUD if >>>> required >>>> on any end-point >>>> device connected to vxlan network and not the Vxlan gateways >>>>themselves. >>> >>> PLMTUD does not use ICMPs. It relies on positive feedback from the data >>> plane, rather than negative feedback from the control plane. >>> >>> Joe >> _______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
