Hi Teco, You wrote:
> I wouldn't say renumbering is always testable, is reliable and is secure. > But I think there are a large number of use cases that this is not a > problem. OK - but are you suggesting that we should pursue any scalable routing solution which is based on the notion that administrators will accept renumbering of entire end-user networks as "routine"? I can't imagine there's anything we could do - other than waving a wand which changes the world's applications host stacks and routers - which would convince administrators that renumbering was sufficiently testable, secure or reliable for them to be happy doing it as part of ordinary operations, for instance when selecting one or more new ISPs. The only scalable approach to multihoming without the major new developments of a core-edge-separation system - such as map-encap - is for IPv6 only: SHIM6. For SHIM6, each end-user network needs at least two ISPs, and each host gets at least two IP addresses for SHIM6 to work with. SHIM6 fails the "Will it be adopted widely and quickly, even when initially hardly anyone uses it?" test of providing substantial benefits to all early adoptors, since it only works when the host at the other end of the communication also runs SHIM6. Also, SHIM6 only provides multihoming and TE without centralised control. It doesn't provide portability without renumbering, which the core-edge separation systems can. > Moreover, IMHO there are use cases where server farms needs very high > availability and this could be provided by using many PA addresses, > multi-homing and automatic renumbering with DNS Dynamic Update. Maybe so - but what if the servers were for websites of hundreds or thousands of customers, and the customers' DNSes were the ones which needed to be updated? I argued that technically and administratively this would be prohibitively costly, fragile, insecure etc. I wasn't suggesting that there be no efforts to make it easier to renumber networks. Just that any such developments are not something we can expect to make administrators regard renumbering as "routine" in the 3 to 5 years we have (my guess) to devise a scalable routing solution. In his most recent message, Eliot seems to agree that within the time frame we have for solving the scalable routing problem we can't move everyone to a whole new set of protocols. I think he agrees that we can't assume that renumbering could be "routine" for larger networks - but I think he is hopeful about it being "routine" for smaller networks. - Robin _______________________________________________ rrg mailing list [email protected] https://www.irtf.org/mailman/listinfo/rrg
