Hi Joel, Further to my previous message, my concern about these recursive DNS lookups is probably reduced by what I expect a DNS server would do when returning a reference to a second DNS server which is more authoritative: including the Locator and Identifier records of the second server along with the FQDN which properly identifies the second server.
Still, as far as I know, whenever a DNS server replies to a querier (a sending host or its resolver) it has not recently sent a packet to, it needs to perform a DNS lookup on that querier. As far as I can tell, this would be a reverse lookup of the Identifier part of the source address in the query packet. This would give it the FQDN, which it probably doesn't need, but would also give multiple Locators by which the querier could be reached. In the case of the querier being a mobile host, the DNS server will need to perform a reverse lookup even if it recently sent a packet to this host, because the TTL on DNS records for mobile hosts will be "very short". When the mobile host sends a query to a resolver, the resolver will be doing this stuff with one or more other DNS servers. But when the resolver needs to send the response back to the mobile host, it will likewise typically need to do a reverse lookup on that host's Identifier, because any cached reverse lookup records will have a "very short" TTL. So with ILNP, these DNS lookups keep multiplying like the Sorcerer's Apprentice's brooms. Just because DNS delay is judged to be "acceptable" in today's Internet doesn't mean it will be with these ILNP changes. I believe that even today's DNS delays are unacceptable when they add to the time required to complete an exchange of initial packets. All DNS lookups today are potentially slow or very slow due to DNS being a global system and because they are all subject to failure and the need for a retry if the query or response packet is lost. I still have no idea how multiple Identifiers can be related to multiple Locators. ILNP has been in development for years and I would have thought that obvious problems like this would have been sorted out long ago. If this is regarded as mere "engineering", then I think the "architects" should take a greater interest in engineering before being confident their overall structure is sound. The solution is probably to have a set of Locators for each of the Identifiers. Likewise, what is the purpose of Identifiers and Locators with priorities if the current IDs do not specify how a host is to choose between them? This isn't just low-level detail - it is fundamental to the protocol design and therefore to the "architecture". - Robin _______________________________________________ rrg mailing list [email protected] http://www.irtf.org/mailman/listinfo/rrg
