Hi Joel, Thanks for your response. Here's the shortest complete statement I can think of about how ILNP would degrade performance in general, and how this would be worse when one or both hosts are on slow or unreliable links.
This involves some points which were not in my previous message - about the DNS server having to do potentially multiple DNS lookups in order to reply, and how each such lookup could itself involve more DNS lookups . . . which themselves involve more DNS lookups. Today, two hosts can perform an initial exchange of application packets without the need to send any other packets or perform any kind of DNS or mapping lookup. So the total time for the exchange is the RTT between the two hosts. Before two ILNP hosts can complete an initial exchange of application packets, the first host must perform a DNS lookup before sending its packet, and the second must do the same before sending its response. Since DNS is a global system, and since response and query packets can be lost, this typically involves significant delays, even for hosts with fast, low-latency links to the DFZ. The DNS lookups themselves - whether performed by the host or by a nearby and better-connected resolver - may involve multiple queries to different servers in order to find an authoritative server. This is exacerbated by the requirement to keep TTLs very short for any lookup concerning a mobile ILNP host. The total delay in receiving the DNS response is further increased by some or many of the DNS servers being on ILNP addresses - being identified by a FQDN and so requiring a DNS lookup to find the Identifier and Locator, just in order to send the query to the server. Furthermore, if the querier is on an ILNP address, the DNS server cannot reply without also performing a DNS lookup (which itself may involve multiple DNS servers and further sets of DNS lookups in order to be able to send queries to these DNS servers). This burden of DNS lookup activity and the delays it imposes on any initial packet exchange would be a serious disincentive to the adoption of ILNP. These DNS lookup delays are significantly worsened for any host which is on a wireless link - such as a 3G link with cost, latency and reliability problems. In the future mobile hosts with 3G wireless links are likely to comprise a significant proportion of all hosts - probably the majority. The performance degradation due to these DNS lookups over wireless links is a serious disincentive to adoption, as is the very high load placed on the global DNS system by the short RTT for all records concerning these billions of mobile hosts. This won't fit in your 500 word critique, but at least it is here as part of the RRG discussion. - Robin _______________________________________________ rrg mailing list [email protected] http://www.irtf.org/mailman/listinfo/rrg
