Re: [ga] No more FUD -- what is the current size of the .com database?
I also forgot to add that it is very easy to build a balancing unit for UDP based DNS that takes a look at the query names and directs them to servers based on some characteristic of the question in the DNS query. Thus for example, one could dispatch queries based on the last character of the question name, thus resulting in up to a 37 way split of the traffic. (The zone file could be split among those servers so that each only needed to load the part relevant to itself, but that might make "additional" record generation a bit less useful than it is today; so it's probably better if each server has the full zone but receives only a portion of the entire query load.) I mentioned to IANA during their queries about how to vet the quality of TLD servers whether they were concerned with the source IP address on the response. They felt there was no reason to be concerned if the response to a query comes back with a source IP address different from that to which the query was sent. The point of this is that if this is indeed acceptable, then the load balancer/splitter only has to do a destination IP mapping on the incoming packets, not on the responses. That saves some computing iron. What I am getting at is this - it is quite easy, and cheap, to build an array of servers to provide a reliable query response service, even for obese zones such as .com. The machines are the easy, and less expensive, part - it's the facility, the management, the physical protection, the power and cooling, and swap-outs of failed gear that are the harder part. And the bandwidth cost is likely to be a major, and perhaps the dominant, cost component of the entire system. --karl--
|