Question regarding load balancing

[Kevin Darcy]

Forrest Aldrich wrote:

> In the case of using a forwarder, how does this function actually work.    For example, we relay a query to our "forwarder" (as defined in named.conf), if the forwarder doesn't have a response, what happens at this point.  Do we send a query out to the internet ourselves or is the forwarder burdened with this task, relaying that response instead of what's in its cache.

Forwarding uses recursives queries, so as long as the forwarder provides recursive service, then it is obligated to resolve the query on your behalf. If the query times out and you have "forward first" configured, then you'll attempt to resolve it iteratively, but the timeout is fairly long -- this was no doubt intended as a fallback mechanism in case of failure, rather than a load-balancing
mechanism _per_se_.If you don't want to burden your forwarder, and you have enough connectivity and machine resources to resolve all of the names yourself, why forward in the first place?

> In the case where you can define multiple IP addresses for the same DNS server, in an effort to load balance, how does this algorithm work.   Take for example a situation where you want to have multiple DNS servers in all of your edge POPs.  You would want each POP to use a DNS server that is closer in proximity.  I don't think the algorithm for load balancing really has anything like this.

Forwarding configuration uses IP addresses, not names, so I'm not sure what you're trying to achieve by defining multiple IP addresses for the same DNS server.

When using iterative resolution, BIND will favor "faster" nameservers over slower ones. I hear this functionality is being added to forwarder lists as well (see CHANGES #997). Doesn't this give you what you want?

                                                                        - Kevin