> Right. We used the probe's local resolver, and the TCP/UDP and DO > choice is the client's (think dig +tcp etc.). > > If that +tcp switch changes the error rate, then the bottleneck > may be between the stub and the resolver, not upstream. So, yes, > interpretation is necessary.
If the client requests A or AAAA with DO=0 then TCP or not should not have a significant effect. If it does, then something weird is going on. I now wonder if some software has a rather low limit on the size of records or RRsets. In that case we can expect failures even with perfect transport connections. _______________________________________________ DNSOP mailing list -- [email protected] To unsubscribe send an email to [email protected]
