It appears that Kazunori Fujiwara  <[email protected]> said:
>These limits are useful because they are already implemented and seem
>to be effective for stable operation of the resolver.

This may be more work than you planned to do, but I think a document
that described the limits in popular DNS caches such as unbound, Knot,
and PowerDNS as well as BIND would be very useful. It would both give
us a list of the things that can cause performance problems, and also
some guidance about limits we should be aware of.

>I agree, but an RRSet with a large number of RRs cannot be carried
>over UDP transport and will eventually break down as they approach the
>65535 octets limit.  I think that apex TXT will become smaller once
>domain verification techniques are standardized and widespread.

I added junk to the TXT record at _dmarc.johnlevine.com so it is over
50K, and then I went to a bunch of DMARC test sites to see if they
handle it properly. Almost without exception they work. This suggests
that large DNS responses may not be fast, but they work.

I suppose I could try another experiment with a hundred 500 byte
junk records and see if they pick out the good one.

R's,
John

PS: There was a bug in NSD that broke when a single record was
over 32K but they fixed it quickly when I reported it.

_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to