Re: [dnsext] Fwd: New Version Notification for draft-bellis-dnsext-multi-qtypes-01.txt

I think this is a bad idea:  

There are two kinds of secondary resource records: ones that can be asked in parallel (e.g. A and AAAA records associated with the same name), and ones that are sequentialized (e.g. the A and AAAA records corresponding to MX entries).

As far as I can tell, this draft only addresses the former, not the latter, as for the latter the client doesn't know yet which question to ASK, so it can't construct a multiple question query.

And in the former case, the efficiency savings are actually negative:  It imposes a needless sequentialization on what was previously a parallel activity of asking both questions separately.  

If the server has one record in the cache but not the other, it will have to wait until it has the other record.  This slows "happy eyeballs" approaches where A and AAAA records are looked up simultaneously, and then the contacts also proceed in parallel.

So this approach can NEVER be faster for the server to compute, only equal time or slower, thus the only speedups can occur if this made communication substantially smaller.

But it doesn't.  You only save the packet framing one one question and one response per additional RTYPE queried, which is, in practice, roughly 200B of traffic when one includes ALL overhead (including interframe gaps, ethernet framing and minimum packet sizes).  On a slow 1 Mbps home connection, this still represents ONLY a 1.6 ms savings.

The only case where this would have more than a 10ms savings is on a <200 Kbps link, and such a link is so slow that nobody will care about the savings because everything else is so glacial, and such links are likely to not be querying for AAAA records anyway (the most common case)

Thus I think this proposal is a bad idea:  It adds a substantial amount of complexity for something that can substantially slow things down by removing parallelization, and even when everything works right, is only 1.6ms of savings on a 1 Mbps link (and .16ms savings on a 10 Mbps link).

_______________________________________________
dnsext mailing list
dnsext@ietf.org
https://www.ietf.org/mailman/listinfo/dnsext