[dns-operations] suggested DNSKEY type

[A. Schulze]

Hello Viktor,

thanks for the detailed response.
I think I follow your suggestion and my preference: ECDSAP256SHA256

Andreas

Am 27.03.2018 um 17:02 schrieb Viktor Dukhovni:
> On Tue, Mar 27, 2018 at 12:22:49PM +0200, A. Schulze wrote:
> 
>> I've some zones signed with DNSSEC Key type 10 (RSASHA512)
>> I like to change the key type and would like to ask about the `better` choice:
>>
>>  - type 8  / RSASHA256
>>  - type 13 / ECDSAP256SHA256
> 
> Use type 13, any system bleeding-edge enough to support DANE, will
> also support ECDSA for DNSSEC, and there is in any case already
> widespread support for it.  The most significant benefit is smaller
> DNS packet sizes, which avoid potential issues with UDP fragmentation.
> 
> If you do choose RSA, I'd recommed a 2048-bit KSK and a 1280-bit
> ZSK.
> 
> On Tue, Mar 27, 2018 at 03:28:33PM +0200, A. Schulze wrote:
> 
>> I would prefer ECDSAP256SHA256 because smaller response size.
>> But how many user will get lost because their resolver don't support ECDSAP256SHA256?
>> What's with MTAs no longer deliver email messages to my MX because DANE fail?
> 
> As explained in another response that won't happen.  Worst case some
> rather unusual MTA will use opportunistic TLS instead.
> 
>> do have other DNS operators experiences?
> 
> Here are the DNSKEY RR algorithm frequeies from the DANE survey of
> 5.3 million domains:
> 
>       count  | alg
>     ---------+-----
>      6555386 |   8
>      3823984 |   7
>      1331242 |  13
>       261719 |   5
>        89726 |  14
>        65592 |  10
> 	2315 |   3
> 	  37 |   6
> 	  36 |  12
> 	  10 |  15
> 
> The top 3 choices are all sufficiently mainstream, but algorithm
> 10 is also a reasonable choice, if you do choose RSA, go with a
> 1536-bit or 2048-bit KSK and a 1280-bit ZSK to keep packet sizes
> more modest.  Some crypto maximalists might make noises at you,
> but these are fine choices.
>