Hello John,

Il 2026-10-05 19:08 John R Levine ha scritto:
That reminds me, are we including RSA keys? The argument against was that ed25519 keys are better in every way, argument for was that there are a lot of published RSA keys people might want to keep using.

On the RSA question, a measurement rather than an argument. One mailbox on my server, 61815 DKIM-Signature headers, delivery years 2024 to date. Continuation lines were unfolded before the a= tag was extracted. The tag was matched only at a tag boundary, so base64 ending in “a=” is not counted. These are signatures as received, including those added by intermediaries on the path.

What follows is a statistic and not a fact about the Internet. It is one mailbox and another one would give different figures. What makes it worth reporting is that nothing in the mechanism behind these numbers is specific to my site. The senders are the ordinary mix of personal correspondence, commercial mail and mailing lists that reaches any small operator. The signing software is theirs and not mine. I would expect the orders of magnitude to hold elsewhere and I would be glad to see a sample where they do not.

year    sigs     ed25519-sha256      rsa-sha1       rsa-sha256
2024   21107      1    0.005%     1497   7.092%    19609  92.903%
2025   19356      0    0.000%      531   2.743%    18825  97.257%
2026   21352     39    0.183%       76   0.356%    21237  99.461%
ALL    61815     40    0.065%     2104   3.404%    59671  96.532%

Two transitions of the same age are visible here, moving in opposite directions. rsa-sha1 is going away. It is down from 7.1% to 0.36% in two years. Looking at the signing domains, most of that decline is one large retailer that stopped during 2025, and the remainder is the turnover of short-lived domains of the kind associated with bulk unsolicited mail. It reads as a long tail in exhaustion rather than as a prohibition being defied.

ed25519-sha256 has been available since RFC 8463 in September 2018. In 2026 it is 0.183% of the signatures arriving here, which is about half the share of the algorithm that has been a MUST NOT since January 2018. All 39 of those signatures come from seven domains and every one of them is a mail software project or the personal domain of someone who writes mail software, several of whom post on this list. None is a provider or an ESP. That last point carries a selection effect I should state plainly: I receive ed25519 because I correspond with the people who implement it, which is also why the count jumps between 2025 and 2026. The effect does not weaken the observation. After eight years, the only ed25519 reaching one ordinary mailbox is the working group’s own mail.

I offer no conclusion on whether RSA should stay. The quantity I would keep in view is the one above. The observed time for a new DKIM signature algorithm to become visible at all outside the implementer community is longer than the interval we are planning across for a post-quantum algorithm.

Regards
Vittorio Moccia
_______________________________________________
Ietf-dkim mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to