[TLS] Re: Discussion about the Deployment Decision

[Paul Romer <[e-mail]>]

Rich wrote:

> Paul, what do you want to happen?
> ...
> The IETF expresses an opinion by saying “RECOMMENDED=N”. Many in the IETF recognize that others — often those deploying and purchasing systems — will have a different opinion.

https://mailarchive.ietf.org/arch/msg/tls/i-3dX5MDa-BcGOJWMRrJxs0mIpk/


Rich,

Fair question. Here is what I want to happen:

1. I would like to see the group work its way out of its current "Lord
of the Flies" culture.

2. I would like to see its discussion become more civilized and scientific.

3. In the discussion, I would like to see more assertions about
specifics like the plausible one you make in the two sentences that
I've quoted. If the group reaches a consensus that what you wrote is
accurate, it follows that the people deploying and purchasing systems
are considering other factors beyond the recommendation flag. They
might value any estimate that the group provides of the effect that a
security downgrade from X25519MLKEM768 to MLKEM768 has on CPU cycles
per TLS 1.3 connection.

4. This estimate could be a useful input into deployment decisions,
but to be clear, I would like to see the group work toward a consensus
about this estimate mainly because the effort might be a helpful
confidence building measure as you rebuild your culture.

5. Not everything I want will be feasible, but I also want the group
to consider the full distribution of changes in CPU cycles per TLS 1.3
connection instead of a single location parameter for that
distribution.

6. After recognizing randomness and uncertainty, I would like to see
the group take account of risk aversion, which a simple cost-benefit
analysis neglects. The consensus from decision theory about the way to
formalize risk aversion is to frame a choice as an attempt to minimize
the expected value of a convex loss function (or equivalently, to
maximize the expected value of a concave utility function).


Paul