[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

comments on rollover requirements



Referring to: http://www.ietf.org/internet-drafts/draft-ietf-dnsext-rollover-requirements-01.txt

The document doesn't describe the usage scenarios that we will face - such as close-knit anchor management versus DLV-type anchor management, or complete compromise versus singleton compromise of a zone's key set.

Quite frankly, if amazon.com has three keys and one is compromised, I won't be able to discern which one. The news will probably be broadcast on some TV station - what news anchor will be able to spit out which key is the compromised one. Maybe if this were all automated, the difference between complete and singleton compromises is significant, but if there's any manual intervention, I don't thin there is a significant difference.

On to comments on section 5...

5.1  Scalability

This section ignores scalability from the point of view of the zone operator. The only mention is how many anchors a resolver will configure.


No solution ought to be dependent on how many resolvers have the anchor configured. There's not necessarily a relationship between a zone operator and the resolver operators. E.g., the resolvers run at AOL (probably) do not have a special arrangement with .ZA. No business link (like even TLD to registrar) at all.

5.2 IP Encumberance

Is this really the second most important point? That's my only comment on this.

5.3 General applicability

IOW, don't special case the root zone.  What more is there to the requirement?

5.4 Support private networks

That sounds like 5.3. All zones ought to be eligible to be anchors. And all zones have parents except the root. (Where would a DS RRSet for the root zone sit?)

Is this section supposed to be a stub for split-zone anchoring? Can the support be to "turn off" validation, as in a null DS record?

5.5 Detection of Staleness

When you live on top of an unreliable protocol (UDP), you can never "detect" anything reliably. The lack of an obtainable DS RRSet may be the result of a disconnected network and not removal of it.

5.6 Manual operations permitted

I think that it should be obvious that an operator can always disable/ignore any automatic attempt to update the configuration of their server. Even the software updating mechanisms for my laptop always ask permission before enacting an upgrade or using upgraded software.

5.7 Planned and unplanned

In an automated protocol, there's not much difference, is there?

5.8 Timeliness

"Distributed to security-aware resolvers" "timely" highlights the missing portion of 5.1 on scaling.

5.9 High Availability

This is needed, and made me wonder why we don't rely more on the CERT RR.

5.10 New RR types

5.11 Support for maintenance

This sounds like the functional requirements.

5.12 Recovery from compromise

This section is lacking substance. It basically says "if the problem is easy to solve, then it should be solvable." No discussion on whether there is a discernable difference between singleton failures or mass failures.

After reading this, in general, what is the interop problem? It is "how does a zone make it's public key (data) known?"

On the one hand, I don't want to reinvent the wheel - meaning, why not rely on the CERT RR, a CA infrastructure, and maybe even a MIB for servers? (Yeah, yeah, I know DNS'ers have this thing about MIBs.)

On the other hand, how much of this is a detail for implementers? Each server has it's own configuration interface, and trust anchors are just configuration entries?

On another hand, do I want to let something as sensitive as trust data be able to be silently updated, especially if it is something I may not understand? (Reading cryptobytes is plaintext hard.)

On the other another hand, if I can do this in band, then the trust data passes along a path as safe as the application data (i.e., fates are shared). But the down side is that if the app data can't get through, the trust data can't either.

I guess I am still leery of putting this mechanism only in to the port 53 traffic stream. But it has it's advantages.

I just don't see this document describing the problem we are setting out to solve.

--
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Edward Lewis                                                +1-571-434-5468
NeuStar

Nothin' more exciting than going to the printer to watch the toner drain...

--
to unsubscribe send a message to namedroppers-request@ops.ietf.org with
the word 'unsubscribe' in a single line as the message text body.
archive: <http://ops.ietf.org/lists/namedroppers/>