Hi Steve,

On 10/09/2013 06:22 PM, Stephen Kent wrote:
> Stephen,
> 
> Thanks for creating a new thread to discuss this topic. It's a good
> starting point for
> 
> an important discussion.

I agree its important.

> I think MTU (vs. MTI) is a very hard argument to make, for several reasons,
> some of which I noted in my response to Dean.

Well, MTU vs MTI is not quite what the subject line says, but is
clearly one of the options worth discussing.

> Internet protocols are used in a very, very wide range of contexts,
> e.g., enterprises of various sorts, the IoT, and public environments. In
> principle one ought to make decisions about what security measures to
> deploy and enable based on a perception of threat. Threats differ in
> different contexts. Our current, MTI-based approach to security enables
> responsible parties in each environment to make decisions about what
> security to offer based on perceived threats and tradeoffs, e.g.,
> performance, processing overhead, user experience, etc.

All good.

> I admit that, in my experience, very few parties appear to make such
> decisions in a well thought-out fashion, but they could. 

That's the catch though isn't it? They don't, for whatever reason(s)
as seems to be shown by the SIP discussion.

I conclude that that means we're doing something wrong. (Maybe we're
not the only ones doing something wrong, but I do think we contribute
to the problem.)

> Suggesting that
> we can make such decisions for the folks who are ultimately responsible
> for the operation of services in a wide range of contexts strikes me
> hubristic.

Hmm. But we're happy enough to suggest that URI scheme names
can't contain a ":" and as a more relevant example, we don't have
a problem that web sockets requires a Sec-WebSocket-Key header. So
I don't buy the "we can't do that" argument to be honest.

> We've already made concessions for the Smart Grid context in the case of
> IPsec as MTI for IPv6. To me this suggests that we were persuaded that
> even MTI can be a barrier to adoption and deployment in some contexts.

But on the other hand maybe we were wrong there. Not sure myself.

> Pursuing a single set of MTUstandards for a wide range of contexts seems
> doomed to failure. 

Perhaps. But I'm not sure what a "single set of MTUstandards" means
to be honest. If we did have consensus for more than MTI then we
could clearly mess up in loads of ways;-)

> Generating MTU RFCs for various contexts might be an
> alternative. That would imply MTI protocol standards, augmented with
> BCPs. Is that what you envision?

Could be. Defo worth thinking about. Be great if folks had some
suggestions for situations where that might produce a better
outcome than we have today.

> Evaluating tradeoffs of security and privacy vs. other factors is hard
> when one deals with a wide range of contexts. For example, end user
> devices range from big servers to laptops, to tablets, to smart phones.
> Battery use if a big issue for some of these devices, as is bandwidth.
> Some of the more extreme TFS mechanisms discussed would have adverse
> implications for both. That's an example of why MTU, at the protocol
> spec levelk,
> strikes me as a bad idea.

Hm. I don't see why that applies to just this aspect of protocol
development. Sure, crypto involves some more CPU but that's not
that big a deal (far less than having the radio on in a challenged
device) and some round-trips which turns out to be a problem in
unchallenged-envirionments.

> There are a lot of middleboxes in the Internet. I am no fan of them; I'm
> a true believer in the e-t-e model for everything, not just security.
> But middleboxes are a fact of life and they exist because the folks who
> purchase equipment and offer services have found them to be operational
> necessities. Middleboxes provide ways to deal with backward
> compatibility and migration issues, broken implementations from vendors,
> security services for enterprises, network traffic engineering, etc. 

I'm surprised that you're convinced by the existence of bad practice:-)
But yes, middleboxes are there and we can't ignore that.

> If
> we argue that security against widespread nation-state surveillance is
> more important than all of these other considerations, I think we risk
> having the IETF be described as out of touch with reality.

Equally, if we ignored pervasive monitoring, we'd actaully be out of
touch with reality.

But there are definite trade-offs here. Its entirely true that we can't
improve privacy without affecting other things. One of my conclusions
from recent revelations is that we've not done a good enough job on
privacy so fixing that will involve such trade-offs if we're serious
about it and not just pretending.

S.

> 
> Steve
> 
> 
> 
_______________________________________________
perpass mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/perpass

Reply via email to