On 10/05/2016 08:54 PM, Peter Saint-Andre - Filament wrote:
> Over on the CORE WG list, we've had a little discussion about the
> desirability (or not) of unique identifiers for devices in the Internet
> of Things. The message below provides some context.
> I'd be curious to learn more about the attack surface lurking behind
> Stephen Farrell's comment that having long-term stable identifiers for
> IoT devices is a privacy-unfriendly practice because people will abuse
> such identifiers.
> To be clear, the scenarios I have in mind are not specific to CoAP and
> don't always involve IP-based networking (the technology I'm working on
> these days enables mesh networking over long-range radio), but they do
> involve discovery and eventual communication that is both end-to-end
> encrypted and as close to metadata-hiding as possible.
> -------- Forwarded Message --------
> Subject: Re: [core] Implications of IP address / port changes for CoAP & Co
> Date: Thu, 6 Oct 2016 00:11:26 +0100
> From: Stephen Farrell
> To: c...@ietf.org <c...@ietf.org>
> Hi Peter,
> On 06/10/16 00:03, Peter Saint-Andre - Filament wrote:
>> On 10/5/16 4:28 PM, Stephen Farrell wrote:
>>> On 05/10/16 23:22, Dave Thaler wrote:
>>>> It is important that every device have a unique UUID that is
>>>> endpoint-address-agnostic and protocol-agnostic.
>>> Considering the privacy implications I'm not at all sure I'd
>>> accept that argument. In fact I'd argue we ought encourage
>>> that devices not have globally unique long-term identifiers at
>>> all unless there is a real need for those, and unless we
>>> understand how to control their (ab)use.
>> By "identifier" do we necessarily mean "network identifier"? It seems to
>> me that it is useful to have a unique long-term identifier for every
>> device, based on its public key. Whether you can obtain a network
>> connection to that device based on such information is another story.
> It is undoubtedly useful to have long term stable identifiers of
> various kinds. I'd include key IDs and public keys as such.
> Turns out that it's also fairly universally privacy unfriendly
> as people will abuse such identifiers for good and bad reasons.
> So I think we need to get much better at analysing when such
> things are really needed and in what scope. My bet is that a lot
> of the time a locally or probabilistically unique more transient
> identifier would be just fine.
> But yeah, I can't prove that. OTOH there is a hint in the term
> "IMSI catcher" isn't there?
At the risk of sounding our own horn, draft-gont-numeric-ids-generation
might be useful for guidance.
For instance, our I-D essentially argues that you should be asking
yourself the question: "what are the interoperability properties you
need for such IDs?".
PGP Fingerprint: 6666 31C6 D484 63B2 8FB1 E3C4 AE25 0D55 1D4E 7492
perpass mailing list