Re: <DKIM> Re: WPD Discovery Options

De: "Martin Thomson"

>On 4 November 2014 07:55, Martin Nilsson <nilsson@opera.com> wrote:
>> I think it is important to find a single method that works everywhere, if
>> possible. Once that is in place, it would be possible for other actors to
>> optimize that for specific domains. E.g. if we standardize on a DNS based
>> discovery, you could push the relevant records to the device in PDP context.
>> Higher layer code still only needs to know about a single method.

> I think that if anything has a chance of working everywhere, it would
> be a DNS-based approach (intentionally handwaving about RFC 6762
> here).  My experience here is that the operational hurdles are
> significant.  Discovery is hard in general because it relies on
> properties of the network below the IP layer.  That means dealing with
> all sorts of wireline and wireless access network technologies, the
> operational considerations imposed by each, and all the complications
> of middleboxes on those networks.

My experience is the same. Unless you have a technically homogeneous centralized network any solution that relies on network properties is not practical. Corporation-side part of the network growth will always be organic and omit such niceties as "advanced" network properties (for very basic understandings of "advanced").

Even DNS is not always reliable because local entities will deploy local caches that ignore original DNS TTLs (in the name of speed) or even override part of the naming.

DNS is the least bad solution but it is a very good idea to limit its scope as much as possible and keep the load balancing properties (for example) inside the WPD file. And when DNS fails, being able to replace discovery with pre-configured settings in the web client can be a lifesaver (otherwise people compound the problem with another layer of DNS overrides).

-- 
Regards,

Received on Wednesday, 5 November 2014 08:19:26 UTC