RE: Informal meeting on blind caching

[Mike Bishop <[e-mail]>]

I would go a little further and say that, at least initially, how the closer node gets the content is entirely an implementation detail and out of scope for the IETF, regardless of working group.

While you’re right that this isn’t restricted to CDNs, that’s the mental model that makes it easiest to understand.  There are several use-cases, each of which will have a different means of acquiring the content – and those means don’t need to be specified by this group in most cases.  If the origin is a CDN while the less-trusted server is a CDN node in an unsecured location, the content acquisition is entirely internal.  For the blind caching case, both servers are the same machine, so the acquisition is internal in an even-more-literal sense.  If the origin is a customer while the less-trusted server is a CDN, that mirrors existing deployments; we don’t seem to have a problem configuring where the CDN goes to get updated versions of a cached resource today, and I would imagine similar techniques could be applied here satisfactorily.

If use-cases arise beyond those that require an explicit “how do I find encrypted content <foo>?” protocol, we can pursue it then.  Let’s not muddy the waters in advance.

From: [e-mail] [mailto:[e-mail]] On Behalf Of Erik Nygren
Sent: Thursday, July 28, 2016 7:32 AM
To: Richard Bradbury <[e-mail]>
Cc: [e-mail] Group <[e-mail]>
Subject: Re: Informal meeting on blind caching

For the "Blind caching in a CDN edge cache that is acting as a delegated origin server" scenario I think we'd be better off separating out the high-level mechanism (ie, getting oob-encoding) right from the actual business model and implementation details for how the caches get populated.  The former is something we can tractably do here (have a way to delegate content bodies out to third-parties) but the latter is something that could easily fill up its own working group to get all of the details worked through.

I also think that we shouldn't refer to "CDN" in this scenario as much as referring to delegating between a server acting as an authoritative origin and a collection of servers at a lower trust level.  (In some deployment scenarios, the former might be a CDN and the latter might be either a different trust tier of CDN nodes or federated-but-less-trusted partners.)
The specific blind caching with a configured proxy cache is a well-defined use-case, but the others have many different ways to implement that trying to standardize something now before there has been more experience with implementations and business models seems like asking for trouble (and is something that some of us may be less interested in than in defining common underlying mechanisms and affordances to allow developing technologies in this area).

        Erik



On Thu, Jul 28, 2016 at 5:07 AM, Richard Bradbury <[e-mail]<mailto:[e-mail]>> wrote:
On 27/07/2016 11:33, Richard Bradbury wrote:
On 17/07/2016 18:03, Martin Thomson wrote:

We are having a meeting on blind caching on Tuesday 08:30 in

Charlottenburg I.  Folks who are interested are invited to join us.



Rough agenda is to discuss what this is, the status of specs and

implementations.  Then we are looking to get some input on the work

that has been done and what might need to happen next.



For those not familiar with this, the following drafts are recommended reading:

https://tools.ietf.org/html/draft-thomson-http-scd

https://tools.ietf.org/html/draft-thomson-http-bc

https://tools.ietf.org/html/draft-reschke-http-oob-encoding

https://tools.ietf.org/html/draft-thomson-http-mice

https://tools.ietf.org/html/draft-ietf-httpbis-encryption-encoding


Hi. Thanks for an interesting session last week.

Based on my reading of the Internet Drafts I think these are the two main Use Cases under consideration here:

  1.  Blind caching in a CDN edge cache that is acting as a delegated origin server.

     *   Described in draft-thomson-http-scd.
     *   Making use of the techniques described in draft-reschke-http-oob-encoding, draft-thomson-http-mice and draft-ietf-httpbis-encryption-encoding.

  1.  Blind caching in an explicitly configured proxy server.

     *   Described in draft-thomson-http-bc.
     *   Making use of the techniques described in draft-reschke-http-oob-encoding, draft-thomson-http-mice and draft-ietf-httpbis-encryption-encoding.

The discussion in Berlin last Tuesday morning flowed fairly freely between the two Use Cases, so I just wanted to check my understanding.


OK. Now I feed I understand Use Case 2 better, I attach a couple of sequences describing my (incomplete) understanding of Use Case 1 above. The main question in my mind here is whether the resources are explicitly pushed into the CDN cache (what I have called "Mode P") or whether they are retrieved on demand from the origin server by the CDN only on cache miss ("Mode G").

Perhaps you are looking to support both modes of operation? The Internet Draft could usefully clarify this.
Mode G
In the cache pull-through mode, there appears to be a requirement for the CDN to know the resource mapping in order to support the URL obfuscation feature. But that appears to negate one of the key benefits of the feature – keeping the mapping secret from the CDN provider. I'm clearly missing something here. Or maybe URL obfuscation just doesn't fly in this mode, which would be a shame since this is our primary mode of operation for some media types.
Mode P
With the push mode, the resource mapping can remain secret to the origin server, which is great. But the mechanism for pushing content onto the CDN remains a bit of a mystery, so I have made an educated guess on my diagram. Perhaps you intend this mechanism to remain private? In which case, perhaps the Internet Draft could include a note to this effect?

Clarifications and comments gratefully received.

Regards,

--

Richard Bradbury | Lead Research Engineer

BBC Research & Development

Centre House, 56 Wood Lane, London  W12 7SB.

T: 0303 040 9672  F: 020 8811 8815