At 2:15 AM -0700 7/29/08, Bruce Lowekamp wrote:
>
> > overlay-node://22301203/
>
>so my understanding is that 22301203 is a node-id

Yes.

> >
>> that can be decorated with the context id, if there is no existing
>> context or there would be ambiguity:
>>
>> overlay-node://22301203/;context="http://introducer.example.net/\
>>    example.odd"?
>>
>> but it remains a pointer to the node.
>>
>
>and this is a pointer to a node-id with a reference to the overlay-pointer

Yes.



> > To point to a resource at that node, the draft uses this syntax:
>>
>> overlay-node://22301203/;context="http://introducer.example.net/\
>>    example.odd"?resource=service-instance
>>
>
>so this is where I get confused.  A "resource at that node" is an
>unusual thing to want a pointer to (unless you're interested in
>diagnostic information).

As an external pointer, it might be rare.  I think it would be more
usual in an internal context (possibly diagnostic, possibly to have
a URI pointer to a specific instance).

> What I would expect to want to be retrieved
>would be a resource in an overlay.


> > or
>>
>> overlay-node://22301203/;context="overlay://enrollment.example.org/\
>>    ;otype=pastry"?resource=example.iso
>>
>> when using the overlay URI scheme.
>>
>>> So if this is pointing to a resource, I would prefer the host-part of
>>> the URI (is that the right name now?) to be a search term rather than a
>>> hash, i.e. [EMAIL PROTECTED]
>>
>> For host part, I think you mean authority section.  The syntax allows
>> you to elide that, so that you can reference a resource within a context
>> without including any node.
>>
>>
>>> My understanding of the URI in Section 5 is that resource= would be a
>>> kind in reload's current terminology.
>>
>> It could actually be a filename, if you want to say something like
>> "I want the debian.iso file stored at this node id in this overlay
>> context"
>>
>> overlay-node://22301203/;context="overlay://enrollment.example.org/\
>>    ;otype=pastry"?resource=debian.iso
>>
>> I think this is different from "kind" in the current document.
>>
>
>kind is a type, like a certificate or a sip registration.  A resource
>name is something like [EMAIL PROTECTED]   So what I might want to do
>is retrieve the certificate for [EMAIL PROTECTED] from the example.com
>overlay.  Normally this is done by hashing [EMAIL PROTECTED] to produce
>a Resource-ID and then the peer responsible for that Resource-ID is
>located.  But since you don't know what peer will be storing that
>resource under normal conditions, you wouldn't want to look up a
>resource on a peer.  Evolving from your syntax, if I use a resource-id
>instead of a peer-id, what I could see would be
>
>overlay-resource://resource-id/;context=overlay-pointer;kind=cert
>
>One of my initial thoughts was that I would rather use the resource-name
>than the id since it's expected to be human-readable.
>
>although the ordering of this bothers me.  Hierarchically, wouldn't
>something like
>
>overlay://overlay-pointer/resource-name;kind=cert
>
>be a bit better from a top-down perspective?
>
>>> Separate from those questions, my other thought when reading this is
>>> whether the overlay specification needs to be different for something
>>> that is trying to direct a node to make a query to an overlay as a
>>> client vs a spec that is intended to be used by new peers.  I think
>>> reload would support a one-shot query as it exists now, but it certainly
>>> hasn't been a design focus.
>>
>> I'm not sure what you mean by overlay specification here.  The
>> overlay URI scheme and odd files?
>>
>
>I was thinking of the overlay-pointer type.  Essentially random
>wondering of whether there should be a shorter description available of
>how to simply make a query to an overlay as a client than is required to
>join the overlay as a peer.  Or maybe there would be more peers
>available for clients to contact to make query than are available to
>serve as bootstrap peers.
>
>Bruce

_______________________________________________
P2PSIP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/p2psip

Reply via email to