Hi Scott and Tony,

On Thu, Apr 23, 2009 at 2:07 PM, Scott Brim <[email protected]> wrote:
> Excerpts from Tony Li on Wed, Apr 22, 2009 08:43:00AM -0700:
>>
>> Hi Scott,
>>
>>> Oh.  Then I think we actually agree on procedure, but let's make it
>>> explicit:
>>>
>>>   The goal is to figure out which classes of solutions are
>>>   appropriate, and one input to that is to look at the implications
>>>   each solution class has on properties of identifiers.
>>
>> Agree.
>>
>>> Some of us have been saying that you can't do "properties of
>>> identifiers" in general, you have to talk about which kinds of
>>> identifiers, in which context.  I guess you agree.  If so, sorry, I
>>> hadn't understood.
>>
>> No, I think it's reasonable to talk about the classes of solutions
>> and  the reasonable properties of the (host, interface, stack, ...)
>> identifier that is bound to the locator within each class.
>
> This demonstrates my point: You are assuming the identifier of
> interest is one that is "bound to the locator", i.e. the thing that
> just the locator gets you to is what you want to describe in this
> exercise.  In this case one could just as easily envision having
> identifiers of interest name things at a finer level of granularity,
> where a locator gets you to a particular device (possibly virtual) but
> initial packets must include both locator and identifier, so that when
> the packet arrives at the device named by the locator, the device can
> figure out which particular communication endpoint (named by the
> identifier) you want to talk to.  The "identifier that is bound to the
> locator" could name the device but would not be interesting to us and
> would not be used in normal communications, only by device management.
>
> I'm happy to look at properties for identifiers that are "bound to
> locators" but let's not assume that each class has such things.  Also
> some classes do not say anything about endpoint identifiers at all.
>
I agree with Scott here.

For 3.3.2.2. Identifier variants
B2a Each host has a single identifer to which the locators are
attached. This identifier is used by the layer-4/5 and higher
protocols to compose the SID.

IMHO, the identifier should not be bound to the locator.
The reason is that you could create an Endpoint Layer  - as discussed
in the paper "Breaking Up the Transport Logjam” from
http://conferences.sigcomm.org/hotnets/2008/papers/HotNets2008proceedings.pdf
- between the network and transport layer in the stack. Then if the
identifier is used in a similar way as a HTTP cookie (RFC2965) in the
Endpoint Layer you might create a generic session state machine for
every application when the underlying locator(s) are changing. So when
a mobile client is changing its attachment point in Internet any
application session (not only HTTP) could be continued though the
locator(s) has changed - if the timers are allowing it.
Another place where you could use the "Endpoint cookie"=identifier is
in multihoming. A site has two separate service providers that are
offering two different Area Locators for the server that is using one
Local Locator. In DNS the server have two records, ALOC1:LLOC and
ALOC2:LLOC. The client is choosing ALOC1:LLOC for communication,
session is established and the identifiers (Endpoint cookies) are
exchanged but at some stage the session via ALOC1 fails. After a while
the client notice that the session has stalled but in the DNS cache
there is a second entry, ALOC2:LLOC. So the client stack decides to
try ALOC2:LLOC instead and the connection is re-established with the
help of the identifier (Endpoint cookie). Security issues arises, need
to look into that and if somebody has ideas feel free to make
proposals.

-- patte
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg

Reply via email to