[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [RRG] What do we have consensus on?




On Jun 19, 2008, at 1:41 AM, Tony Li wrote:


|through a phone chat with Tony, I finally got what's the above tried
|to say.
|It's all due to from which angle one looks at the picture.
|
|I take the view of scaling routing by provider-aggregatable prefixes:
|for a multihomed site, both SHIM6 and Handley proposal take the
|multiple provider-based addresses all the way *into* each host.  As
|far as routing scalability is concerned, it makes no difference
|whether these multiple PA addresses stop at a shim layer, or go to
|transport layer (of course it can make a big difference as far as
|transport functions are concerned).
|This is top-level branch#1 in my earlier msg
|http://psg.com/lists/rrg/2008/msg01220.html
|
|Tony's above msg took the view from transport layer and *looking
|downward*: SHIM6 has this notion of EID, and the actual IP packets
|*may* be sent using IP addresses that are different from EID, hence
|"this is a translation" view, i.e. a translation from an "EID" to an
|"IP address".
|(to be more precise, SHIM6's notion, as I understand it, is really a
|binding of a set of IP addresses to one transport session, so anyone
|in the set can be used. But lets not argue this semantic detail).


Agreed. So if we look at things from the network side looking in, we have
the classes where:

- Locators propagate all the way to the host
- Locators are terminated at the site border

From the feedback that we got from the SHIM6 discussion, there really aren't enough tools today to deal with the case where locators propagate to the host. This isn't to say that it couldn't be fixed, just that there's an
(engineering?) issue to be dealt with.

Do we agree with this separation?  And is it useful?

While I agree with the separation as a conceptually useful distinction, for me it isn't really the "top-level branch" in the architectural tree because one could always put a map-n-encap module in the routing layer of each host and then there would be three host- oriented approaches: - topologically-aggregatable addresses and EIDs inside the routing layer of the host. Only EIDs visible to transport - topologically-aggregatable addresses mapped to/from an EID in a shim layer between routing and transport layers of the host. Only EIDs visible to transport - topologically-aggregatable addresses directly managed by the transport layer of the host

Others will of course see this differently, but for me the top level branch is in fact whether the solution *requires* code and state in the hosts, or conversely can be done by routers with the hosts totally oblivious. A consequence of this branching is that if you require new code in the hosts, you can exploit state in the hosts, which has benefits which have been articulated many times on this list.

In the taxonomy above, only the first approach shields the hosts from the mapping/translation/whatever. That's pretty fundamental in my book.

DaveO.

Tony


--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg


--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg