Re: [rrg] GLI-Split Locator Gleaning explanation

[Robin Whittle <[e-mail]>]

Here are two other questions about things I can't yet understand in
the proposal (2009-12-22 version):


http://www3.informatik.uni-wuerzburg.de/~menth/Publications/papers/Menth-GLI-Split.pdf

According to section 2.3.2, a host application running on a GLI-host
(all applications are the same with GLI-split - the same as before,
with no new functions, API, IP address or DNS behavior etc.) can look
up a FQDN in the DNS, and will always receive an IP address with a GL
locator and the GAP bit set.

The application opens up a TCP socket with this address.

Now consider the top left of Figure 4.  This is the transport layer
(the TCP layer) in the sending host.  The source and destination
addresses are shown as being in "Identifier" format: without any GL
or LL and without the GAP bit set.

I am trying to write up my own understanding of how GLI-split works,
from a GLI-updated host in one GLI-domain to a host in another
GLI-domain, when that recipient host is either GLI-updated or an
ordinary IPv6 host.

One set of explanations, around pages 6, 7 and 8, refers to the hosts
(nodes) as if they were black-boxes.  Another explanation in section
3.2.x refers to what happens inside the host stack, in a general way
- but not in reference to all the various modes of operation of the
GLI-Split system.

I am unable to reconcile the application using an address with a GL
and GAP bit set with the Figure 4 and 3.2.x explanation.


Hosts today can have multiple IP addresses, so they can behave, to
the outside world, like multiple hosts (since in today's IP naming
model, the identity of a remote host IS its IP address).

How does GLI-Split work when a host has multiple Identifiers - which
would be the equivalent of a host having multiple IP addresses today.

  - Robin