On Apr 3, 2009, at 10:09 PM, Tony Li wrote:
Hi Dow,
I am ok defining things this way. To ensure we're on the same
page, this means that only the network part of an IP address (from
the perspective of the host) can ever be considered a locator, right?
I think we're aligned, but there are some nits that happen here
because the IP address again this overloaded use of both locator and
identifier.
To demonstrate a nit: what's the network part of the IPv4 address?
I have a prefix 192.168.1/24 that points to a particular Ethernet.
I also have a host 192.168.1.100 that is NOT on that Ethernet. I
establish host routes everywhere I need access to this host. I'd
claim that 192.168.1.100/32 is both a locator and identifier.
Such is the way of semantic confusion.
Agree (I think). I would claim that all 24 bits of 192.168.1.100
could be considered to contain location information, but other hosts
attached on the 192.168.1/24 segment would only have 24 "locator
bits". However, I think this means that it is the combination of the
(physical) point of attachment, and how that point is described in the
routing system, that fully define location.
If the topology were a perfect binary tree, and nodes were addressed
in a manner that perfectly matched this tree, then you wouldn't need
additional information in the routing system. Since the topology is
not a tree, the address semantics (even in the best case of address
allocation) can only approximately describe topological location - it
takes the information in the routing system to complete the mapping of
address to topology.
Also, since the mask that is in effect (for forwarding) at various
points in the topology differs, then the part of the IP address
that we can call a locator also differs at various points in the
topology.
Similarly, I'd make the slightly different distinction that you
simply can't say what part of an IPv4 address is a locator without
full information about the particular point of attachment. Note
that whatever bits are relevant to the locator do NOT change over
time or the point where they are examined. The fact that the
observer has incomplete and insufficient information doesn't change
the real semantics of the bits.
I am ok with this, but would you then say that the semantics of LPM
are an integral part of encoding location information in the IP
address? (and subsequently, IP addresses were not locators prior to
CIDR?)
Note that this is more clear in other proposals (e.g., GSE) where
the locator is a fixed number of bits.
In other words, if we say that:
(a) on an ethernet segment (that maps to only a single prefix in
the IP routing system), only the network part of the IP address of
connected hosts is considered to be a (L3?) locator, then
(b) a similar property exists closer to the "middle" of the
topology/routing system if aggregation is used.
In (b), more specific prefixes "hidden behind" the aggregate are
topologically flat from the perspective of processes that are only
aware of the less-specific, aggregate mask.
Thus, I disagree with this, as with the above. Aggregation in the
routing system is (and should always be) independent of the
semantics of the namespace.
I think I agree from a definition perspective - that is - we can make
use of proximity semantics in the address space even if there is no
aggregation in the routing system. However, from an engineering
perspective, the only reason to employ such semantics in the address
is if you plan to aggregate (somewhere/somehow), right?
R,
Dow
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
