Andrea has a good example of an L3 link that consists of 802.11 link
and Ethernet link, so let me discuss based on the example.
Obviously there can be multiple UEs attached to the same AP
something
like:
802.11 link Ethernet link
UE1--------------------- AP ------------------- AR
/ /
UE2-------------------+ /
/
UE3------------------+
Do you call UE2 and UE3 "L3 PoAs of UE1" just because they are
adjacent nodes for UE1 in terms of L3? I have a problem with
calling
UE2 and UE3 as PoAs because UE1 is not really attaching to UE2 and
UE3. In fact, there is no notion of attaching to a particular node
in IP (though there is a notion of attaching to a network in DNA
WG), so I think it is hard to have an agreement on bringing the
notion of PoA to IP layer which is not defined in the IETF.
Please see my further comment below.
On Mon, Oct 03, 2005 at 12:50:46PM -0400, Andrea Francini wrote:
Hi Mike,
I really appreciate your comments. I think they are moving this
part of the discussion (definitions and their necessity) in a good direction.
As I wrote in earlier messages, I recognize well that the
definitions I proposed have plenty of room for improvement. Your
comment points out that now also the notion of adjacency would
need to be explicitly spelled out for a better common understanding.
So this time, instead of generating another controversial
definition on top of the previos ones, I'll take a step back and
try to be more clear with the problem I would like to see solved.
We started with the attempt to converge on a good definition of
Point of Access (PoA).
Why do we need such a definition?
1. We need a name for the point in the network from which the UE
obtains connectivity.
2. We need a name for the first point in the network with which
the UE can exchange MIH-related information (for any of the three MIH services).
Can a single definition of PoA address both needs at the same time?
I don't think so, because a single definition cannot capture all
possible combinations (in particular, it cannot provide a good
distinction between the cases where MIH is immediately available
at the AP/BS and where it is only available deeper inside the network).
This is why I thought that it made sense to have a general notion
of PoA (the first point in the network from which the UE obtains a
certain service) and then further qualify it with the appropriate
attributes (L2, L3, and MIH are the ones that I originally
considered relevant in the
802.21 context; TCP can be another one).
Still in general terms, being the PoA the first point in the
network that the UE can use as a peer for message exchanges
related to a certain service, defining the entity that exists
between UE and PoA with respect to that service is the next step. I thought that "link"
could be a good term, although I realize more and more that "link" has a strong L2 connotation that may not be easy to relax.
Whatever term we use, it must convey the notion that the UE and
the PoA are next to each other on it, otherwise the PoA is no
longer "the first point in the network". I used "adjacent" in the
definition of link, with the idea that the adjancency is not
necessarily physical, but only logical within the same context in which the service obtained from the PoA is defined.
Think for example of a network of routers that is built on top of
a network of ATM switches. I consider two routers adjacent if they
are on IP hop away from each other, independently of the number of
ATM switches that exist in the physical path between them.
Similarly in 802.21, two L3 entities are adjacent if no other L3
entity exists in between, independently of the number of L2 segments (or "L2 links") that connect them.
In a situation like the following:
802.11 link Ethernet link
UE --------------------- AP ------------------- AR
how can we capture the fact that the Access Router (AR) is the
first
L3 entity in the access network path of the UE?
With the definitions I proposed, the 802.11 and Ethernet links are
L2 links, the AP includes the L2 PoA for the UE, the AR includes
the
L3 PoA for the UE, and an
L3 link exists between the UE and the AR (of course only after the
UE has obtained its IP address).
If the 802.11 interface on the AP is MIH-capable, the AP also
includes the MIH PoA for the UE, independently of the MIH
capabilities of the AR. If only the Ethernet interface on the AR is MIH-capable, the AR includes the MIH PoA.
With the definitions I proposed, every entity of the example gets
an unambiguous name. I don't have any problem with changing the
names or the wording of the definitions, but I still want to have
a name for each of the entities. Any other proposal should pass the same test.
I think that unambiguous naming is also possible even if we call the
immediate L2 link to UE as "link", and the end point of the
immediate
L2 link as "PoA", for example:
If the 802.11 interface on the AP is MIH-capable, we can call the AP
the MIH PoA (or MIH AP) for the UE, independently of the MIH
capabilities of the
AR. If only the Ethernet interface on the AR is MIH-capable, we can
call the AR the MIH AR.
On the other hand, regardless of which direction we take (i.e.,
defining link and PoA for each layer vs. defining link and PoA for
L2 only), definition of the "immediate L2 link for the UE" still
depends on each media, e.g., a PDP context can be the immediate L2
link for GPRS, and an association between a STA and an AP can be the
immediate
L2 link in 802.11, etc. I think this is more important thing to
consider.
With respect to your specific concerns:
I think the link between two adjacent layer 3 entities is actually a layer 2 link. To me, the > purpose of any link at layer N is to provide a PDU transfer service to layer N+1.
Do you mean that two layer-3 entities are adjacent only if they
are connected by a single L2 link, i.e., only if no other L2 node
exists between the respective
L2 interfaces? What term could then be applied to the two L2
interfaces to express the fact that no other L2 node exists between them?
In the IP-over-ATM example, two neighboring routers can be
attached to ATM nodes that are connected by a single ATM VC or by
a chain of ATM VC's. With my definition of adjacency, the two
routers are adjacent in both cases (i.e., neighboring = L3-adjacent). Would it be the same with your definition?
I'm not trying to say that they should necessarily be called
adjacent, but to understand how your definition would apply to the example.
If you took your definitions as they are, then the layer 3 link coming up would not allow TCP > or UDP to flow - you'd still have to wait for IP address assignment - and that sounds wrong to > me.
I think it should be implicit to have IP addresses in place before
an L3 link can exist. If the implication is not clear, an explicit
statement can be easily added.
For IP, just having an IP address is not sufficient to claim that
there is L3 connectivity between an UE and an IP node on the same IP
link. In IPv6 Neighbor Discovery protocol (RFC 2461), at least
there must be a neighbor cache entry for the communicating node with
"REACHABLE" state to *roughly* claim that the node is reachable
(within ten seconds ago).
Regards,
Yoshihiro Ohba
Thanks,
Andrea
Mike Moreton wrote:
Andrea,
I really like these definitions - I think they are clear and precise, which gives a good basis to argue from.
Which is what I'm going to do!
I think the link between two adjacent layer 3 entities is actually a layer 2 link. To me, the purpose of any link at layer N is to provide a PDU transfer service to layer N+1.
If you took your definitions as they are, then the layer 3 link coming up would not allow TCP or UDP to flow - you'd still have to wait for IP address assignment - and that sounds wrong to me.
Mike.
-----Original Message-----
From: Andrea Francini [mailto:francini@LUCENT.COM]
Sent: Friday, September 30, 2005 8:41 PM
To: STDS-802-21@listserv.ieee.org
Subject: Re: [802.21] [DNA] Prefix information for link
identification in DNA
Hi Yoshihiro,
I definitely don't mean to contradict what I wrote yesterday.
I still think of
the PoA as a link endpoint.
Your comment rightly brings up the necessity of providing a
clear definition of "link" since link and PoA are tightly
inter-related.
With a generic definition of PoA as a link endpoint, defining
"L2 PoA", "L3
PoA", and "MIH PoA" implies corresponding definitions of "L2
link", "L3 link", and "MIH link".
I assume from now on that a layer-agnostic notion of link is
accepted and that "link" is not strictly a Layer-2 notion. The
group can debate if this is a valid assumption. If not (i.e.,
the group prefers to assign a strong L2 flavor to "link"), we
can find a better term (e.g., "connection", or
"relationship") and
base on the new term both the generic and the specific
definitions of PoA. In this latter case, "link" would be
synonymous of "L2 connection" (or "L2 relationship", or
whatever other term the group may identify).
I can think of the following generic definition for a
layer-agnostic link:
"Communication relationship for the exchange of messages
between adjacent peer protocol entities."
Where:
"Peer protocol entities" always belong to the same protocol
layer (e.g., L2, L3, MIH).
"Adjacent" emphasizes that there is no other interposed peer
entity between the ones that terminate the link (e.g., there
cannot be another
L3 entity between
the endpoints of an L3 link; if such entity is present, there
are two and not one L3 links). This does not prevent a link
from having more than two endpoints:
in a multicast link, for example, all endpoints are adjacent
to each other and none of them is necessary to enable
connectivity between others.
The layer-specific definitions easily follow:
L2 link: "Communication relationship for the exchange of L2
messages between adjacent L2 entities."
L3 link: "Communication relationship for the exchange of L3
messages between adjacent L3 entities."
MIH link: "Communication relationship for the exchange of MIH
messages between adjacent MIH entities."
Having the notions of "L2 link", "L3 link", and "MIH link" in
place, the PoA definitions I previously proposed can easily be
mapped as follows:
L2 PoA: network-side endpoint of L2 link involving the UE
L3 PoA: network-side endpoint of L3 link involving the UE MIH
PoA: network-side endpoint of MIH link involving the UE
As for identifying the endpoint entity as part of a network node:
The L2 PoA is an L2 interface on the network node, identified
by an L2 address.
The L3 PoA is an L3 interface on the network node, identified
by an L3 address (on a router, the same physical interface can
co-locate L2 and L3 interfaces).
The MIH PoA is an MIH interface on the network node, i.e., an
interface (either
L2 or L3) with which the MIH function of the network node is
registered for any of the MIH services. When referring to both
transport and MIH capabilities of the interface, we may have
an
"L2 MIH PoA" or an "L3 MIH PoA".
The main purpose of the endpoint vs. node distinction in the
PoA definition is to avoid ambiguities when the same network
node can terminate multiple links and present for each of them
different capabilities and behaviors (i.e., MIH capability can
be activated on one interface and not on another, or the node
can be a hybrid L2/L3 box with both L2 ports and L3 ports).
Defining the PoA with
respect to a specific link (or connection) brings the focus of
the PoA definition on the functionality that the corresponding
UE can obtain from that point in the network, without
requiring any unnecessary assumptions on the overall nature of
the network node that includes it.
While I am sure that the wording for the definitions I am
proposing can be dramatically improved, I am convinced of the
absolute necessity to single out the respective entities and
provide clear definitions for each of them.
Thanks,
Andrea
Yoshihiro Ohba wrote:
Andrea,
The PoA definition below is going to the direction that the
notion of
PoA is less associated with the notion of "link", as opposed
to what you made in your previous general statement which I
have
fully agreed.
Or you may be introducing a new definition of "link" as "a
specific type of communication relationship", which seems to
be too
ambiguous.
Yoshihiro Ohba
On Fri, Sep 30, 2005 at 11:25:04AM -0400, Andrea Francini wrote:
Trying to finalize one part of the ongoing discussion:
the PoA definition.
I have the impression that some people consider the
capability of supporting MIH
as part of the definition of PoA, while other people
don't, giving it only a
network connectivity value.
What about the following:
1. General definition of PoA:
a. "PoA is the first point in the network that acts as
the UE counterpart for a
specific type of communication relationship (e.g., L2, L3, MIH)."
2. Accordingly, the following three specific definitions
could be added:
b. "L2 PoA is the network-side endpoint of the L2 link by
which the UE connects
to the network."
c. "L3 PoA is the closest network counterpart for the UE
that requires an L3
address to be identified in UE-generated messages."
d. "MIH PoA is the closest network counterpart of the UE
for MIH exchanges."
Thanks,
Andrea
"Stefano M. Faccin" wrote:
Peretz, nobody denies that. The issue here is that what
you have been saying doe not allow for deployments that do not
use any MIH services at L2. Even if you may not believe these
deployments will happen, there are vendors and operators that
do believe that their networks will only use MIH services at
L3, at least for the initial deployments.
Thjerefore our model and definitions must allow for this. In
this model, there is no MIH @ L2, and the PoA is in the subnet
where the UE gets its IP address.
Stefano
________________________________