Juan,
Since you made some good comments for draft-retana and mentioned RFC
5614, I thought I would introduce the following update for RFC 5614
(OSPF-MDR) by addressing some of your comments as they would apply to
it.
http://www.ietf.org/id/draft-ogier-ospf-manet-single-hop-00.txt
Please see my comments below.
Juan Antonio Cordero Fuertes wrote:
...sorry for not changing the subject.
Juan Antonio
Hi
Alvaro,
Some comments/thoughts/questions about
draft-retana-ospf-manet-single-hop-01
below. Thanks in advance for your attention.
0) The position of routers in the networks addressed in this interface,
is
fixed or may move as long full connectivity (1-hop) remains? In section
6 it
is said that these are fixed networks, but this characteristic is not
listed
when describing single-hop broadcast networks in section 1.1.
*About Hellos:*
1) Incremental Hellos are not used (required) in this proposal.
However, in
a single-hop MANET, (full) Hellos of any router list all routers in the
network every HelloInterval seconds. Since the topology is static (not
considering wireless link failures or changes in the link cost), may it
be
worth to explore a Hello optimization mechanism? Not only incremental
Hellos
but also, possibly, differential Hellos as specified in RFC 5614.
I agree, and have included the option for differential Hellos in
the update for RFC 5614.
*About Smart Peering:*
2) The proposed heuristic in section 3, is expected to replace or to
complement the one presented in RFC 5820? Might be useful to make it
more
explicit. I assume that the proposed heuristic comes right after the
one in
RFC 5820 -- otherwise the number of adjacency-forming processes might be
higher, some of them redundant. Is that correct?
The update for RFC 5614 describes two new procedures, both of which
are optional. One is a new option for constructing LSAs for a
single-hop network, which is interoperable with the other LSA options
and is similar to draft-nsheth. The other is a simplification of the
MDR selection algorithm for a single-hop network, which gives exactly
the same result as the original algorithm when the MANET is a
single-hop network (and is therefore interoperable with it).
3) Wait Time[r] mentioned in section 3 is defined in RFC 2328 as
RouterDeadInterval. Why tying the waiting time before running the SP
state
machine to such RouterDeadInterval value? Wouldn't HelloInterval be
sufficient? If the goal is to ensure that the router has information
about
all neighbors from the network, a user-configurable parameter within the
interval [HelloInterval, RouterDeadInterval] may be useful.
4) If I'm understanding correctly, the number of adjacencies maintained
by a
router may be higher than the parameter "maximum number of adjacencies".
Consider for instance MAX_ADJ=2 and 4 routers in a single-hop MANET
with ids
(actually, (RtrPri,RtrId)) 1,2,3,4, that appear in the network (are
discovered) in order 4-3-2-1: 4 has three adjacencies. Seems that the
"maximum number of adjacencies" corresponds to the number of REQUESTED
adjacencies for a router, which is <= than the final number of
adjacencies,
given that routers accept passively any adjacency-forming request.
When OSPF-MDR is run in a single-hop network with AdjConnectivity =
2, every non-MDR/BMDR router will have exactly two adjacencies (one
with the MDR and one with a BMDR). The MDR will have n-1 adjacencies
(same as the DR in OSPF).
5) It is not clear for me in which sense the proposed heuristic is
deterministic. Consider again a 4 nodes single-hop network with ids
1,2,3,4.
Depending on the order of appearance, the adjacency subgraph changes
(e.g.,
for an appearance order 1-2-3-4 the adjacencies 12, 23, 34 are formed;
for
the appearance order 4-3-2-1 the formed adjacencies are 14, 24, 34).
That
is, the adjacency map cannot be extracted from the final topology of the
network.
I agree that the heuristic for draft-retana is not deterministic.
But the same is true for an OSPF broadcast network, since the first two
routers that appear will be the DR and BDR.
In OSPF-MDR, if one router has higher RtrPri than the others, it will
always be the MDR regardless of order of appearance.
Regards,
Richard
|
