Minutes of the Mobile Ad-hoc Networks (manet) Workging Group.

The manet WG held two sessions of two hours each.  

First Session:

The first manet session began with the ritual agenda bashing which resulted in little or 
no change to the proposed agenda.  Implementations of various manet protocol proposals 
now exist and status was briefly updated relating to the various Internet Drafts (IDs) at 
the beginning of the meeting. Subsequently, new proposals and protocol enhancements 
were presented by various parties. 

1) Cluster-Based Routing Protocol Presentation:

Mingliang Jiang from the University of Singapore presented the Cluster-Based Routing 
Protocol (CBRP).  The protocol concept is based upon a well-known clustering 
algorithm developed at the Naval Research Laboratory and the University of Maryland, 
and also recently adopted in work at UCLA. The algorithm constructs a set of node 
clusters, each cluster having a special node designated as the clusterhead, with the 
property that every non-clusterhead node is directly connected to a clusterhead.  The 
algorithm also builds a tree consisting of clusterheads and nodes (termed gateways) 
interconnecting the clusterheads of neighboring clusters.  The unique design motivation 
of the proposers of CBRP is to use the shared tree to reduce the overhead of the control 
packet flooding required by dynamic source routing. In CBRP, while the shared tree is 
used for more efficient flooding the actual routing of data is performed using a form of 
dynamic source routing that is similar to the scheme proposed for ad hoc routing from 
Carnegie Mellon. In CBRP, source routes are constructed from a source through a set 
of intermediate clusterheads to a destination. Usage of source routing is considered 
desirable in that it has the ability to make use of end-to-end routes which consist fully 
or partially of unidirectional links.  The presenters indicated that IMEP may be adopted 
for link status sensing functionality.  Also, it was pointed out that early cluster-based 
routing (CBR) work done at the Naval Research Laboratory in the mid 1980s may 
provide some optimization in reducing the number of clusterheads formed overtime 
with this sort of algorithm.

2) Optimized Link State Presentation:

Amir Qayyum from INRIA presented a draft on an Optimized Link State (OLS) 
algorithm.  The approach presented is optimized in the sense that it does not require 
propagation of full topology knowledge to all nodes, but still guarantees the generation 
of shortest-path routes to all nodes.  The algorithm is intended to run atop IMEP, and to 
utilize multipoint relaying concepts (presented at the manet WG, 41{\super st} IETF in 
L.A.) to disseminate topology information more efficiently.

3) Ad Hoc Multicast Routing Presentation:

Rajesh Talpade from Bellcore presented a draft on the Ad hoc Multicast Routing 
Protocol (AMRoute). The basic concept behind the protocol's design is to have only 
routers to which multicast hosts are affiliated maintain multicast forwarding tree state. 
Both senders and receivers are on this bi-directional, shared tree. This is accomplished 
by tunneling multicast packets between such routers. This has the advantage of being 
independent of the underlying unicast routing protocol.  Another feature of the protocol 
is that while it is core-based, they are not cores in the traditional sense (e.g. CBT) in that 
they are not central points for data distribution.  These cores assist in member detection 
and tree formation, and can dynamically migrate among the members nodes. Thus the 
cores serve a group management functionality, making this an interesting aspect of the 
protocol. 

4) Carnegie Mellon Simulation Effort:

Dave Johnson from CMU presented an overview of work he and several others have been 
doing in adding mobility extensions to the NS-2 simulation toolkit and beginning to 
simulate particular MANET protocols. The NS-2 simulator is a desirable software 
platform for manet routing simulation in a number of respects, but previously did not 
have necessary mobility modeling features (see manet 40{\super th}IETF minutes).  
CMU's contribution consists of two parts: generic MANET simulation objects such as 
wireless channel and mobility models, and a first cut at simulating several of the proposed 
manet routing protocols including DSR, DSDV, AODV and TORA.  Dave indicated that 
they were planning on making these models available to the rest of the working group and 
community soon.

5) Sun Microsystems Simulation Effort:

Charlie Perkins from Sun also briefly presented work he has been involved with on 
developing MANET simulation technology for the NS-2 simulator. While the motivation 
is the same as for the CMU work, the two approaches differ in some of the details and 
level of simulation.  For example, the CMU channel model is possibly being aimed at a 
highly accurate representation of the signaling environment, whereas the Sun approach is 
simpler, choosing to model the channel at a coarser level of fidelity.  Also, the two efforts 
have initially developed slightly different mobility models for MANETs. 

Second Session:

The meeting with a short introduction and review of the planned agenda by Joe Macker.

1) LAM Presentation

Scott Corson from the University of Maryland presented a draft on a Lightweight 
Adaptive Multicast (LAM). LAM creates and maintains a group-shared forwarding tree 
for the group. Conceptually, the protocol can be viewed as a fusion of CBT and TORA. 
The design of the protocol follows the concept of vertically-coupled design to achieve 
efficiency (low overhead, fast reaction), which is desired in a MANET environment. 
LAM is specifically built and reliant upon TORA features, and serves as an integrated 
component of an envisioned IMEP-TORA-LAM manet routing suite.  The presenter 
indicated that this coupling enables LAM to benefit from TORA's mechanisms while 
reacting to topological changes. Also, during periods of stable topology and constant 
group membership, the LAM protocol does not introduce additional overhead because 
it does not require timer-based messaging during its execution.

The draft was put forth not as a complete solution for multicast routing in some network 
context, but as potentially part of some future solution.

2) AODV Multicast

Charlie Perkins from Sun presented an updated draft of AODV with a new extension to 
support multicast routing. The extension incorporates the notion of a multicast grouphead, 
a special node which is the first multicast group member in a connected network portion. 
The extension reuses the destination sequence number mechanism of unicast AODV to 
maintain loop freedom, and the grouphead is responsible for initializing and updating 
the multicast group destination sequence number. While not explicitly identified as such, 
the grouphead effectively functions as a traditional core (such as in CBT) in that it is a 
part of the shared, bi-directional multicast data forwarding tree, and is central to the 
construction and maintenance of the shared tree. Unlike CBT, should the grouphead 
fail or become partitioned from its previous network portion, AODV specifies a 
mechanism by which some other node in the network portion will dynamically elect itself 
as grouphead and form a new tree. AODV also specifies a mechanism by which two 
previously disconnected trees can merge into a single tree.

3) Manet Authentication Architecture

Stu Jacobs from GTE presented a draft on a proposed Manet Authentication Architecture.  
The presenter pointed out that it is known that wireless links are vulnerable to 
eavesdropping, replay, spoofing, and other attacks. In the absence of sufficient link-layer 
security support, some mechanism is required in many contexts to mutually authenticate 
routers before they begin exchanging network control traffic. The architecture presented 
specified a shared key mechanism based on a keyed-MD5 hash, as well as four levels of 
public key-based authentication requiring periodic verification via a certificate authority.  
The default mode also provided for no authentication when desired. While the architecture's authentication mechanisms where presented as logically independent of IMEP or any 
other routing support protocol, they are presently intended to be implemented in IMEP, 
requiring the addition of authentication and certificate objects to IMEP. This extended 
IMEP security functionality ensures that the authentication mechanisms may be used 
by any network control protocol function using the IMEP specification in the future.

The draft is very preliminary, as many issues such as details regarding the interaction with 
routing are left open.

4) Manet Addressing Discussion

Scott Corson began an open discussion to explore manet specific addressing issues.  A 
discussion took place regarding the use and adoption of an addressing framework to 
include a router identification capability vs. the use of other techniques and assumptions.  
Several group members raised varying issues and points of contention.  Points were 
raised regarding the ability to support address aggregation, multiple mobile interfaces, 
and managing these address spaces within manets.  There was consensus that this issue 
needed more discussion time and should subsequently be brought to the mailing list.  
Scott agreed to provide a strawman outline of addressing issues and present this to the 
mailing list for further exploration.

5) Open Discussion

Joe Macker opened the floor for open discussion of issues.

Some of the issues raised were as follows:

The need for developing applicability statements was reiterated, this helps reinforce 
what technical goals we are working towards.

The chairs agreed to provide a strawman applicability statement outline to the mailing 
list for review and comment.

The need for a common simulation environment was also raised again, which includes the 
development of common baseline scenarios and mobility models underlying the simulations.

Several group members agreed to provide input on mobility models for consideration

There was consensus that the recent ns2 extensions provided a significant vehicle forward 
on this issue

Carnegie Mellon indicated that various manet implementations within ns2 would be soon 
released for adaptation and further investigation by researchers.

Is it sufficient for multicast to rely on the proposed approach that works with any unicast 
protocol?

No group consensus or answer here at the present time

The question was raised regarding IRTF status and appropriateness for manet issues?

An answer was provided that if we can get to a standard(s) it should be an IETF group