Issues in wireless tactical networking & routing

Rohit Khare (rohit@bordeaux.ICS.uci.edu)
Fri, 06 Feb 1998 00:33:04 -0800


This was a really cogent message on issues in munchkin routing regarding
physical layout and wireless propagation issues.

Rohit

------- Forwarded Message

To: manet@itd.nrl.navy.mil
Subject: Concerns with proposed MANET routing protocols
From: Mike Butler <mgb@mitre.org>
Date: Thu, 05 Feb 1998 09:43:58 -0500

Hi all.

We've been reviewing the MANET routing proposals with the military's
tactical networks in mind. After considering the three proposed
algorithms, AODV, TORA, and ZRP, we have several concerns and a few
general questions.

Our concerns are:
1) There is an apparent reliance on reversible links
2) The importance of load balancing seems to be overlooked
3) There seems to be no exploitation of "advantaged" nodes
4) An early quench of route discovery may hide the best route
5) The routing objective function is unclear

The general questions are:
6) What does the charter's "a unicast algorithm" statement mean?
7) Has there been work in link-layer and network-layer coupling?

The following paragraphs will elaborate on each of these points.

- ----
1) There is an apparent reliance on reversible links.

The proposed routing algorithms have an apparent reliance on
bidirectional or reversible links. There are a number of factors
which will make wireless links unidirectional, among these are:

- Differing radio capabilities.

Radios within a network can have different transmit powers or
receive sensitivities. This is quite likely in a tactical
environment where man-pack and vehicular radios exist. Vehicular
radios, being less unconstrained by size and weight, typically
have 10dB greater transmit power than their man-pack counterparts.
Unidirectional links are exceedingly likely in tactical networks.

- Interference

Even a network of identical radios will have unidirectional links.
Interference, either by hostile jammers or by friendly "cosite"
will reduce a nearby receiver's sensitivity. Since it is only one
receiver that is handicapped, unidirectional links may result.

- Wide-area information broadcast

There is an increasing emphasis, especially in tactical networks,
on the wide-area broadcast of information. Satellite-based
transmitters (GBS) are being used for the forward links while the
the return links use alternative paths. (The commercial sector
display similar topologies in Direct-PC and some HFC
configurations.)

- EMCON operations

An extreme instance---applicable only in military networks---is when
some not cannot transmit due to an impending threat (EMCON). In
such a case, it may be necessary to have some other node provide a
"likely" route in response to a route discovery request.
(Obviously, an EMCON node can not participate in bidirectional
communications, but it still needs to receive information.)

- ----
2) The importance of load balancing seems to be overlooked

Another critically important issue in tactical networks will be
load balancing. Specifically, a node's purview will change with
it's position. A node in a particularly advantageous position
(e.g., mountain top) will "see" much more of the network. The
current routing proposals seem to favor routing traffic through
this advantaged node, thereby causing a bottleneck. The result is
that the introduction of an advantaged node can cripple the
network.

We expect that "advantaged nodes" will occur in most wireless
topologies; a person on stage is likely to have a much better view
of the wireless network throughout an auditorium. Advantaged
nodes will certainly be an issue in tactical networks. In fact,
there are efforts underway to introduce highly advantaged
(air-borne) communications nodes into tactical networks.

The approach to load balancing in a wireless network will likely
differ from that used in wired networks. In wired networks, load
balancing is only beneficial when multi-homed hosts (multiple
network segments) are present. In the wireless environment, load
balancing is a problem even if there is a single, flat network
segment and no multi-homed hosts.

- ----
3) There seems to be no exploitation of "advantaged" nodes

Although "advantaged" nodes are a liability for the currently
proposed routing algorithms, we believe that other routing
algorithms could use these nodes to great advantage. For example,
consider an airborne node with limited traffic handling capacity
but purview of the entire terrestrial network. Such a node could
easily collect neighbor information from the terrestrial assets,
construct a theater-wide routing picture, and broadcast a common
routing picture.

Furthermore, a node need not be omniscient to assist routing
decisions. For example, if some advantaged node sees a superset
of another sees, it has a decided advantage in finding and
propagating routes. Given radio's propagation phenomena, we
expect that advantaged nodes will occur in every wireless network,
even in flat, homogeneous networks. A node on stage, in the
center of a group, or near a doorway is likely to have a much
wider view; it makes sense to exploit that in finding routes.

- ----
4) An early quench of route discovery may hide the best route

We are concerned that the current routing algorithms seem to
quench the route discovery as soon as a path is found. The
concern is that the "best" route might easily escape discovery if
one packet along the "best" path is corrupted or lost. Since
wireless bit error rates are quite high (even for good links), it
is not unlikely that a packet along a good link could be dropped.

- ----
5) The routing objective function is unclear

Unlike wired links, there are a large number of objective
functions which make sense when selecting routes in a wireless
network. Although it is not explicitly stated in the
proposals, the tacit assumption seems to be that any
route is acceptable and the shortest path is generally preferable.

The "shortest path" route is frequently a poor choice in a
wireless network. A shortest path route through a wireless
networks would tend to favor longer hops which would (1) require
higher transmitter powers, (2) introduce higher bit-error rates,
(3) make hidden nodes more likely, etc. Furthermore, as mentioned
above, blindly accepting the "first found" route may prevent the
discovery of much better routes.

Objective functions which make more sense for a wireless
(particularly tactical) environment are (1) most reliable path
(best link margin), (2) most stable path (least likely to change),
(3) minimum total power path, (4) least congested path, (5)
minimal spectral use, etc.

We believe that any of these five objective functions might be a
superior choice to the shortest path or first-found routes in
tactical wireless environments. We also believe that route
discovery could provide this with little additional overhead---a
route's objective "cost" could be calculated as the route is
discovered and the high "cost" routes could be selectively pruned.
In short, we think that it is highly desirable to discover
multiple routes and down-select based on some criteria, and
first-found and shortest-path are unlikely to be provide a good
selection criteria.

- ----
6) What does the charter's "a unicast algorithm" statement mean?

According to the working group's charter, the intention is to
select "a" unicast routing algorithm. It seems highly unlikely
that a single algorithm will work well for all network sizes,
topologies, and rates of change. Is the working group making some
tacit assumptions about the networks of interest when they discuss
"an algorithm?" If so, would you please make those assumptions
explicit.

- ----
7) Has there been work in link-layer and network-layer coupling?

In "Mobile Ad hoc Networking (MANET): Routing Protocol Performance
Issues and Evaluation Considerations" the authors pointed out the
need for "interaction/interface with link-layer protocols". At
the link level there is a tremendously rich design space which
trades-off power, bandwidth, latency, coding, bit error rates,
etc. It would seem that these can be readily exploited in routing
through a wireless network, but none of the proposals addresses
these issues explicitly.

Has any work been done in this area? Does the working group
intend to explore this area at some future time? Is there a tacit
assumption that the proposed algorithms can be retrofitted with
these link-layer embellishments?

- ----

Mike Butler (mgb@mitre.org)
Tom Ferguson (ferguson@mitre.org)

The MITRE Corp
202 Burlington Road
Bedford, MA 01730

------- End of Forwarded Message