Hi,
As we are talking about link characteristics, please let me share some of
my thoughts on the problem.
I am primarily concerned with link characteristics in a wireless ad-hoc
network setting. In such a network all nodes are mobile, there are no base
stations, and no notion of cells. Links get established and broken as nodes
move about. In such a scenario I feel there are three major issues to be
addressed:
1. the lossy nature of the links due to noise,
2. the unidirectionality of links,
3. temporary disruptions of paths between source and destination of packets.
Due to the lossy nature of the links several data or ack packets can get lost.
As a result no acks may be received by the sender. The sender may then
interpret the non-arrival of packets as congestion and invoke congestion
control. Several solutions have already been proposed for this problem in
the context of cellular networks (FEC as in Snoop, I-TCP, SACKs, etc.). I
am sure I am missing several other references.
I find the 2nd and 3rd issues especially interesting. For example, what if
links break temporarily due to which the source and destination are
partitioned? The moment that happens the route rediscovery protocols at the
network layer (being discussed by the MANET WG) should initiate a new route
search. For the duration that there is no path does it make sense for the
source to send packets, which will get dropped anyway? If the source can be
sent some "quench" information to "shut up" while route rediscovery is in
progress, and is sent a "resume" indication on the discovery of a route, the
problem may be ameliorated.
The issue of unidirectional links need more attention. When asymmetry is
mentioned it usually means asymmetry in bandwidth. But, in wireless networks,
due to the hidden terminal problem and a few other reasons, there may be
a significant number of unidirectional links. So, the path from A to B may
be quite different from the path from B to A. Moreover, there may be
times when there exists a path from A to B, but no path from B to A. If such
a situation arises in the middle of a TCP session, data packets may flow from
A to B, but there is no way that acks can flow from B to A. So, routing
asymmetry needs to be considered as well.
I guess I have taken enough of everybody's time. So, I will get off my
soap-box. If there is some interest, I would be glad to discuss these issues
in greater detail.
Thanks,
----Ravi.
===========================================================================
Ravi Prakash (ravip@utdallas.edu) www.utdallas.edu/~ravip
Computer Science Program
The University of Texas at Dallas
Richardson, TX 75083-0688. Phone: (972) 883-2289, Fax: (972) 883-2349
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 09:12:19 EST