EE384Y Project Proposal:
------------------------


by Amit Gupta <amit.gupta@stanford.edu> and
   Amit Vyas <amitvyas@stanford.edu>


Wireless networks are inherently more lossy than traditional wired
networks.  As a result of this, a number of retransmissions are often
required before a packet successfully reaches it's destination.  In turn,
this leads to an increase in the expected delay per packet as compared to
wired networks. This has an adverse effect on certain interactive
applications, where the delay requirements are very critical.

We propose a scheme which uses extra bandwidth in order to reduce the
expected delay per packet. In order to achieve this goal, we exploit path
diversity by sending the same data across multiple paths in the network.
We expect that in each transmission, this will increase the probability
that the packet is delivered successfully to it's destination. In turn,
this will reduce the expected number of transmissions for successful
delivery and hence the delay per packet.

In order to create multiple paths, data will be sent on separate paths at
certain switches chosen by a branching algorithm. Design issues include:

- Choice of switches where data is sent on separate paths (branching
  algorithm)
- The number of paths on which data is sent (relates to bandwidth
  used)

We can model the wireless links using a simple model with a certain
probability of loss over a single hop. Then we can analyze the probability
of successful end-to-end delivery of a packet in one transmission, which
will also give us the expected number of transmissions per packet.
Alongwith the path metrics, this can be used to get an estimate of the
expected delay per packet.

So, we will first decide appropriate metrics and come up with a branching
algorithm. Then, for different network topologies, loss probabilities and
number of paths, we will perform mathematical analysis and simulations to
determine the expected delay per packet. Then comparing these results with
those obtained for conventional unicast, we will determine the tradeoff
between excess bandwidth used and improvement in delay.