[next] [prev] [prev-tail] [tail] [up]
Developed for ARPANET.
- A distributed adaptive routing algorithm using estimated delays as cost criteria.
- Each node holds an estimate of the delay, and the next destination, for each possible
target.

- Periodically, each node gets the estimated delay vectors of its neighbors, and its own delay
information.

- Each node modifies its estimates and destinations.

- Shortcomings
- Queue length depends not just on speed of delivery between nodes, but also on other factors like
processing time within the nodes
- The algorithm responds slowly to congestion within the network
- Measures delay time between adjacent nodes, by looking at the time it takes to positive
acknowledgments to come back.
- Uses flooding to inform other nodes when significant changes in delay occur
- Delay time is a good criteria for low traffic loads, but not for high traffic loads where the algorithm
can easily cause oscillations.
- Observation: oscillations are implied by the conflicting objectives of providing the best routes for
all destinations

- Solution: provide good path to the average route instead of providing the best path to all
routes.

- Algorithm to calculate link cost:
- Measure average delay t on link
- Get service time
s = (average packet size) / (data rate on link) = (600 bits) / (data rate on link)
- Use queuing theory to calculate the utilization
= (2s - 2t)/(s - 2t)
- Let the cost be the average of the old and new utilization:
Unew = (Uold +
)/2
The averaging is to reduce the effect of oscillation.
[next] [prev] [prev-tail] [front] [up]