Speaker
Dr
George Riley
(Georgia Institute of Technology)
Description
Large Scale Network Simulation Methods
Dr. Riley has been working in the field of large-scale network simulation
since his PhD thesis under the direction of Dr. Richard Fujimoto
and Dr. Mostafa Ammar. In that work he was the first to show that
distributed simulation techniques using conservative synchronization
protocols could be applied to the popular ns-2 network simulation tool
which resulted in the developemnt and release of "pdns".
However, just breaking a large topology into different logical
processes proved to be insufficient for achieving larger simulations.
He discovered that routing table models consumed large amounts of
systems memory, growing in proportion to N-squared (where N is the
total number of network nodes). This problem was alleviated using a
novel approach for packet routing called "NIx-Vector" routing, which
was applied to the ns-2 distributed simulation and resulted in successful
experiments with more than 100,000 nodes.
The lessons learned in the distributed ns-2 work were then applied to a
new network simulation environment developed by Dr. Riley called
the "Georgia Tech Network Simulator" (GTNetS), which was designed from
the outset for scalability and efficiency. Using a large number of CPUs
on the Pittsburgh Supercomputer Center, GTNetS was successfully demonstrated
to execute a network topology of more than 1 million network elements.
This was the first ever simulation experiment with packet-level detail
that exceeded the 1 million node threshold.
More recently, Dr. Riley is co-PI on the new ns-3 network simulator
development effort. Again, ns-3 has been designed to support distributed
simulation using conservative synchronization protocols. The inter-process
communications in ns-3 is performed using the ubiquitous MPI
message passing interface. Researchers at the Army Research Lab
in Aberdeen Maryland have used ns-3 to model networks of more than
100 million network elements, clearly two orders of magnitude larger
than has been possible previously.
Primary author
Dr
George Riley
(Georgia Institute of Technology)