Radio Network Simulator

In the '70s while Ward was still in school he imagined computers costing a few hundred dollars controlling amateur radios similarly priced to provide a message delivery service nationwide.

> I modeled this in operation between 173 locations. It worked. And it exhibited unexpected behavior that interests me to this day. site

YOUTUBE hqnhNOlbc5s Published March 18, 2013.

Ward simulated a store-and-forward network composed of 173 vhf radios.

> I implemented a hierarchal routing and showed that it could route traffic backwards to use a discovered transcontinental link.

Page one of a Mimeograph report explaining the store-and-forward radio network.

The program was written in pascal using dynamically allocated structures for both radios and messages.

Ward later translated this to Java and found that the decomposition of structs served well as objects.

> The program exhibited behavior that I could not understand based on the time series and scatter plots from the first generation. I saw messages delivered with hop counts much higher than the "diameter" of the network.

He later discovered areas of congestion could create temporary routing dilemmas that sent packets in circles. These all cleared as the congested network drained once the load abated.

The model remains an object of study as a self-contained complex system. The exact configuration of stations and the logic they use to communicate have been duplicated now in three implementations offering progressively improved means of study.

As we undertake renewed explorations based on this now historic model we will here note aspects that run through all implementations and improvements we have or will make in the most recent incarnation.

Visualization

Ward stored lat/lon in the dataset and used it to draw a large map on the CalComp plotter. He took this to the Dayton Hamfest and showed the then young K1ZZ, ARRL technical editor, his vision for ham radio networking.

He collected stats out of some runs and render them as a map in Bill Mills' Picture Model abstraction. He drew bubbles at cities varying size for backlog and color (on a thermal scale) for throughput.

Ward used an early raster graphic monitor to render frames for an animation. Joe Pruitt helped with the c coding. We shot it in Super-8 movie film which got shown at an ARRL digital communications conference.

Load and route visualization that runs concurrently with the Java version of the simulator.

He wrote a interactive map for the Java version and added interactive trace-route capability. Ward debugged this with Test Point. He jittered the route endpoints which gave the lines a sense of width before Java had more than single pixel width lines.

Performance

Runs were reasonably expensive on the CDC 6500. Ward traced this to insertion overhead in the event queue.

He later discovered Leftist Trees in Knuth volume III. This was the structure Ward needed for his priority queue.

~

In the '70s while I was still in school I imagined computers costing a few hundred dollars controlling amateur radios similarly priced to provide a message delivery service nationwide. I modeled this in operation between 173 locations. It worked. And it exhibited unexpected behavior that interests me to this day. site