Performance Experiment: Graph Length

[Previous: Performance Experiment: Graphs vs Streams vs Synchronous Kernels]

In this experiment, we are going to test the hypothesis that the completion time of an iterative process will get shorter the more of its iterations are built into a graph. For example, if we are going to perform a task 64 times, we would expect that a graph which has 16 of those iterations built into it (that is repeated 4 times) would complete the task faster than one which only has a single iteration built in but is repeated 64 times.

To do this, we are going to use the same graph version of the quadratic equation solver seen in the explicit graph building tutorial, which we will then use with the child graph node feature in order to produce graphs with 1 built-in iteration, 8 built-in iterations, 64 built-in iterations, and 512 built-in iterations. Then each graph will be run so that the quadratic solver process is run a total of 16382 times by each graph which will be timed so we can observe any trends that appear.

Creating the Initial Graph

To use a graph as a child node of another graph, we must first create the base graph that the child node should contain.

In our case, this is the same graph that we built in the tutorial on explicit graphs.