In the sample code section of this website, you may download code that I have written titled “throughput”. These pictures were produced by this code.



The throughput code attempts to measure the performance (or throughput) of a single queue with varying numbers of producer and consumer threads, in search of that “sweet spot” that maximizes throughput through the queue. To make this simulation realistic, producer and consumer threads are given both work and I/O to perform. The amount of work and I/O given to each thread varies according to a normal (Gaussian) distribution. For a description of the code, see:

The above pictures are from identical runs (that is, identical input parameters), with one run executing on my quad core intel i7, and the other executing on my raspberry pi. The color codes on the right of each picture indicate the number of work items that actually got through the queue. The X and Y axis are the number of consumer and producer threads.

One of the things that is very interesting (and a bit unexpected) to me is that the “sweet spots” are not nearly as far apart as I would have expected (with respect to the number of producer and consumer threads).

Comments, anyone?

Leave a Reply