It's just my playground with Amdahl's law. I know that this law realization as code thoroughly has been written many and many times, but I decided to do it myself as research and learning. I've written several useful methods for computing desired values
speedUpRate- Calculates how fast will your system with a specific number of processes, and with some proportion of parallel computingdrawPlot- Draws graph with plots. Pass into the method: 1) number of processes/threads 2) Sequences with values of parallel percentage (To compare with each other)
calculateSerialProportion- It's something new... I was wondering I have no clue where and how I should know how many serial/parallel works have some computing.
However, I could gather a processing time, and change the number of threads(as an example).
With this knowledge now we can determine how much serial processing proportion we have, and aim to reduce it. The method returns the most effective number of processes. This method takes Tuple(processing time, number of processes). More values, better and precise output.
| Processing time in ms | 14199 | 16922 | 4568 | 3527 | 2904 | 2409 | 2102 | 1886 | 1892 | 1871 |
|---|---|---|---|---|---|---|---|---|---|---|
| Number of processes | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
Serial proportion: 13.28%.
The most optimal number of process: 6
There's my attempt at modeling parallel work: src/main/scala/SomeApp.scala