Parallel Implementation

This guide is for the Rust extension of the coursework, which is optional.

In this stage, you are required to write code to evolve Game of Life using multithreading on a single machine.

The following pages are some suggested steps to help you get started, you are not required to follow them.

Your implementation will be marked against the success criteria outlined here.

Step 2

Parallelise your Game of Life so that it uses worker threads to calculate the new state of the board.

You should implement a distributor that tasks different worker threads to operate on different parts of the image in parallel.

The number of worker threads you should create is specified in struct Params.

We recommend using threads or rayon (a data parallelism library), to parallelise the Game of Life computation.

We do NOT recommend tokio as it is designed for handling IO-bound tasks, it is not the best choice for the compute-bound tasks for this coursework.

If you use rayon, you can use rayon::ThreadPoolBuilder::new().num_threads().build() to manually generate a fix-sized thread pool, allowing you to specify how many threads rayon should use.

Tips on improving the performance - worker pool

Unlike the Median Filter in Lab 1, the Game of Life involves rapid, repeated iterations, which requires frequent thread creation and destruction.

However, frequent creation and destruction of threads can incur significant overhead. Hence, our core approach is to reuse threads, avoiding recreation in each iteration. We can abstract the design of the worker pool into four components: distributor, workers (thread), tasks (input) and results (output).

Workers are initialised ahead of computation and run continuously, processing tasks from the task channel and sending results back to the result channel. The distributor is responsible for assigning tasks to workers and collecting results for each iteration.

It is important to manage the lifecycle of the workers properly; when the distributor exits, ensure the workers are gracefully terminated.

It is recommended to back up your various implementations; this will be helpful for analysing how well you did in the later stages.

Test

To test your implementation, type the following in the terminal.

cargo test --release --test gol