(Optional) Do it yourself

Overview

Teaching: 0 min
Exercises: 0 min

Questions

• What is the best way to write parallel code from serial?

Objectives

• Work on a second example, possibly your own research code.

Going from Serial to Parallel

Choose a working serial code to work on. It can be your own research code, or the example below. Choose a function or a compact region of the code and write a parallel version of it.

Follow the same steps as in lesson 9:

1. Identify parallel and serial regions. Decide if the potential speedup is worth the effort.
2. Decide how to split the data among a number of ranks. Think about how much data needs to be communicated and how often.
3. Set up tests. Write a program that runs the function with different parameters and checks that it’s running correctly.
4. Write the parallel code. When using MPI, this means writing the function that will run on a single rank and communicate with the others. If you can, split this into multiple parts for different kinds on input.
5. Profile when the program runs correctly.
6. Optimise, making sure the tests still succeed. Goto: 5.

Serial code for the Ising Model

As an example, here is the code code code for simulating the Ising model.

The Ising model describes the behaviour of a ferromagnetic material. The material is formed of atoms arranged in a 2 dimensional square lattice. Each atom has a magnetic field pointing either up or down. The atoms only interact with their immediate neighbours. When neighbouring atoms point in different directions, they increase the energy of the system a bit. Similarly when they point to different directions, they decrease the energy by a bit.

At zero temperature the system would just fall to the state with lowest energy, but with a nonzero temperature it will sample states with different energies. Increasing energy is still alway less likely than decreasing it. Precisely, the probability of changing the direction of an atom is:

$$P = \min\left( e^{-\frac 1T \Delta_E}, 1 \right),$$

Where $\Delta_E$ is the change in energy. The code runs multiple iterations of a loop over all atoms in the lattice and attempts to flip the direction of each one at a time. After enough iterations, the system should find a natural equilibrium for the given temperature.

The code reads parameters from a file called parameter. It needs to contain two lines for two parameters, beta and iter. For example:

beta 0.5
iter 100

solution

A possible MPI implementation for reference: ising2d4_mpi.c ising2d4_mpi.F08 ising2d4_mpi.py

Key Points

• Start from a working serial code.

• Write a parallel implementation for each function or parallel region.

• Connect the parallel regions with a minimal amount of communication.

• Continuously compare with the working serial code.