Multithreading speedup for large models

In this tutorial we illustrate how to make use of multi threading in BeforeIT.jl to allow for faster executions of single simulation runs.

import BeforeIT as Bit
using Plots, StatsPlots

First, we initialise the model, this time we use the Italy 2010Q1 scenario

parameters = Bit.ITALY2010Q1.parameters
initial_conditions = Bit.ITALY2010Q1.initial_conditions
model = Bit.Model(parameters, initial_conditions);

The model is in scale 1:2000, so it has around 30,000 households

model.prop.H

29916

Note that the households number is actually the sum of active and inactive households, the owners of firms and of the bank

length(model.w_act) + length(model.w_inact) + length(model.firms) + 1

29916

Let's fist check how many threads we have available in this Julia session

Threads.nthreads()

2

Then we need to first compile the code not to count compilation time, we can do that just by executing the function one time

T = 50
Bit.run!(model, T; multi_threading = false);

Let's now compare the performance of single threading and multi threading

model = Bit.Model(parameters, initial_conditions);
@time Bit.run!(model, T; multi_threading = false);

model = Bit.Model(parameters, initial_conditions);
@time Bit.run!(model, T; multi_threading = true);

  5.098689 seconds (422.57 k allocations: 6.104 GiB, 14.62% gc time)
  2.537830 seconds (428.84 k allocations: 6.157 GiB, 9.12% gc time)

Is the speedup in line to what we would expect? Yes!