PDES-MAS

PDES-MAS is a framework and a distributed simulation engine for multi Agent-based system models (MAS).

Getting Started

How to compile

Use Git to clone it to your local repository:

git clone https://github.com/PDES-MAS/PDES-MAS.git
cd ./PDES-MAS

PDES-MAS uses CMake to manage its build process. Typically, you should use the Release target to achieve the best performance, and use Debug only when you're debugging it. To build PDES-MAS, first, ensure that you have CMake installed. You can either use a package manager of download it manually from CMake official site.

Then, use CMake to generate the Makefile, build it in the ./bin directory:

mkdir bin
cd bin
cmake -DCMAKE_BUILD_TYPE=Release ..

After the Makefile has been generated, use make to build it:

make -j5

Running Example Code

There are already some standard examples of multi-agent system model, Tileworld is one of them. And we've provided an example of this kind of agent. You can find it in your build directory after successfully built PDES-MAS. To run and test it, just use mpirun:

mpirun -np 15 tileworld

Agent API Reference

The Agent Programming Interface

To program multi-agent models used to run on PDES-MAS, we provided a set of programming interface. The interface includes two key components: the Simulation class, which is used to control the settings of the whole simulation, and the Agent class which let you program your own agent models without knowing the detailed implementations inside PDES-MAS.

First, let's see a basic example about how to construct a simulation:

int main(int argc, char **argv) {
  spdlog::set_level(spdlog::level::debug);
  Simulation sim = Simulation();
  sim.Construct(7, 8, 0, 10000); // 7 CLPs, 8 ALPs, start from timestamp 0, end at 10000

  spdlog::info("MPI process up, rank {0}, size {1}", sim.rank(), sim.size());
  sim
      .attach_alp_to_clp(7, 3)
      .attach_alp_to_clp(8, 3)
      .attach_alp_to_clp(9, 4)
      .attach_alp_to_clp(10, 4)
      .attach_alp_to_clp(11, 5)
      .attach_alp_to_clp(12, 5)
      .attach_alp_to_clp(13, 6)
      .attach_alp_to_clp(14, 6)
      .preload_variable(10701, Point(0, 0), 0)
      .preload_variable(10702, Point(1, 0), 0)
      .preload_variable(10801, Point(2, 0), 0)
      .preload_variable(10802, Point(3, 0), 0)
      .preload_variable(10901, Point(4, 0), 0)
      .preload_variable(10902, Point(5, 0), 0)
      .preload_variable(11001, Point(6, 0), 0)
      .preload_variable(11002, Point(7, 0), 0)
      .preload_variable(11101, Point(0, 1), 0)
      .preload_variable(11102, Point(0, 2), 0)
      .preload_variable(11201, Point(0, 3), 0)
      .preload_variable(11202, Point(0, 4), 0)
      .preload_variable(11301, Point(0, 5), 0)
      .preload_variable(11302, Point(0, 6), 0)
      .preload_variable(11401, Point(0, 7), 0)
      .preload_variable(11402, Point(0, 8), 0)
      .Initialise();

  spdlog::info("Initialized, rank {0}, is {1}", sim.rank(), sim.type());
  if (sim.type()=="ALP") {
    for (int i = 0; i < 4; ++i) {
      TestAgent *test = new TestAgent(0, 10000, sim.rank() * 100 + 1 + i);
      sim.add_agent(test);
    }

  }

  sim.Run();
  spdlog::info("LP exit, rank {0}", sim.rank());
  sim.Finalise();
}

First, we initialize the simulation object sim, which will be in control of the simulation settings, variable loading and controlling of the simulation execution process. After calling Construct(), MPI processes will be initialized and all the following code will run in different MPI processes.

void TestAgent::Cycle() {
  if (this->GetEndTime() - this->GetLVT() <= 1000) {
    this->time_wrap(this->GetEndTime() - this->GetLVT());
  } else {
    this->time_wrap((random() % 500) + 500);
  }
  this->SendGVTMessage();
}