pysimgrid

The main goal of the project is to create an easy way to implement and benchmark scheduling algorithms while leveraging the powerful SimGrid simulation framework.

Features:

• Simulate different scheduling distributed platforms
• Implement your own scheduling algorithms
• Compare them with well-known approaches (e.g. HEFT). Batteries included )
• Run your simulations in parallel

Documentation:

https://alexmnazarenko.github.io/pysimgrid/index.html

Acknowledgements:

This work is supported by the Russian Foundation for Basic Research (projects 15-29-07043, 15-29-07068).

Examples

Running a simulation:

from pysimgrid import simdag
import pysimgrid.simdag.algorithms as algorithms

with simdag.Simulation("test/data/pl_4hosts.xml", "test/data/basic_graph.dot") as simulation:
  scheduler = algorithms.Lookahead(simulation)
  scheduler.run()
  print(simulation.clock, scheduler.scheduler_time, scheduler.total_time)

Implementing your very own scheduling algorithm:

from pysimgrid import simdag
import networkx

class RandomSchedule(simdag.StaticScheduler):
  def get_schedule(self, simulation):
    schedule = {host: [] for host in simulation.hosts}
    graph = simulation.get_task_graph()
    for task in networkx.topological_sort(graph):
      schedule[random.choice(simulation.hosts)].append(task)
    return schedule

Dependencies

C++

• C++11 capable compiler (tested with GCC/G++ 4.8.4)
• SimGrid built with graphviz support (build script provided, tested v3.13)
• CMake (for building SimGrid)

Python

• python 2.7 or 3.4+
• setuptools
• Cython
• numpy
• networkx

Build instructions

Ubuntu 14.04+

Install system dependencies (list is not full):

sudo apt-get install libboost-context-dev libboost-program-options-dev libboost-filesystem-dev doxygen graphviz-dev

Use provided scripts to get dependencies:

./get_simgrid.sh

Installation:

python3 setup.py install --user

Development build

Inplace build

python3 setup.py build_ext --inplace

Test the build:

python3 run_tests.py

FAQ

1. Where to get platform definition files? They look scary.

SimGrid source distribution contains quite a few of platform examples of different complexity and scale.

2. Where to get some ready DAGs?

• Pegasus workflows, DAX format
  • See Pegasus Workfow Generator
  • Direct download (warning: ~400Mb) [link]

3. What about multi-core tasks?

They are not supported for now. Two main reasons:

• Existing SimGrid task parsers do not support them, so you need custom task format to even setup this.
• There are some discouraging words in SimGrid documentation about validity of such simulation.
• API seems to be "in progress"