OMNeT++ Goes Python
András Varga
andras@omnetpp.org
OMNeT++ Core Team
OMNeT++ Community Virtual Summit – November 2-3, 2022
Introduction
• After finishing OMNeT++ 6.0...
• We started playing around with Python because we saw huge
potentials there
– enthused by the Python-powered Analysis Tool
– during INET development and maintenance, a Python REPL with a home-grown lib has
grown to be an indispensable tool
– discovered the cppyy C++ FFI Python package
OMNeT++ Community Virtual Summit – November 2-3, 2022
(Snapshot from the office)
OMNeT++ Community Virtual Summit – November 2-3, 2022
Areas where Python can be useful
• Setting up models (e.g. topology building)
• Extending the capabilities of NED expressions
• Implementing simulation-specific ad-hoc components (e.g. scenario managers,
custom traffic generators, etc.)
• Simulation control (e.g. custom stop conditions)
• Managing and running simulations and simulation campaigns (workflow automation)
• Result analysis (inside and outside of the IDE)
• Disadvantages of Python in OMNeT++
– speed (or lack thereof)
• you want to keep Python code out of the “hot” parts of the model
– the word “module” has become ambiguous
• do you mean “importable Python file”, or “OMNeT++ simulation component”? ;-)
OMNeT++ Community Virtual Summit – November 2-3, 2022
Upcoming Python-related tools
• Python in NED expressions
– pyeval() and pycode() NED functions
• Modules implemented in Python
– @pythonClass NED property
– modules can be written from scratch, or subclassed
from existing (e.g. INET) modules
• Simulation library as a Python package
– from omnetpp.runtime import *
– based on cppyy
• Python package for processing simulation results
– from omnetpp.scave import results, ...
• Python package for managing and running simulations
– from omnetpp.simulation import *
OMNeT++ Community Virtual Summit – November 2-3, 2022
The pyeval(), pycode() NED functions
• pyeval(<expr>,...) - evaluates a Python expression string
– pyeval("2*3")
– pyeval("x: 2*x", 3) --> 6
• pycode(<block>,...) - evaluates a Python statement block
– block must end in a "return" statement (like a function body)
– pycode("import math; return math.factorial(15)")
– pycode("a,b: import math\nif a<0 or b<0: return math.nan\nreturn math.gcd(a,b)",
70, 62)
More details about pyeval()/pycode(), @pythonClass and cppyy in
Attila Török’s Summit presentation: Using Python within Simulations
OMNeT++ Community Virtual Summit – November 2-3, 2022
The @pythonClass NED property
• Denotes that module is implemented by a Python class
// sink.ned
simple Sink {
parameters:
@pythonClass;
...
}
# sink.py
from omnetpp.runtime import *
class Sink(omnetpp.cSimpleModule):
def handleMessage(self, msg):
...
OMNeT++ Community Virtual Summit – November 2-3, 2022
The cppyy package
• Dynamic runtime Python bindings for C++
– allows cross-inheritance and callbacks, template instantiation and more
– simple example:
import cppyy
cppyy.include("iostream")
cppyy.cppdef("""class A { public: void sayHello() { std::cout << "Hello" << std::endl; } };""")
A = cppyy.gbl.A
a = A()
a.sayHello() # prints “Hello”
• Foundations:
– Cling, the interactive C++ interpreter from CERN
– Cling itself builds on Clang and LLVM
– (at least this is what they want you to believe, but actually it relies on magic)
OMNeT++ Community Virtual Summit – November 2-3, 2022
The omnetpp.runtime package
• Based on cppyy, it exposes the simulation library as a Python package
– both simulation kernel and the “envir” infrastructure
– its essence:
import cppyy
cppyy.include("omnetpp.h")
• OMNeT++ is undergoing extensive refactoring:
– Reusability of “envir” part improved
– Python readiness
– Multi-thread support
• Allow multiple threads to be used for simulation (but simulations are still single-threaded!)
./aloha -u Cmdenv --cmdenv-num-threads=8 ...
• Global variables became thread_local or simulation-scope (cSimulation::getSharedVariable<T>(name) - new)
OMNeT++ Community Virtual Summit – November 2-3, 2022
The omnetpp.scave package
• Part of OMNeT++ 6.0
Slide from last year’s Summit
OMNeT++ Community Virtual Summit – November 2-3, 2022
The omnetpp.simulation package
• Library and toolset for managing and running simulations and campaigns in
various ways
– in-process / local / distributed (e.g. ssh cluster using Dask)
– for results, for regression testing (e.g. fingerprints), etc.
• Grown from the needs of INET development and maintenance
– “run all simulations, utilizing all computing resources I have access to”
– “refresh list of simulations to be fingerprint-tested”
– “re-run failing fingerprint tests”
– “re-run simulations that contain WiFi” (after WiFi model change)
– “compare results to those with INET version X”
• Human time is expensive:
– automate/assist as much as possible (high-level tools, REPL, etc)
– store instead of recompute (fingerprints, simulation results, etc)
– utilize available computing resources (multi-core, ssh cluster, etc)
OMNeT++ Community Virtual Summit – November 2-3, 2022
DEMO
OMNeT++ Community Virtual Summit – November 2-3, 2022
Recap of the Demo
Python added to several OMNeT++ sample simulations:
• pyfifo: pure Python simple modules
• routing: various examples for making use of Python
• aloha: examples for running simulations from Python
To try, use the OMNeT++ version at the following URL:
https://drive.google.com/file/d/1jDD-vtzi9YVzShrw2fP6q1SUU5VMVsDM/view?usp=share_link
OMNeT++ Community Virtual Summit – November 2-3, 2022
samples/pyfifo
• A fairly verbatim re-implementation of the fifo example
• With all modules written in Python
OMNeT++ Community Virtual Summit – November 2-3, 2022
samples/routing
Configurations in omnetpp.ini:
• [FromCsv]: Setting up a network specified in CSV, using Python
• [RandomTree]: Setting up a network generated with NetworkX Python package
• [DumbbellFaultyLink]: Shows a scenario manager written in Python
• [AppExt]: Extending a C++ simple module from Python
• [App2]: Parameter values produced using pyeval()/pycode()
OMNeT++ Community Virtual Summit – November 2-3, 2022
samples/aloha
Python scripts:
• example1.py: instantiating a simulation
simulation = omnetpp.cSimulation(...)
• example2.py: all simulations in an omnetpp.ini parameter study
ini.getNumRunsInConfig("PureAlohaExperiment")
• example3.py: manually organizing a parameter study
for numHosts in [10,15,20]:
for iaMean in [1,2,3,4,5,7,9]:
...
• example4.py: utilizing multiple CPU cores
with multiprocessing.Pool() as p:
p.map(alohaJob, taskList)
• example5.py: simulation-based optimization
scipy.optimize.minimize(...)
• example5a.py: preserve and also plot all simulations
OMNeT++ Community Virtual Summit – November 2-3, 2022
Questions?
17
OMNeT++ Community Virtual Summit – November 2-3, 2022
