Path Computation Element

Path Computation Element, also called PCE, is a component of Atlanticwave SDX project.

The problem PCE aims to solve is this: given a network topology and a set of connection requests between some nodes in the topology that must satisfy some requirements (regarding bandwidth, latency, number of hops, packet loss, etc.) how do we find the right path between the given nodes?

See background for a description of the approach.

Using PCE

PCE's API is still evolving. With that caveat, and omitting some details, the general usage is like this:

from sdx_pce.load_balancing.te_solver import TESolver
from sdx_pce.topology.temanager import TEManager

temanager = TEManager(initial_topology)
for topology in topologies:
    temanager.add_topology(topology)

graph = temanager.generate_graph_te()
traffic_matrix = temanager.generate_traffic_matrix(connection_request)

solution = TESolver(graph, traffic_matrix).solve()

breakdown = temanager.generate_connection_breakdown(solution)
for domain, link in breakdown.items():
    # publish(domain, link)

Note that PCE requires two inputs: network topology and connection requests. For testing, a random topology generator and a random connection request generator is available.

In the intermediate steps, network topology is generated by NetworkX and the traffic matrix computation is executed by Google OR-Tools Solver.

Network Topology

The Network Topology should be in the format of NetworkX graph. For each link, three attributes need to be assigned: cost, bandwidth and latency. The current unit of each attribute is abstract, but the unit in the Network Topology should be consistent with the unit in Connections.

Connection Requests

Connection requests should carry an identifier, source, destination, required bandwidth, and required latency, among other things. They are in JSON format. For an example, see test_request.json.

Working with PCE code

Working with PCE in a virtual environment is a good idea, with a workflow like this:

$ git clone https://github.com/atlanticwave-sdx/pce.git
$ cd pce
$ python3 -m venv venv --upgrade-deps
$ source venv/bin/activate
$ pip install .[test]

PCE can read topology data from Graphviz dot files, if the optional pygraphviz dependency is installed with:

$ pip install .[pygraphviz]

In order to be able to install pygraphviz, you will also need a C compiler and development libraries and headers of graphviz installed.

Running tests

Use pytest to run all tests:

$ pip install --editable .[test]
$ pytest

If you want to print console and logging messages when running a test, do:

$ pytest --log-cli-level=info [-s|--capture=no] \
    tests/test_te_manager.py::TEManagerTests::test_generate_solver_input

Use tox to run tests using several versions of Python in isolated virtual environments:

$ tox

With tox, you can run a single test verbosely like so:

$ tox -e py311 -- --log-cli-level=info [-s|--capture=no] \
    tests/test_te_manager.py::TEManagerTests::test_generate_solver_input

The test that depend on pygraphviz are skipped by default. If you are able to install pygraphviz in your setup, you can run that test too with:

$ tox -e extras

Test data is stored in tests/data as JSON files.

There are also some code checks (ruff, black, and isort) that you can run with:

$ tox -e lint