P4 Model Repository

This repository maintains a sample set of the P4 programs and allows building P4 for the P4 Behavioral Model.

Important: git submodules

We often update the submodules for this repo. This is why we recommend that you run the following command whenever pulling the latest version of master:

git submodule update --init --recursive

Quickstart

To install all the Ubuntu 14.04 dependencies, run

./install_deps.sh

Before running the simulator, you need to create veth interfaces that the simulator can connect to. To create them, you need to run:

sudo p4factory/tools/veth_setup.sh

We use autoconf tools to generate makefiles. Run the following commands to prepare your workspace.

cd p4factory
./autogen.sh
./configure

To validate your installation and test the simulator on a simple P4 target, do the following:

cd p4factory/targets/basic_routing/
make bm
sudo ./behavioral-model

To run a simple test, run this in a different terminal:

cd p4factory/targets/basic_routing/
sudo python run_tests.py --test-dir tests/ptf-tests/

Building and Running a Target

Each P4 program (called a 'target') is set up in a directory under targets/. Inside the target directory is a Makefile with the instructions on how to build the behavioral model for that P4 program.

To build the target "project_name":

cd targets/project_name
make bm

This should result in an executable in the same directory called "behavioral_model"

To add Openflow support to a target, please refer here.

Integration with Mininet

Integration with Mininet provides a way to instantiate a network of nodes each running a data plane described in P4.

We provide a Mininet integration for one of our existing targets: simple_router

To run it, do the following:

cd p4factory/targets/simple_router/
make bm
./run_demo.bash

To install some table entries, run the following (in a different terminal):

./run_add_demo_entries.bash

You can then type commands in the Mininet CLI:

mininet> h1 ping h2

Integration with Mininet and Docker

Integration with Mininet and Docker provides a way to instantiate a network of nodes with each node running a data plane described in P4 along with its own control plane instance (eg. Quagga) packaged into a Linux container (Docker).

Mininet: Install Mininet from http://mininet.org/download/.

Docker: Install Docker from http://docs.docker.com/linux/started/.

To build the docker image for a target, include the file "makefiles/docker.mk" in the Makefile, set the variable DOCKER_IMAGE to the Makefile target to build and build the target "docker-image".

For example:

# In target/switch/Makefile, add the following lines
DOCKER_IMAGE := bm-switchlink
include ${MAKEFILES_DIR}/docker.mk

# To build the docker image
make docker-image

The docker image is called "p4dockerswitch".

Sample output:

sudo docker images
REPOSITORY          TAG                 IMAGE ID            CREATED             VIRTUAL SIZE
p4dockerswitch      latest              84f6c028ad6c        14 hours ago        1.234 GB
ubuntu              14.04               6d4946999d4f        3 weeks ago         188.3 MB

We provide a few topologies that showcase Mininet and Docker integration.

SAI:

mininet/sai_l2.py : Simple L2 topology with two switches and two hosts.

mininet/sai_l3.py : Simple L3 topology with two switches and two hosts.

Switchlink with SAI:

mininet/swl_l2.py : Simple L2 topology with two switches and two hosts.
                    The topology is loop free (no spanning tree protocol).

mininet/swl_stp.py : L2 topology with four switches and two hosts. It runs
                     MSTPD to form a loop free topology.

mininet/swl_l3_static.py : Simple L3 topology with two switches and two
                           hosts. The setup is statically configured.

mininet/swl_ospf.py : Simple L3 topology with two switches and two hosts.
                      The setup runs OSPF (Quagga) to learn and advertise
                      networks.

mininet/swl_bgp.py : Simple L3 topology with two switches and two hosts.
                     The setup runs EBGP (Quagga) to learn and advertise
                     networks.

Please see README.md under target/switch for a specific example on how to build the docker image and run the test topologies.

Creating a New Target

To add a new target, cd to targets/ and run:

p4factory/tools/newtarget.py project_name

where project_name is the name of the P4 program (without the .p4 extension). This will create a new directory in targets/ called project_name/, set it up to build the behavioral model, and create a template for the P4 program there named project_name.p4. Then, edit that file or copy your P4 program to that file and make in that directory.

P4 Dependency Graph Generator

The relationships between tables of more complex P4 program can be difficult to comprehend. The p4c-graphs utility parses through the the P4 program and generates a dependency graph using graphviz. The dependency graph can be generated with the following command:

p4c-graphs <p4 code>

The resulting files can be viewed using xdot or with a PNG viewer.

Towards a better behavioral model: bmv2

We have released a new version of the behavioral model, written in C++. Some targets already support this new model -in addition to the original version, p4c-behavioral. If you see a target with a bmv2 directory, it means the new model is supported and you can try it out!

The new model splits the switch logic and the auto-generated PD API (drivers) into 2 different processes.

For example, the l2_switch target supports bmv2. To run the code, you can do the following:

  cd targets/l2_switch/bmv2/
  make bm
  ./run_bm.sh       # to start the data plane 
  sudo ./drivers    # in a second terminal, to start the PD APIs (RPC server)

You can then run the tests in a third terminal, by going up one directory:

sudo python run_tests.py --test-dir tests/ptf-tests/

The switch.p4 target already supports bmv2. For more information take a look at the bmv2 README.

The new behavioral model code is also hosted on p4lang, in this repository.