This is the second version of the P4 software switch (aka behavioral model), nicknamed bmv2. It is meant to replace the original version, p4c-behavioral, in the long run, although we do not have feature equivalence yet. Unlike p4c-behavioral, this new version is static (i.e. we do not need to auto-generate new code and recompile every time a modification is done to the P4 program) and written in C++11. For information on why we decided to write a new version of the behavioral model, please look at the FAQ below.
On Ubuntu 14.04, the following packages are required:
- automake
- libjudy-dev
- libgmp-dev
- libpcap-dev
- libboost-dev
- libboost-test-dev
- libboost-program-options-dev
- libboost-system-dev
- libboost-filesystem-dev
- libboost-thread-dev
- libevent-dev
- libtool
- flex
- bison
- pkg-config
- g++
- libssl-dev
You also need to install thrift and nanomsg from source. Feel free to use the install scripts under build/travis/.
To use the CLI, you will need to install the nnpy Python package. Feel free to use build/travis/install-nnpy.sh
To make your life easier, we provide the install_deps.sh script, which will install all the dependencies needed on Ubuntu 14.04.
Our Travis regression tests run on Ubuntu 12.04. Look at .travis.yml for more information on the Ubuntu 12.04 dependencies.
1. ./autogen.sh
2. ./configure
3. make
To enable logging, you probably want to use the following flags when running configure:
./configure 'CPPFLAGS=-DENABLE_SIMPLELOG'
In 'debug mode', you probably want to also use the following as well:
'CXXFLAGS=-O0 -g'
To run the unit tests, simply do:
make check
If you get a nanomsg error when running the tests (make check), try running them as sudo
To run your own P4 programs in bmv2, you first need to transform the P4 code into a json representation which can be consumed by the software switch. This representation will tell bmv2 which tables to initialize, how to configure the parser, ... It is produced by the p4c-bm tool. Please take a look at the README for this repo to find out how to install it. Once this is done, you can obtain the json file as follows:
p4c-bm --json <path to JSON file> <path to P4 file>
The json file can now be 'fed' to bmv2. Assuming you are using the simple_switch target:
sudo ./simple_switch -i <iface0> -i <iface1> <path to JSON file>
In this example <iface0> and <iface1> are the interfaces which are bound to the switch (as ports 0 and 1).
The CLI code can be found at tools/runtime_CLI.py. It also requires the json file generated by p4c-bm and can be used like this:
./runtime_CLI.py --json <path to JSON file> --thrift-port 9090
The CLI connect to the Thrift RPC server running in each switch process. 9090 is the default value but of course if you are running several devices on your machine, you will need to provide a different port for each. One CLI instance can only connect to one switch device.
The CLI is realized using the Python's cmd module and supports auto-completion. If you inspect the code, you will see that the list of supported commands. This list includes:
- table_set_default <table name> <action name> <action parameters>
- table_add <table name> <action name> <match fields> => <action parameters> [priority]
- table_delete <table name> <entry handle>
The CLI include commands to program the multicast engine. Because we provide 2 different engines (SimplePre and SimplePreLAG), you have to specify which one your target is using when starting the CLI, using the --pre option. Accepted values are: None, SimplePre (default value) and SimplePreLAG. The l2_switch target uses the SimplePre engine, while the simple_switch target uses the SimplePreLAG engine.
You can take a look at the commands.txt file for l2_switch and simple_router to see how the CLI can be used.
To enable event logging when starting your switch, use the --nanolog command line option. For example, to use the ipc address ipc:///tmp/bm-log.ipc:
sudo ./simple_switch -i <iface0> -i <iface1> --nanolog ipc:///tmp/bm-log.ipc <path to JSON file>
Use tools/nanomsg_client.py as follows when the switch is running:
sudo ./nanomsg_client.py --json <path to JSON file> --socket <ipc address>
The script will display events of significance (table hits / misses, parser transitions, ...) for each packet.
We will provide more information in a separate document. However you can test the Mininet integration right away using our simple_router target.
In a first terminal, type the following:
- cd mininet
- sudo python 1sw_demo.py --behavioral-exe ../targets/simple_router/simple_router --json ../targets/simple_router/simple_router.json
Then in a second terminal:
- cd targets/simple_router
- ./runtime_CLI --json simple_router.json < commands.txt
Now the switch is running and the tables have been populated. You can run pingall in Mininet or start a TCP flow with iperf between hosts h1 and h2.
- The new C++ code is not auto-generated for each P4 program. This means that it becomes very easy and very fast to change your P4 program and test it again. The whole P4 development process becomes more efficient. Every time you change your P4 program, you simply need to produce the json for it using p4c-bm and feed it to the bmv2 executable.
- Because the bmv2 code is not auto-generated, we hope it is easier to understand. We hope this will encourage the community to contribute even more to the P4 software switch.
- Using the auto-generated PD library (which of course still needs to be recompiled for each P4 program) is now optional. We provide an intuitive CLI which can be used to program the runtime behavior of each switch device.
- The new code is target independent. While the original p4c-behavioral assumed a fixed abstract switch model with 2 pipelines (ingress and egress), bmv2 makes no such assumption and can be used to represent many switch architectures. Three different -although similar- such architectures can be found in the targets/ directory. If you are a networking company interested in programming your device (parser, macth-action pipeline, deparser) with P4, you can use bmv2 to reproduce the behavior of your device.
You can take a look at the targets/ directory. We will soon publish a separate document with detailed information.
- Arithmetic is now possible on arbitrarily wide fields (no more limited to <= 32-bit fields).
- We finally have unit tests!
- While it is still incomplete, we provide a convenient 'event-logger' built on top of nanomsg. Every time a 'significant' event happens (e.g. table hit, parser transition,...) a message is broadcast on a nanomsg channel and any client can consume it.
No. Some features described in the P4 spec are missing compared to the original behavioral model:
- mirroring, cloning, resubmit
- direct meters
- registers, stateful memories (will be added soon)
- parse value sets
Some features which were absent in the original behavioral model are still missing:
- variable length fields
If you find more missing features or if you would like to request that a specific feature be added, please send us an email (p4-dev@p4.org) or submit an issue with the appropriate label on Github. Do not hesitate to contribute code yourself!
Please submit an issue with the appropriate label on Github.
You can fork the repo and submit a pull request in Github. For more information send us an email (p4-dev@p4.org).
All developers must sign the P4.org CLA and return it to (membership@p4.org) before making contributions. The CLA is available here.