Using AI/ML algorithms with ns-3.
If we want to implement AI algorithms in your custom network protocol/application under ns-3, then the direct way is to incorporate an existing AI/ML/DL framework with it. If it will be an AI/ML C++ based framework, then it can be incorporated with the C++ code of ns-3. But the existing AI/ML frameworks are huge and getting rapidly improving over time. Further, there are popular frameworks that are based on python and make it hard to merge with ns-3 natively.
So instead of directly incorporating such huge AI/ML frameworks with ns-3, ‘ns3-ai Model’ provides a way to communicate with independently running AI/ML frameworks from an ns-3 simulation.
Features of ns3-ai
ns3-ai does not include any AI algorithms in it. Further, it will not rely on one particular framework. This framework provides the ways to communicate and use the AI framework, which is installed separately in the system.
The following are the features outlined by its developers
- High-performance data interaction module (using shared memory).
- Provide a high-level interface for different AI algorithms.
- Easy to integrate with other AI frameworks.
Installation of ns-3 under chroot Jail
No specific version was mentioned by the developers of ns3-ai. In this article, I am doing the installation with my previous installation of ns3-33 which was installed under “chroot jail”. Specifically, I am going to use the ns3.33 version in which I installed ms-van3t extension; because in future, I may try to use some ns3-ai nodes in a VANET simulation.
Hopefully, this installation procedure will also work with higher versions of ns3 also.
So one may follow any procedure from  to install any version of ns-3 before proceeding the installation of ns3-ai module or simply follow any chroot jail-based installation procedure.
Installation of ns3-ai Module
Change to “contrib” folder of your ns-3 installation.
Clone the ns3-ai repository under the “contrib” folder of your ns-3 installation.
$ ./waf configure
The following screenshot shows the successful install of ns3-ai under chroot jail.
Configure Python interface for ns3-ai
$sudo apt install python3-pip
pip3 install . –user
Testing the Installation with simple a_plus_b code
a_plus_b is a very simple but useful example for the ns3-ai model to illustrate the data exchange between Python-based AI frameworks and ns-3.
We may use this example to test the ns3-ai installation.
In this example, the python code run.py will create a common memory pool and start an ns-3 simulation called a_plus_b which shares some variables from the common memory pool. In this case, the value of ‘a’ and ‘b’ are randomly chosen from the python part of the code and put into the common memory pool.
The ns-3 simulation will have access to these ‘a’ and ‘b’ values and calculate the value ‘c=a+b’ and put and display the result from the ns-3 simulation itself. In the following output, we are having 10 outputs. The reason is, the python code will initiate the ns-3 code 10 times within a loop.
As per the ns3-ai documentation, the following set of commands should run the a_plus_b iexample.
$ cd scratch/a_plus_b/
$ python3 run.py
But, under our ‘chroot jail’ based installation, this command doesn’t able to run the ns-3 part of the code. (just indefinitely waiting after displaying “build”)
We tried to run the python script in the background (&) and then manually started the ns-3 part of the simulation.
So, the following are the commands that we used under chroot jail.
$ cd scratch/a_plus_b/
$ python3 run.py&
$ ../../waf –run a_plus_b
This example simulation is not using any AI/ML features. So, this will run without having any AI/ML framework installed on the system. But, if we will try to run an example that uses AI/ML features, then before that, we have to install the necessary AI/ML framework and other python dependency modules.
Further, under chroot jail, the programs are not getting started as expected. The reason for these buggy behaviours should be found and addressed.