Enabling Federated Learning at the Edge through the IOTA Tangle

Abstract

The proliferation of Internet of Things (IoT) devices, generating massive amounts of heterogeneous distributed data, has pushed toward edge cloud computing as a promising paradigm to bring cloud capabilities closer to data sources. In many cases of practical interest, centralized Machine Learning (ML) approaches can hardly be employed due to high communication costs, low reliability, legal restrictions, and scalability issues. Therefore, Federated Learning (FL) is emerging as a promising distributed ML approach that enables models to be trained on remote devices using their local data. However, "traditional" FL solutions still present open technical challenges, such as single points of failure and lack of trustworthiness among participants. To address these open challenges, some researchers have started to propose leveraging blockchain technologies. However, the adoption of blockchain for FL at the edge is limited by several factors nowadays, such as long waiting times for transaction confirmation and high energy consumption. In this work, we conduct an original and comprehensive analysis of the key design challenges to address towards an efficient implementation of FL at the edge, and analyze how Distributed Ledger Technologies (DLTs) can be employed to overcome them. Then, we present a novel architecture that enables FL at the edge by leveraging the IOTA Tangle, a next-generation DLT whose data structure is a directed acyclic graph (DAG), and the InterPlanetary File System (IPFS) to store and share partial models. Experimental results demonstrate the feasibility and efficiency of our proposed solution in real-world deployment scenarios

Create the Containers

Launch the following commands in the respective directories:

docker build -t as-image .
docker build -t client-image .
docker build -t client-python-image .

Deployment

Deploy a private one-click-tangle instance and an IPFS cluster. This will result in the creation of Docker containers on the nodes where the deployment was executed. It is necessary to take note of the IPs and ports of the containers to reach them from outside. After doing so, it is possible to start the remaining components.

Start the Authorization Service (AS) on a node. From the respective directory:

docker run -i --name="as" -v $(pwd)/src:/mnt --network="host" as-image

Number of clients refers to the total count of clients.
N represents the i-th client.

Modify the code in each clientN (main.rs and lib.rs) to connect to the Tangle, IPFS, and AS. In main.rs:

• In lib::create_builder() [line 30], insert the IP and port of a Tangle node.
• In TcpStream::connect() [line 81], insert the IP and port of the Authorization Service.
• In IpfsClient::from_str() [line 191], insert the IP and port of a node in the IPFS cluster.

In lib.rs:

• In create_client_iota() [line 185 and line 209], insert the IP and port of a Tangle node.
• In create_client_identity() [line 211], insert the IP and port of a Tangle node.

Modify the code in the Authorization Service (main.rs and lib.rs) to connect to the Tangle. In main.rs:

• In lib::create_builder() [line 124], insert the IP and port of a Tangle node.

In lib.rs:

• In create_client() [line 143], insert the IP and port of a Tangle node.

Please note: Although each component can be deployed on different nodes, it is necessary for clientN and the respective client-pythonN to be on the same node.

After doing so, it is possible to launch the various containers of the components. From the respective directories:

docker run -i -v $(pwd)/src/clientN:/mnt --network="host" --name="clientN" -e PORT="555N" -e CLIENTS="Number of clients" client-image
docker run -i -v $(pwd)/src/clientN:/mnt --network="host" --name="client-pythonN" -e PORT="555N" -e CLIENTS="Number of clients" client-python-image

Citation

If you publish work that uses this code, please cite us as follows:

@article{MAZZOCCA202417,
    title = {Enabling Federated Learning at the Edge through the IOTA Tangle},
    journal = {Future Generation Computer Systems},
    volume = {152},
    pages = {17-29},
    year = {2024},
    issn = {0167-739X},
    doi = {https://doi.org/10.1016/j.future.2023.10.014},
    url = {https://www.sciencedirect.com/science/article/pii/S0167739X23003849},
    author = {Carlo Mazzocca and Nicolò Romandini and Rebecca Montanari and Paolo Bellavista}
}