Industry 4.0 is the next big thing in the ever-turning wheel of industrial revolution. With increasing interconnectivity, automation, and new technologies like the internet of things (IoT), artificial intelligence (AI) and robotics, the automation and processing of large amounts of data are already a reality.
The IoT is becoming very popular in domains such as environmental monitoring and healthcare. Researchers are constantly trying to apply IoT along with Industry 4.0. But the challenge is that there is a large amount of data that needs to be handled which cannot be adequately carried out by IoT devices.
As AI and machine learning (ML) are gaining in popularity, it makes sense to implement AI/ML technologies along with IoT-based Industry 4.0.
In a typical Industry 4.0 IoT system architecture, the data is forwarded to other intermediate nodes, such as aggregation devices and gateway/sink nodes, before being forwarded to a Central Management Monitoring and Control entity.
A solution for handling the large amounts of data generated by IoT devices, is the use of In-network computing (INC). INC can reduce the volume of transmitted data and can combine the energy of IoT devices. Using this, it also becomes possible to generate AI/ML models based on the observed data, and apply the models on newly generated data in order to meet desired system objectives.
One way to forward packets of data from IoT devices, is through a network switch. However, the processing limitations of IoT devices do not make them suitable for such a computation. The network switches need to be flexible and programmable so that different models can be implemented on the switches, depending on the applications’ requirements.
This recent study has been conducted by Dr. Ganesh C. Sankaran, Prof. Krishna M. Sivalingam, and Dr. Harsh Gondaliya from the Department of Computer Science and Engineering, Indian Institute of Technology Madras, Chennai, India. Dr. Ganesh is also currently with the Networking and Cybersecurity Division, Information Sciences Institute, Marina Del Ray, California, USA. The researchers used programmable data planes (PDPs) as the hardware for network switches, along with programming protocol-independent packet processors (P4) programming language as the software. With this technology, data plane switches can now be used to process a packet and take some action, instead of just forwarding data.
In this work, it was shown that PDP and P4 concepts can be used to realize INC for AI-based Industrial IoT systems.
Application developers are unable to fully leverage the computing capabilities of packet switches due to the following reasons:
1. Lack of a similar level of understanding of a packet switch’s capabilities.
2. The operational boundaries that separate networks and servers.
Leveraging a network switch helps to unleash the full potential to improve IoT application and data processing performance.
Another issue for INC is the problem of security. Recent hardware security vulnerabilities like ‘Spectre’ and ‘Meltdown’ enable unauthorized programs to access the memory of other running programs. Hence secure design is a main concern, but has not received much attention. In this study, security is also focused upon. A secure execution model has been proposed.
Here, the in-network instructions are embedded in the packet and executed ahead of conventional network functions, such as routing. Also, the execution of instructions in a packet are not allowed to modify data outside of the current context.
Memory access restrictions were also imposed to resist security attacks that exploit unintended memory access.
Thus, the proposed architecture, which has a Mininet/P4 software environment on a Xilinx NetFPGA SUME based hardware platform, which is a programmable data plane, was implemented successfully with a secure execution model. Several application-level use cases were also demonstrated such as Internet checksum computation, Decision Tree interference model computation, Packet header parsing for security, and Domain name server lookup at network switches.
Dr. Praveen Tammana, Assistant Professor from the Department of Computer Science and Engineering, Indian Institute of Technology, Hyderabad, Telengana, India, has acknowledged the importance of the work done by the authors with the following comments: “In a typical Industrial 4.0 IoT system, AI/ML-based data classification techniques are commonly used. However, behind the scene, tens of thousands of IoT devices generate fine-grained data and forward the data to a central location. This architecture has huge computing, storage, and network resource overheads. As an alternative, data aggregation, computation, and classification at the intermediate nodes (e.g., aggregation servers, network devices, gateways) would not only reduce resource overheads significantly but also reduce AI/ML-based inference request/response latency. However, in this new paradigm, the security of intermediate nodes plays a major role to realize its full potential in practice. This work takes a step to address security concerns and proposed a secure execution model, which enables the secure execution of ML/AI model-specific instructions on the intermediate nodes. The work developed a prototype of the proposed model on two P4-based programmable switch targets that are considered for intermediate nodes and demonstrated the utility with real IoT data processing use cases. Overall, the work has great potential to enable Industry 4.0 ML/AI-based IoT applications to meet both performance and security requirements.”
Article by Akshay Anantharaman
Click here for the original link to the paper
