Application of Machine Learning Techniques for Secure Traffic in NoC-based Manycores

TL;DR

This paper explores using machine learning-based intrusion detection systems to identify Denial of Service attacks in Network-on-Chip (NoC) systems of manycore processors, developing a platform for data collection and validation.

Contribution

It introduces a novel approach combining machine learning and temporal series analysis for detecting NoC DoS attacks, with a new platform for data extraction and validation.

Findings

Development of a high-level platform for traffic data collection

Validation of platform data with low-level platform results

Progress towards an IDS for NoC-based manycore systems

Abstract

Like most computer systems, a manycore can also be the target of security attacks. It is essential to ensure the security of the NoC since all information travels through its channels, and any interference in the traffic of messages can reflect on the entire chip, causing communication problems. Among the possible attacks on NoC, Denial of Service (DoS) attacks are the most cited in the literature. The state of the art shows a lack of work that can detect such attacks through learning techniques. On the other hand, these techniques are widely explored in computer network security via an Intrusion Detection System (IDS). In this context, the main goal of this document is to present the progress of a work that explores an IDS technique using machine learning and temporal series for detecting DoS attacks in NoC-based manycore systems. To fulfill this goal, it is necessary to extract traffic data from a manycore NoC and execute the learning techniques in the extracted data. However, while low-level platforms offer precision and slow execution, high-level platforms offer higher speed and data incompatible with reality. Therefore, a platform is being developed using the OVP tool, which has a higher level of abstraction. To solve the low precision problem, the developed platform will have its data validated with a low-level platform.