| Sumario: | Key Performance Indicators (KPIs) are essential to modern enterprises to keep track of the well-being, performance, and stability of sophisticated digital infrastructure. With recent developments in infrastructures moving towards highly distributed, cloud-native, and microservices-based models, it has become difficult to detect what the causes of the deviations of KPI are. Conventional methods of root-cause analysis, commonly based on fixed rules or centrally trained machine learning, are unable to scale, adapt, and give timely explanations in these types of settings. In the given paper, we suggest a multi-Agent systems (MAS) to analyze the root cause of KPIs and identify anomalies, where several autonomous agents are working together to reach a diagnosis based on causal relationships, and validate the diagnosis hypotheses. The agents have different analyses, i.e., KPI monitoring, dependency analysis, causal inference, or hypothesis evaluation, and the wellorganized communication allows them to think, in an effective and adaptive way, system-wide. The proposed aids in decentralized decision making, gives more localization while fault and boosts resistance to noisy or incomplete data. By experimentally assessing the framework on representative cases of KPI, it is shown to have a higher diagnostic accuracy, lower latency in analysis, and is better scalable than centralized methods. The results indicate that the multi-agent systems (MAS) have the potential to become the base technology of the next-generation AIOps systems, and simultaneously, the issue of governance, transparency, and ethical concerns that must be addressed to make the system work in practice are also discussed.
|
|---|