Thursday 23 January 2025
The ever-growing deluge of data from connected devices has created a pressing need for more efficient and adaptive management strategies in computing continuum architectures. The traditional cloud-centric approach is struggling to keep pace with the dynamic nature of these environments, where infrastructure demands and data analytics requirements can fluctuate rapidly.
Enter a team of researchers who have developed a novel framework for managing elasticity in the computing continuum. Their solution takes a multi-dimensional approach, encompassing not just resource management but also the specific needs of data analytics. By integrating reinforcement learning (RL) concepts, they’ve created an orchestrator that can dynamically adapt infrastructure resources and analytics configurations to meet changing demands.
The researchers’ system is designed to respond to three fundamental dimensions: quality, resources, and costs. These dimensions are interrelated, with changes in one impacting others. For instance, increasing the sample size for data analysis requires more computing resources, which in turn affects energy consumption and cost.
To evaluate their framework, the team conducted a series of experiments using a realistic use case from the smart cities domain. They simulated a distributed infrastructure with nodes equipped to process and store data analytics, and observed how CPU usage changed as data analytics requirements were adjusted. The results showed that increasing coverage, sample size, or freshness demands all led to significant rises in CPU utilization.
The team also demonstrated the effectiveness of their RL-based predictor by simulating an environment where four nodes processed data with different parameters. By adjusting these parameters, the orchestrator was able to balance system performance and resource availability. The results showed that the QL agent could fulfill specific SLOs (Service Level Objectives) close to 100% of the time when given priority.
This research has significant implications for the development of more efficient and adaptive computing continuum architectures. By integrating RL concepts, these systems can better respond to changing demands and ensure optimal performance in real-time. As the world becomes increasingly reliant on data analytics, this innovative framework is poised to play a critical role in shaping the future of distributed computing.
Cite this article: “Adaptive Computing Framework Empowers Real-Time Data Analytics”, The Science Archive, 2025.
Computing Continuum, Data Analytics, Reinforcement Learning, Resource Management, Adaptive Architecture, Cloud-Centric, Smart Cities, Distributed Infrastructure, Cpu Utilization, Service Level Objectives
