MLSysOps: Machine Learning for Autonomic System Operation in the Heterogeneous Edge-Cloud Continuum

Description

In response to the escalating deluge of data processed by computing systems, there has been a shift towards processing data as close to its origin as possible, known as edge computing. The emergence of cloud-edge computing exacerbates the already complex task of managing diverse and dispersed resources, this time on an immense scale, rendering human-in-the-loop management entirely impractical. To achieve a system and application management approach that is dynamic, flexible, and requires minimal user intervention, the concept of autonomic computing systems was introduced long ago, referring to systems capable of self-management based on high-level objectives set by administrators.

The Horizon Europe project MLSysOps will extend the autonomic paradigm by introducing a control framework powered by machine learning (ML) that interfaces with available management mechanisms. MLSysOps will also introduce hierarchical, distributed, explainable, and adaptable ML models for autonomous system operation of the cloud-edge-IoT continuum. To achieve adaptability, MLSysOps incorporates continual ML model learning with intelligent retraining concurrently with application execution. The project prioritizes openness and expandability, making use of explainable ML techniques and providing an application programming interface (API) for interchangeable ML models.

MLSysOps considers crucial aspects like energy efficiency (including the use of sustainable / green energy sources), performance optimization, minimizing latency, efficient and robust storage, devices with limited resources, and network connectivity. It employs ML models to co-optimize such objectives in a challenging computational environment.

MLSysOps will demonstrate its efficacy through two well-defined use cases in precision agriculture and smart cities.