DECICE Digital Twin Requirements. Figure 4 shows a block diagram representing the DECICE Digital Twin component. Starting top-down the digital twin has to interface with different parts of the DECICE framework and computing continuum. It receives sensor information from different real HW and SW platforms across the computing continuum and DECICE framework, ranging from HPC to edge. The services provided by the Digital Twin are split into two main parts - the Digital Twin Core and the Digital Twin modules. Digital Twin Core The digital twin core consists of (i) the system structure description that includes the available components and their compositions, as well as the (ii) current and historical status of the system. The digital twin core receives information from various sensors and APIs located in different parts of the physical/real twin or physical/real entity. It also provides the necessary information to other parts of the DECICE environment, such as the AI scheduler and Control Manager. Therefore, the digital twin core is primarily responsible for receiving the state of real entities as input and providing raw data and processed information to the other parts of the DECICE environment. Digital Twin Modules The digital twin modules are a set of modules that extend the capabilities of the digital twin. They consume the raw data provided by the digital twin core to provide an augmented representation of the real entity capable of simulating the behaviour in different condition, predicting future trends and estimating hidden states of the system (extracting features which are not directly observable). Figure 4: DECICE Digital Twin Component. The digital twin modules consist of two parts: the Behavioral Model and Forecasting/Evaluation. It creates the Virtual Training Environment that can forecast and evaluate different scenarios (more information in Section 7.3.3), which is useful for conducting what-if analysis. One of the main functions of this module is to use the current status of the job description and exploit it to forecast future outcomes. Additionally, the digital twin module includes visualization/GUI tools (e.g., Grafana) and feature extraction capabilities for NOT measurable parameters, which we can generate different approaches using human experts, ML tools, and Mathematical tools. This part may also contain various tools that utilize different techniques, such as statistical and machine learning methods, to generate new features or information, such as anomaly detection a...
