Software systems usually provide numerous configuration options that can affect performance metrics such as execution time, memory usage, binary size, or bitrate. On the one hand, making informed decisions is challenging and requires domain expertise in options and their combinations. On the other hand, machine learning techniques can search vast configuration spaces, but with a high computational cost, since concrete executions of numerous configurations are required. In this exploratory study, we investigate whether large language models (LLMs) can assist in performance-oriented software configuration through prompts. We evaluate several LLMs on tasks including identifying relevant options, ranking configurations, and recommending performant configurations across various configurable systems, such as compilers, video encoders, and SAT solvers. Our preliminary results reveal both positive abilities and notable limitations: depending on the task and systems, LLMs can well align with expert knowledge, whereas hallucinations or superficial reasoning can emerge in other cases. These findings represent a first step toward systematic evaluations and the design of LLM-based solutions to assist with software configuration.

Models at runtime have been initially investigated for adaptive systems. Models are used as a reflective layer of the current state of the system to support the implementation of a feedback loop. More recently, models at runtime have also been identified as key for supporting the development of full-fledged digital twins. However, this use of models at runtime raises new challenges, such as the ability to seamlessly interact with the past, present and future states of the system. In this paper, we propose a framework called DataTime to implement models at runtime which capture the state of the system according to the dimensions of both time and space, here modeled as a directed graph where both nodes and edges bear local states (ie. values of properties of interest). DataTime provides a unifying interface to query the past, present and future (predicted) states of the system. This unifying interface provides i) an optimized structure of the time series that capture the past states of the system, possibly evolving over time, ii) the ability to get the last available value provided by the system's sensors, and iii) a continuous micro-learning over graph edges of a predictive model to make it possible to query future states, either locally or more globally, thanks to a composition law. The framework has been developed and evaluated in the context of the Intelligent Public Transportation Systems of the city of Rennes (France). This experimentation has demonstrated how DataTime can deprecate the use of heterogeneous tools for managing data from the past, the present and the future, and facilitate the development of digital twins.