Abstract: “Computation” has become a massive part of our daily lives: in science, a lot of experiments and analysis rely on massive computation, in AI we use vast resources to train and use massive models, and in engineering we use complex simulations and digital twins to increase efficiency and productivity. Under the assumption that computation is cheap, and time-to-result is the only relevant metric, we often use significant computational resources at low efficiency. In this talk, I argue this approach is an unacceptable waste of computing resources, and demonstrate we can do better! By means of a couple of case-studies, I will show how performance engineering can be used for zero-waste computing, proving how efficiency and time-to-result can be happily married. I will further propose a co-design methodology that leverages such performance engineering methods to enable the selection of algorithms _and_ their effective deployment on suitable infrastructure. The approach relies on design-space exploration, driven by efficient search methods and compositional performance models. I will conclude by reflecting on the next steps and open questions that need answers to make this co-design approach feasible and applicable for more applications and systems.

Abstract: Our society has turned digital: From science to business, from online shopping to online gaming, from education to government, digital applications depend every moment on Information and Communication Technology (ICT) infrastructure. To pilot this infrastructure and navigate through it, we need detailed information and decision-making support, provided on-time, cheaply, and reliably. (We briefly argue non-data-driven approaches are currently unable to cope with real conditions.) In this talk, we argue that Operational Data Analytics (ODA) can provide these capabilities, with advances not only in monitoring and observability, but also in data sourcing and ontology mapping, data cleaning and filtering, data characterization and modeling, and process management. Furthermore, we argue a digital twin, capable of simulating both the current conditions and making long-term predictions, should be integrated with the ODA system - providing an advanced form of A - to help with resource management and scheduling, pipeline optimization, energy awareness, general system tuning, capacity planning, etc. We present a reference architecture for ODA, a partial analysis of the state of the art, and experience with data collection and its use in the digital twinning of datacenters. This work aligns with and benefits from collaboration with the SPEC RG Cloud Group and, among others, the EU Graph-Massivizer project, and the OffSense and 6G Future Network Services projects in the Netherlands.