Data Center Digital Twin for Sustainable Exascale Computing
Graph-Massivizer targets “sustainable science throughput” through scalable energy-aware, exascale operation and traceable TCO 107 understanding, including sustainability indicators and their environmental effects (e.g., GHG emissions). The Graph-Massivizer tools will enable the creation of a novel, graph-based digital twin of a data centre; this digital twin will further support the construction of sustainable exascale computing operational models to support scientific discovery in the next decade.
Objective
Design massive DC-MG models capturing the spatiotemporal dependencies between computation, nodes, and cooling equipment and conduct analytics to predict the impact of the spatial power distribution on cooling efficiency and cost.
Result
Green Data Centre Digital Twin and open data modelling of the Marconi 100 and EuroHPC Leonardo supercomputers at exascale
The Data Center Digital Twin for Sustainable Exascale Computing Use Case in detail
High Performance Computing (HPC) plays a crucial role in scientific progress, but as systems approach exascale, maintenance becomes increasingly difficult. This Use Case introduces a graph-based digital twin of HPC centers that addresses anomaly handling, energy efficiency enhancement, and carbon emissions reduction.
The main objective of this Use Case is to develop and implement a sustainable framework to tackle various challenges encountered in HPC systems, including handling anomalies, enhancing energy efficiency, reducing carbon emissions, and ultimately optimizing system performance.
The logic of the Use Case involves the integration of various components, such as Graph Inceptor to output the graph representation of the telemetry data of the HPC for anomaly prediction models, as well as the Graph Scrutinizer to execute the BGOs, as well as the Optimizer and Choreographer to provide the computational resources for performing inference/graph queries.
Expected Outcomes of the Use Case
The targeted outcomes are the implementation of a sustainable model for data centers based on graphs, the development of an ontology for HPC systems, and the creation of a data center data model. The outcomes will be achieved by creating a graph-based digital twin of HPC centers to optimize system performance and make data centers more scalable and sustainable.
This solution will enable the implementation of complex queries that make the work easier for facility managers and engineers that are not directly possible for current monitoring systems.
The solution plans to export open-source ontologies of the Marconi100 public dataset, as well as to provide technologies to be tested in data centers in production, and it will target data centers and HPC systems for exploitation. The stakeholders involved in this use case are data center owners, operational data analytics framework developers, and data center operators.
