6G-OpenLab - Open Scientific-Technologic Laboratory of Research in 6G of the UPC

Data Center

The data center is distributed across the Baix Llobregat Campus and Campus Nord, connected by a bidirectional DWDM system. This distributed DC architecture enables a low-latency, multi-user environment, crucial for providing experimental services for future 6G use cases and applications. The project includes Kubernetes, OpenStack, and HPC clusters running on high-capacity servers with high-end GPUs.

A fixed, private network interconnects all infrastructure elements, with its core in the data center.

The three technologies (Kubernetes, OpenStack and HPC clusters) deployed in the data center servers.

This cluster enables the deployment and management of microservices across various pods.

Use Cases

  • Model Deployment and Inference. Deploy pre-trained ML models and serve them as REST APIs or gRPC services in a containerized environment.
  • Distributed Training with Kubeflow. Use Kubeflow, an ML platform built for Kubernetes, to conduct distributed training of ML models using frameworks like TensorFlow, PyTorch, or MXNet. Kubeflow allows to run large datasets and scale across multiple containers.
  • Hyperparameter Tuning and Automated ML Pipelines. Kubernetes can manage parallel hyperparameter tuning tasks where multiple experiments are run simultaneously with different configurations. Optuna or KubeFlow Pipelines can be used to automate these processes.
  • Data Preprocessing Pipelines. Deploy data preprocessing pipelines (e.g., for cleaning, transforming, and normalizing large datasets) that run in parallel on multiple Kubernetes pods.
  • Interactive Jupyter Notebooks as a Service. Provide access to Jupyter Notebooks or JupyterHub hosted on Kubernetes, allowing you to prototype, develop, and test ML models interactively.

Two OpenStack clusters, one in CN and one in CBL, connected by the DWDM system.

Use Cases

  • Cloud computing. Network elements virtualization.

High-Performance Computing (HPC) servers are powerful computer systems designed to perform complex and resource-intensive tasks that require substantial computing power.

Use Cases

  • Large-Scale Distributed Deep Learning. Training large neural networks.
  • HPC simulations.Run complex simulations that require large amounts of computation and memory.