EMbeDS HPC: HIGH-PERFORMANCE COMPUTING CLUSTER
EMbeDS is equipped with two PowerEdge R740 high-performance nodes, which will serve the scientific computing needs of a broad community of researchers and students who perform in silico experiments, large simulations and advanced statistical analyses of big data.
The two nodes have the following specifications:
- 2 * Intel Xeon Gold 6252
- 384 GB RAM
- 2 GPU Nvidia Volta V100
These multi-core and multi-gpu nodes are attached to a Storage Area Network with 40 terabytes of storage for high performance and/or "reliability", interconnected in star topology with a ToR switch. The cluster is managed using the VMware virtualization middleware layer to optimize its use and dynamically share computational resources.
The EMbeDS IT system integrates the NVIDIA GRID vGPU solution, which allows one NVIDIA Tesla GPU card to be shared across multiple virtual machines (VMs) creating multiple logical vGPU devices, each of which can be assigned to a virtual machine. Our target is a computational architecture with performance specificity + capacitive/scratch storage, which will offer both classical computational power and computational power supplied by GPUs mounted on servers (sharing GPUs core among VMs).
EMbeDS HPC program
The EMbeDS HPC program makes high-performance computing resources available to researchers and students conducting projects within the mission of the Department of Excellence. A virtual machine can be requested through a short proposal at any time by affiliated members. Proposals will be evaluated and accepted by the HPC Manager, or by the Managing Board, as needed.
Two types of VMs can be requested:
- VM vCPU: 8vCPU, 24 GB RAM
- VM vGPU: 8vCPU, 24GB RAM, vGPU V100 4GB RAM
Additional computational resources can be dynamically allocated to each virtual machine according to the needs of specific projects and research groups (again with the authorization of the HPC Manager or the Managing Board). Access to VMs can occur via SSH protocol or Remote Desktop using the internal network of the Scuola Superiore Sant'Anna.
Internal evaluation of HPC resources for the Institutes of Economics and Management
Two standard virtual machines (8vCPU, 24GB RAM, vGPU 4GB RAM) have been set up for the Institutes of Economics and Management in order to promote the accessibility, experimentation and use of the EMbeDS vGPU technologies. Two administratore have been designated for these VMs; namely, Dr. Andrea Vandin for the Institute of Economics, and Dr. Valentina Lorenzoni for the Institute of Management. The administrators can approve access requests and create credentials for users, who will be able to:
- test the EMbeDS computational resources, e.g. assess whether to submit their own proposal for a dedicated VM (see above)
- test parallel computational code on CPU/GPU, Machine Learning (ML) and Deep learning tools.
Interested affiliates of the Institutes of Economics and Management should contact their respective VM administrators.
More information is provided in the downloadable document "EMbeDS HPC Regulation" (In Italian).