The EHBA infrastructure consists of 19 machines, supported by a centralized storage service.
These machines are organized as Kubernetes nodes and run a customized version of Kubernetes called Kaptain.
Five machines function as master nodes, while the remaining machines serve as worker nodes.
Among the worker nodes, five are equipped with GPUs: four have previous-generation GPUs (V100S 32GB), and one has three NVIDIA A30 GPUs.
ICE4AI is a containerized service that hosts JupyterHub, which runs inside a containerized notebook scheduled within the Kubernetes cluster.
Users access the system via ice4ai.ehbda.lan, the container hosting JupyterHub.
They can only use local resources on ehbda.dieti.unina.it; access to AWS resources is no longer available.
Users can request notebooks, but they cannot request containers. Notebooks are assigned automatically through a resource allocation system.
Please note that GPUs, like all local resources, cannot be reserved. They may be used as needed for processing tasks, subject to availability. In particular, the allocation system allocates the nodes with the GPUs, if available, to the notebooks when users request a notebook with a GPU. Then, if the notebook is not working for a few hours, the system de-allocates it, leaving all the files of the user on the storage space of such user.
When requesting an account, users must describe the goal of their research.
If they are bachelor’s students, master’s students, PhD candidates, borsisti, or assegnisti, they must also provide the name of their supervisor.
A form is available on this website to request a notebook:
In case a notebook is requested, access will be provided to a jupiter instance backed a container. Users can choose the notebook configuration at login time. The configurations differ in terms of hardware resources as well as in terms of python libraries installed. Some configuration are also equipped with a GPU. The container backing the notebook will be deallocated if not used for 2 hour, so to free resources (including the few available GPUs) for other users. If the resources requested are not available when launching the notebook, the notebook creation fails and the user is informed. At this point, he/she can try with a container with less resources or they can wait for the resources to be available. The notebook is attached to a personal storage space that is persistent: files in this space will not be deleted also when the container is de-allocated and are available in all the notebook instances launched by such user. Additional libraries can be installed but should be reinstalled when the container is re-lauched. Access to the notebook is provided through a VPN.
For the Nvidia DGX Station please contact directly Prof. Gianni Poggi who manages this resources.
The system is described in more details in the paper below, please use it as a reference:
Botta, Alessio, and Elio Masciari. "A Research Infrastructure for E-Health Big Data Analytics." SEBD. 2021.