LCRC has the ability to host dedicated nodes and storage for Argonne users/groups that require of resources to advance their research. This ability is contingent on limited availability inherent in the design of our existing clusters and requests will be dealt with on a first come first served basis.
Such users can contact LCRC (support@lcrc.anl.gov) to discuss their computing requirements and the anticipated time-frame for setting up these dedicated computational resources. Users/groups purchasing additional nodes can still apply for time on the shared LCRC computing resources.
By default, when adding nodes to an existing cluster, we will attempt to add the same or very similar CPU or GPU architectures, and can provision them with the same or more memory than the typical compute nodes.
This can be a cost-effective way to set up a number of dedicated nodes as they get the benefits of the existing cluster management. For large purchases over 40 nodes, LCRC will need to offset some labor and utility costs. Any purchase of nodes will necessitate buying the necessary cabling and potentially a switch as needed to incorporate it into the cluster. The research group will provide any licenses for special software.
As these nodes are made part of a production cluster, the support by LCRC of these nodes is tied to the lifetime of the cluster. When the cluster is decommissioned, we make no guarantees of running the nodes past the decommissioning. Owners can take physical ownership of the nodes if they so desire and LCRC will work with them at that time.
If a standalone mid-range cluster is desired, it may be possible if there is sufficient power and space available in the data center. In this scenario, the researcher/project is responsible for all acquisition and support costs of the cluster and their own storage and will need to cover some labor costs and all relevant utility costs. If connection to the LCRC’s storage is alternatively desired, then the researcher/project would need to purchase the necessary equipment (cables, switches, and potentially servers) to make such a connection.