// SCU10&13-Decommission

As many of you know, the Discover supercomputer undergoes regular hardware upgrades to stay current with the latest technology. The next upgrade, SCU17, is planned for this winter and will require some facilities upgrades to the cooling capabilities in addition to the compute upgrade. To prepare for the arrival of SCU17, we will need to decommission both SCU10 and SCU13 (ALL of the Discover Haswell nodes; compute, login, and gateway nodes). This will be done in stages to allow users to continue using the Haswells while we work on the decommissioning process.

We are working closely with the SCU17 vendor to assess the likely delivery date but with supply chain issues, this is particularly challenging. At the moment, we plan to start decommissioning SCU10/13 no earlier than Nov 14, 2022. The process will take between 5 and 6 weeks and it's not till they are all fully disconnected and moved out of the way that we can start the installation of the new Cooling Distribution Unit (CDU) which will take up to 3 weeks.

NOTE: We will reassess the delivery schedule on Nov 10 and may decide to postpone till after the Thanksgiving holiday and start on Nov 28, but users should plan to have the nodes begin to disappear on Nov 14.

We anticipate the nodes will be decommissioned in this order, but exact dates may change as we progress through this process and as we receive updates regarding SCU17's delivery date and accommodate holidays:

Week 1: 432 nodes

Week 2: 432 nodes

Week 3: 432 nodes

Week 5: 432 nodes (plus login and gateway nodes)

NOTE: We are providing access to limited HPC resources in AWS during this time for projects with small to moderate storage and I/O requirements. Please send email to support@nccs.nasa.gov if you are interested in discussing this option or if you have any questions about the decommission.