I was wondering if anyone could advise on what compute resources individual users will have in the operations-era RSP.
In the current DP0 version of the RSP the maximum compute resources for individual users are 4 nodes with 12 GB of RAM. How much is this likely to change when operations start? Just a rough estimate will be fine. The FAQs on the differences between the DP0-era RSP and operations-era RSP (FAQ: Technical Aspects of Rubin Science Platform Accounts for DP0) suggests that the operations-era RSP may have larger resources per user but doesn’t give specific details.
If anyone could help in the next two days (I am submitting a grant proposal on Thursday) that would be much appreciated.
Hi, for the RSP Notebook aspect 4 cores with 12 GB or RAM are the baseline estimate for individual users. The RSP Notebook aspect is intended for interactive data analysis. Users may also apply for computing on the production batch system for sizeable loads - @ktl can advise on the sizing of that.
We are investigating providing dask capability to the RSP Notebook instances but the sizing for that is not yet determined and is likely to be elastic.
These are estimates; the final numbers will be set by a cost envelope and therefore may be more if the cost of computing falls or the number of users is lower than expected.
Nominally, DMTN-135 §2.1 states that 10% of the Data Release Production compute capacity has long been planned to be made available to users (including batch computing and the Rubin Science Platform). The resulting core numbers during the Operations period are given in Table 43, ranging from 517 to 4664 cores as the survey progresses.
The original plan was to augment these guaranteed Rubin-provided resources with co-located supercomputing that could be applied for outside the project. We are thinking about similar models for US Data Facility operations, but any progress on that front is dependent on the Rubin Operations Proposal being reviewed and accepted.