PAC Meeting Summary, March 30, 2021
Inaugural meeting in which committee was formally charged by Dr. Chaouki Abdallah and Daren Hubbard (link).
Neil Bright reviewed the new cost accounting model. Further details are at https://pace.gatech.edu/update-gts-research-cyberinfrastructure-cost-model. Key points: PACE charges based on actual computing and storage resources used. This consumption model is based on an approved rate study and cost per CPU-hour is based on corresponding hardware cost and standard benchmarks. Sponsored funds used for PACE services and commercial cloud services are granted an F&A Waiver (through at least FY22). The new model has more transparent accounting, can support rapid provisioning (vs old model where 6-8 month delays were the norm), hardware flexibility, a free tier for all PI-eligible faculty, and operational capacity planning and forecasting. Solicitations that explicitly target hardware/equipment in the RFP (e.g., DURIP, MRI, CC*) can still be handled, and such equipment would typically not be included in the model.
Lew Lefton reviewed feedback and concerns already voiced. Key points: The decision to change the model was made with insufficient faculty input, concerns about expiration of credits (both existing credits from converted Rich Building equipment and startup/retention), concerns about wait times in the new model as compared with dedicated resources, request to pool credits (library model) across larger groups, concern that F&A will be start being charged on PACE after initial 2 year evaluation period, cost share uncertainties (including using existing credits as cost share), concerns about negative sponsor reaction, concerns that new model will increase cost on grants, concerns that unfunded computational explorations (e.g., graduate student research and thesis work) will not be supported, concerns that this will create additional workload on faculty who will have to submit secondary proposals for compute time on national resources, concerns that national resources cannot provide the necessary level of computing support and control that local resources provide, e-routing checkbox confusion.
Additional concerns discussed, including queue system, how to manage credits on multiple projects and grants, need for HPC researchers to have limited but dedicated access to a large cluster for benchmark runs, software version churn
We will poll the Committee to rank/prioritize the concerns and begin working on the highest priority ones.