This looks fine. Recommend we approve 80K for Q4. ----------------------------------------- Ray Bair Argonne National Laboratory On 7/18/17, 5:11 AM, "allocations-admins on behalf of [email protected]" <[email protected] on behalf of [email protected]> wrote: Hello, A change in allocation has been requested: Requester: wild (Stefan Wild) Project: UNEDF_Optimization Title: Optimization for the NUCLEI SciDAC-3 Description: We are building upon our optimization successes under the SciDAC-2 UNEDF (Universal Nuclear Energy Density Functional) project in the successor project NUCLEI. In this project, optimizations over parameter spaces corresponding to energy density functional parameters will be performed using the Argonne-developed POUNDERS optimization algorithm. We will work with DFT codes HFBTHO and HFODD and begin work with ab initio codes such as MFDn. We will also be performing sensitivity analysis and exploration of the design space, we have recently implemented load-balancing for the efficient computation of simultaneous parameter sets on thousands of nodes. We expect to use 120k core-hours for this task. For our parameter optimizations, we typically have 10-20 parameters and require on the order of 300 simulations. Based on the current simulation time, we would thus require 120 8-core nodes for up to 6 hours, or 6k core-hours for each version of the functional we want to optimize. We expect the number of core-hours will remain the same when using the new 16-core nodes. Past experience has let us to expect to do roughly 10 such optimizations in FY17, for a total of 60k core-hours for this task. POUNDERS is available in the Toolkit of Advanced Optimization (TAO), but we still prototype specialized versions of the algorithm using MATLAB/Octave on a single, master node. We will test updates to the TAO version on Blues and compare the performance and correctness against the MATLAB/Octave versions. As in the past, we will have roughly 4 users for this project. Current: undetermined amount Justification: Requested: 80000 A specific reason has been given: Require more time since we are expanding our developer base and moving toward optimization on Haswell and KNL nodes. This needs to be approved and the final allocation amount decided upon. Thank You, The LCRC Accounts System _______________________________________________ allocations-admins mailing list [email protected] https://lists.lcrc.anl.gov/mailman/listinfo/allocations-admins