Hello, A new project on the LCRC cluster has been requested. Please forward the information on to the LCRC Allocation sub-committee. Applicant's name: Stefan Wild Applicant's institution: ANL Applicant's division: MCS Project Name: apsquared Project title: APS^2: Advanced Parallel optimization Solvers for APS upgrade simulations Associated funding: LDRD Other Systems: APS and MCS internal clusters Science: In this project we will test novel derivative-free optimization methods for estimating parameters in state-of-the-art accelerator simulations. Optimization is the single most time-consuming task in numerous simulation-based accelerator design problems because it requires running simulations for many different potential designs. Often, the underlying calculations are inherently serial (or only up to a point), so additional parallelization must occur at the level of the optimization algorithm rather than at the level of the design evaluation. We will test various implementations from a new class of model-based, structure-exploiting algorithms that use multiple design simulations performed in parallel to reduce the time to solution. Project description: Prototype implementations of optimization algorithms will be developed in Matlab/Octave and more advanced codes will be written in C and available in the open-source Toolkit for Advanced Optimization (TAO). The accelerator simulations will be run in ELEGANT (serial) and PELEGANT (parallel), which is developed by collaborator Borland (ASD) and requires the APS' SDDS toolkit (http://www.aps.anl.gov/Accelerator_Systems_Division/Accelerator_Operations_P...). We will test on a variety of problems, from those where the accelerator simulator is run on a single node (such as optimization of nonlinear dynamics in the APS upgrade, which requires serial computation of complex linear optics solutions) to those that scale up to hundreds of cores/dozens of nodes (such as when setting the strength of skew quadrupoles to recalibrate vertical beam sizes). We will test a variety of different algorithmic strategies, where the number of simultaneous designs generated by the optimization algorithm ranges from 2 to in the hundreds. The concurrencies of the simulator and the optimization algorithm are result in multiplicative parallelism, so we expect our largest runs will use thousands of cores for roughly 1 wall-hour. Smaller jobs will require fewer cores but more wall time. Project members will include three MCS developers (Wild, Munson, Sarich) and two ASD developers (Borland and Shang). We are requesting less time initially since we expect to be able to prototype many of our codes on smaller machines initially. Project URL: Requested allocation: 170000 Q1: 30000 Q2: 40000 Q3: 50000 Q4: 50000 Justification: N/a The requester has used undetermined amount hours of their initial startup project. In addition to approving an initial amount, please specify a Category and Subcategory for this project. For a list of the current selection of approved categories, please see: https://wiki.lcrc.anl.gov/wiki/Processes/Categories Once the Allocation committee has approved the project, please go to the Project Management page to create it: https://accounts.lcrc.anl.gov/projects.php Thank You, The LCRC Accounts System