Allocations-admins

Download

[email protected]

August 2015

4 participants
112 discussions

[LCRC Accounts] Project Allocation Request
by accounts＠lcrc.anl.gov 03 Aug '15

03 Aug '15

Hello, A change in allocation has been requested: Requester: dduke (Daniel Duke) Project: XRayFuelSpray Title: Fuel Injection and Sprays Studied Using X-Ray Diagnostics Description: The cavitation modeling component of the project is being performed using the OpenFOAM framework. A number of state-of-the-art numerical solvers have been implemented in OpenFOAM; namely incompressible-liquid and fully compressible homogeneous relaxation cavitation models and more recently a fully compressible homogeneous relaxation cavitation model which includes non-condensable gas modeling (HRMFoam). These codes were developed by Schmidt et al at the University of Massachusetts-Amherst, and we continue to collaborate with them on code development and efficiency improvements. In the last year, we have made some substantial improvements to the cavitation model and the parallel efficiency of its implementation on Blues/Fusion. With the support of LCRC technical staff and our collaborators at the University of Massachusetts-Amherst, we have implemented a custom build of OpenFOAM-3.0 on Blues against the system native MPI, and we have made the necessary changes to the HRMFoam solver’s source code in order to move from the older OpenFOAM-1.6 to the new 3.0 framework. This has led to substantial gains in parallel performance; up to 20% on 32 cores and 12% on 64 cores on Blues. A scalability study has been performed with the new code, and details are given in the Large Allocation Efficiency statement. In the new OpenFOAM-3.0 build, we have successfully implemented both Large Eddy Simulation (LES) and Realizable k-epsilon (KE) turbulence models. Furthermore, we have implemented code which allows us to take virtual projections of the three-dimensional simulations so that they may be quantitatively compared with the x-ray experiments. In the following year, we plan to extend high-fidelity cavitation simulations by undertaking longer simulation times in order to capture lower-frequency hydrodynamic phenomena which we have observed in our experiments. Improvements in efficiency will allows us to pursue longer run-time simulations. We also intend to simulate the effects of varying pressure and flow rate on cavitation, in order to compare against new x-ray fluorescence data from APS. The gas jet simulations which we plan to undertake in Q1 are performed with the CONVERGE CFD software, its use is already well established at LCRC. Tomographic reconstruction of gasoline direct injection spray measurements from APS will be conducted using the TomoPy software package developed at APS. We intend to run single-node jobs to efficiently perform multiple reconstructions of large (multi-GB) datasets. We have also successfully compiled a Paraview server on our Fusion dedicated nodes which allows us to rapidly visualize the results of our simulations and reconstructions. Using Blues and Fusion will allow us to achieve rapid turnaround of results during beamtime. This will be a significant aid to the experimental program. As the quantity of tomography work increases, we may consider purchasing more dedicated nodes. Current: undetermined amount Justification: Our requested allocation for next year will exceed 0.5M core-hours in order to encompass the gas jet simulations in Q1 and tomographic reconstruction work in Q2-4. The majority of core hours will still be dedicated to OpenFOAM simulations, which will be undertaken on both dedicated Fusion nodes and also on Blues, using OpenFOAM-3.0. We have demonstrated notable improvements in scalability on Blues as detailed above. A scalability study of OpenFOAM-3.0 performance with multi-million cell meshes has been undertaken, demonstrating excellent linear scalability. We observe 81% efficiency relative to single-CPU performance, scaling linearly up to 320 cores (20 nodes) with a 15-million cell mesh (a typical upper limit). We have implemented Scotch decomposition in order to ensure good load balancing & minimization of cross-processor communication boundaries. The present OpenFOAM simulations use static meshes, so load rebalancing at runtime is not a concern. We intend to run typical job sizes of 64 to 128 cores; smaller than the maximum job size for which good scalability has been demonstrated. Further details regarding scalability studies can be obtained by contacting the project PIs. Requested: 100000 A specific reason has been given: We would like to request an additional 100k core hours to continue ongoing numerical simulations work for rapid publication, which we intend to complete before FY2016 allocations are assessed. Our prior allocation was exhausted due to the need to perform additional simulations (for conference papers) which were not originally planned, and to perform tomographic reconstructions of x-ray data from APS. We will use the additional hours to perform OpenFOAM simulations; scalability test data are available upon request (up to 300 cores using system MPI). We intend to run 100-200 core sized jobs. If 100k allocation is not possible, we would benefit from any additional allocation that can be made available. This needs to be approved and the final allocation amount decided upon. Thank You, The LCRC Accounts System

1 0

[LCRC Accounts] Project Allocation Request
by accounts＠lcrc.anl.gov 03 Aug '15

03 Aug '15

Hello, A change in allocation has been requested: Requester: leicheng (Lei Cheng) Project: ElectrolyteRedoxFlow Title: discoveries of electrolyte for redox flow batteries Description: In this renewal work, we will continue focusing on fundamental properties of transition metal complexes as redox active molecules for non-aqueous redox flow battery application. In collaboration with experimental group (Prof. Thompson, University of Michigan), we have previously studied redox windows of some novel compounds. We now study the solubilities of these compounds in non-aqueous solutions as the capacity of this type of battery is directly proportional to redox window and concentration of active compound in solution. Insight gained from these studies will greatly guide design and synthesis of the compounds experimentally. In previously proposed work, we have successfully determined the B3LYP/6-31+G* and BP86/6-31+G* combinations as proper computational treatments for these system. The same methods will be used to calculate solvation energies of Cr, V, Mn and Fe based novel compounds. These energies will be calculated using both implicit continuum models such as PCM, CPCM and SMD methods, as well as explicit model with actual solvent molecules added to the quantum calculations. These solvation energies will then be compared with experimentally measured solubilities of the compounds to determine correlation between the two. Calculations with explicit solvent molecules will also be performed to further validate accuracy of the implicit model calculations as well as provide more insights on coordination details of the metal complexes by solvent molecules. Recently, interesting solubility trends of metal complexes in different binary organic solvents have been observed experimentally (private com munication). We will also guide our research efforts towards providing physical understandings of the observed experimental results. All calculations will be performed using Gaussian and NWCHEM. These two code packages are both considered very efficient and reliable for quantum chemistry calculations. Gaussian is not very scalable but it provides really robust geometry optimizer and the G3/G4 methods are built-in. NWChem is very scalable especially for higher level calculations such as coupled cluster methods and perturbation theories. After interesting molecule candidates and appropriate calculation methods are identified, derivatives of these molecules will be subject to high-throughput screening. The high-throughput screening will be performed using NERSC supercomputers. Current: undetermined amount Justification: The size of molecules we propose to study is very typical for the two computational software (Gaussian and NWChem) so they scale reasonably well for most DFT calculation. The higher level calculations will consume more memory as well as computational time. The NWChem code uses a globally-shared memory system, therefore one can request and utilize more memory through node parallelization. Requested: 50000 A specific reason has been given: We have used up the computing time allocated because the systems (metal organic complexes) of interest are a lot harder to converge than expected due to the presence of transition metals with different possible charge and multiplicity states. In addition, our experimental collaborators have discovered some more interesting systems with different metal centers that we are interested in investigating computationally. We are preparing a manuscript based on the work conducted under this proposal. These additional allocation will allow us to finish the extra computations needed. I really appreciate your consideration! Sincerely, Lei Cheng This needs to be approved and the final allocation amount decided upon. Thank You, The LCRC Accounts System

1 0