Hello, A yearly allocation for the LCRC cluster has been requested with the following updated information: Submitter/PI: Iulian Grindeanu Project Name: homme_moab Division: MCS Project title: Scalable Advection Methods for Multi Tracers in Climate Codes Associated funding: SciDAC FASTMATH, ACES4BGC, DOE Other Systems: VESTA / MOAB -scalability , have trouble compiling there Science: One of the important goals of ACES4BGC project (Applying Computationally Efficient Schemes for BioGeochemical Cycles) is to implement and optimize new computationally efficient advection algorithms for large number of tracer species. This work demonstrates a framework that uses intersection algorithms developed in MOAB (Mesh Oriented datABase), linked with a HOMME dynamical core driver. Project description: Several studies were conducted with respect to intersection scalability, initial results were presented at 2015 SIAM-CSE Conference. Studies need to continue, as several scalability bottlenecks were identified: use HDF5 for parallel IO scales satisfactory on blues system, up to 256 nodes / 16 tasks per node; writing/reading scalability was demonstrated for a file with 6 billion entities, of which about 3 billion vertices and 3 billion hexahedrons. The generated mesh file uses 300Gb of allocated disk space. - some memory leaks need to be found / fixed, as it is found that after several thousands intersection iterations the system runs out of memory. I am planning to use some profiling tools, to help identify where the problem lies on. - the setup of the problem is rather cumbersome, HOMME code is written in Fortran 90, it has various dependencies, mostly HDF5, netcdf and pnetcdf, in the configuration used, and a cmake based build system; The new advection transport method is implemented as part of MOAB, in C++, and it uses a scalable mesh intersection algorithm, for the eulerian-lagrangian approach. - to communicate / locate intersection meshes on various mesh partitions, MOAB parallel infrastructure is used heavily; At the same time, HOMME already has a parallel infrastructure, but it is limited to communication within one layer of ghost elements; MOAB infrastructure allows arbitrary communication, but the effectiveness and advantages of this more complicated approach still needs to be demonstrated. MOAB approach would allow running the problem with CFL numbers larger than 1, although we need to demonstrate that this approach is theoretically sound. Industry partnership: Project URL: http://www.aces4bgc.org/ Current FY Hours Used: undetermined amount New FY Requested allocation: 40000 Q1: 10000 Q2: 10000 Q3: 10000 Q4: 10000 Justification: Storage requirements: Thank You, The LCRC Accounts System