Hello, A yearly allocation for the LCRC cluster has been requested with the following updated information: Submitter/PI: Feng Qiang Project Name: StructJuMP Division: MCS Project title: Parallel modelling of stochastic power grid in Julia Associated funding: ASCR,OE, GMLC Other Systems: Science: The objective of this project is to develop and test a parallel, memory-distributed, modeling framework for stochastic optimization problems arising in the optimization of power grid models under uncertainty. We plan to prove scalability and real-time capabilities that make the framework feasible to existing operational practices in power grid industry. Project description: Our previous parallel runs showed that the solver scales very well (>90% parallel efficiency on Intrepid, Mira and Fusion) but determined that the instantiation of the model is far beyond this efficiency. To remove this bottleneck we have moved from AMPL (who is an outdated modeling language for mathematical optimization with virtually no parallel capabilities) to Julia, which has been used by collaborators from MIT to develop JuMP, a modeling framework for optimization. We have build the StructJuMP, a JuMP extension for modeling structured optimization problems. StructJuMP can be used for modeling both linear and nonlinear programming and also fully integrated with our solver, so that we have successfully demonstrate the ability to model and solve stochastic optimization problems from power grid industry in parallel on HPC cluster. The parallelism in the instantiation of such problems comes from the particular structure these problem have (which has been very successfully exploited by the solver, as we mentioned above). Our preliminary results from a test problem on the Blues/Fusion cluster shows that the efficiency of the parallel problem generation is even better than the efficiency of the solver. However, we will need to investigate more into the model setup up time which currently doesn't seem to be scale well. We also plan to solve much larger problem instances than our testing problem. We expect this to scale up to the full Blues/Fusion. We expect a few full machine runs and other smaller runs that are needed to investigate scaling efficiency. Since we want to show a full integration with the solver PIPS, we will need to solve these problems to optimality. For this we expect to use around 80 thousand core hours in total for this year. Running our modeling and solver framework requires MPICH2 library. It would be great if Julia is pre-installed, however, we are also be able to build our own Julia executable from source. I expect the project will have 3 members (collaborators + summer student(s)). Industry partnership: Project URL: http://github.com/StructJuMP/StructJuMP.jl Current FY Hours Used: undetermined amount New FY Requested allocation: 80000 Q1: 10000 Q2: 20000 Q3: 20000 Q4: 30000 Justification: Storage requirements: 1TB will be sufficient Thank You, The LCRC Accounts System