Efficient Static Well Assignment in Parallel Reservoir Simulation

Summary Load balancing poses a challenge to massively parallel reservoir simulation. Inefficient work distribution impedes scalability and leads to inefficient use of computing resources. Rather than addressing the more popular - and admittedly more important - topic of grid partitioning, this paper delves into the problem of the assignment of wells to processes. In the presented method, well activity mask is first obtained using a proxy model. Using this information, the well assignment task is translated to a bin packing problem and First Fit Decreasing algorithm is applied. Depending on the percentage of the simulation time that an average well is active, the results show between 8.2% and 73.8% improvement in terms of the number of bins. The number of bins is directly correlated with the maximum number of processes that can be assigned at most one well at a time which is beneficial in scenarios where the number of wells is greater than the number of processes assigned to the simulation run.