Scalable libraries for graph partitioning

The key problem in efficiently executing irregular and unstructured data parallel applications is partitioning the data to minimize communication while balancing the load. Partitioning such applications can be posed as a graph-partitioning problem based on the computational graph. The partitioning problem is in the class of NP-complete problems; hence exact solutions are computationally intractable for large problems. However, good suboptimal solutions are sufficient for effective parallelization of a large class of these applications. We are currently developing a library of partitioners based on physical optimization and related methods. In this paper, we describe an outline of the different methods and current status of our library.<>