Quest for a New Solver for EPANET 2

Abstract

One of the best known hydraulic water distribution modeling toolkits that is most used by both researchers and practitioners is EPANET 2. Initially the authors aimed to speed up such simulations by utilizing modern multicore processors in the code implementation. The 30-year-old linear solver was replaced in steps by seven different modern multicore capable solvers. Subsequently, speedup tests were carried out with small- (up to 1.6×103 nodes), medium- (up to 6.3×104 nodes) and large-sized (up to 6.3×105 nodes) test cases. None of the tested solvers was found to perform faster than the original solver for networks with a real-world character, although two solvers showed a speedup on medium and large water distribution networks. This is an example that strategies to reduce computation time can produce promising results in a theoretic research environment, but fail in practical engineering applications. Likewise, this paper highlights again the importance of considering realistic test cases during the implementation phase. Further, the authors point out why the different tested strategies in this work have not succeeded. Although two other issues (implemented hash table and node reordering) could be identified for potential improvement of the code, it was concluded that the original solver is still the fastest for practical system configurations.