Speeding up Science Through Parametric Optimization on HPC Clusters

Abstract

Science is constantly encountering parametric optimization problems whose computer-aided solutions require enormous resources. At the same time, there is a trend towards the development of increasingly powerful computer clusters. Geneva is currently one of the best available frameworks for distributed optimization of large-scale problems with highly nonlinear quality surfaces. It is an excellent tool to be used in wide-area networks such as Grids and Clouds. However, it was not user-friendly for scheduling on high-performance computing clusters and supercomputers. Another issue was that it only provided a framework for parallelizing workloads at the population level of optimization algorithms, but did not support distributed parallelization of the cost function itself. For this reason, a new software component for network communication -called MPI-Consumer -has been developed and published as open-source software. As a complement to our previous paper, which explained the MPI Consumer's system design, this article provides an extensive performance evaluation. Comparing the MPI Consumer with two earlier in Geneva developed network technologies shows that the MPI Consumer is an overall improvement in terms of performance. Furthermore, we analyze the impact of system components of the MPI Consumer by testing different configurations. It is observed that asynchronous client requests speed up the time to solution by up to 20%. Furthermore, the multithreading design proves to be very scalable, allowing for a significant speed-up even if on extremely high loads. Tests on GSI's Green IT Cube HPC cluster show that our measurements realistically reflect the expected behavior in a production environment.