No abstract

The extreme scaling pattern of the ComPat project is applied to a multi-scale workflow relevant to the magnetically confined fusion problem. This workflow combines transport, turbulence and equilibrium codes (together with additional auxiliaries such as initial conditions and numerical module), which aims at calculating the behaviour of a fusion plasma on long (transport) time scales based on information from much faster (turbulence) time scales. Initial findings of profile measurements are reported in this paper and indicate that, depending on the chosen performance metric for defining ‘cost’, such as time to completion, efficiency and total energy consumption of the mutliscale workflow, different choices on the number of cores would be made when determining the optimal execution configuration. A variant of the workflow which increases the inherent parallelism is presented, and shown to produce equivalent results at (typically) lower cost compared with the original workflow. This article is part of the theme issue ‘Multiscale modelling, simulation and computing: from the desktop to the exascale’.

No abstract

In this work, we consider the integration of MPI one-sided communication and non-blocking I/O in HPCcentric MapReduce frameworks. Using a decoupled strategy, we aim to overlap the Map and Reduce phases of the algorithm by allowing processes to communicate and synchronize using solely one-sided operations.Hence, we effectively increase the performance in situations where the workload per process is unexpectedly unbalanced. Using a Word-Count implementation and a large dataset from the Purdue MapReduce Benchmarks Suite (PUMA), we demonstrate that our approach can provide up to 23% performance improvement on average compared to a reference MapReduce implementation that uses state-of-the-art MPI collective communication and I/O.