About

The objectives of the VECMA project:

Our central objective is to automate as many stages in VVUQ as feasible for the diverse set of multiscale computing applications deployed in the project, developing generic algorithms and approaches that will carry over to third party multiscale applications, and exploiting the required computational power offered by existing petascale and emerging exascale computing environments.

We will develop a software toolkit to support VVUQ for multiscale applications. This VECMA toolkit will be made available to the community in stages throughout the project, under an open source release. The toolkit will implement UQPs and VVPs which are optimized and flexibly applicable for multi-petaflops and exascale performance multiscale computing.

We will develop a systematic collection of Uncertainty Quantification and sensitivity analysis Primitives (UQPs), tailored to efficiently use current HPC infrastructures, and to incorporate expected requirements for use on exascale architectures. The UQPs will capture, in modular form, specific sub-activities required for the uncertainty quantification and sensitivity analysis in multiscale applications. We will develop UQPs to capture key control flow properties required for UQ, as well as optimization UQPs that can be flexibly added to optimize the performance and reduce the computational cost and minimize required data movements and communication instances of these procedures.

We will design and implement a constrained set of multiscale Verification and Validation Primitives (VVPs) to specifically capture and formalize activities which support the extension of single scale verification and validation procedures to multiscale settings, again tailored to the exascale.

We will establish VECMA middleware, adapting existing components to enable the automated execution of the VVUQ procedures created by the VECMA toolkit. The middleware will support the efficient execution of these on leading multi-petaflop computing resources, using task optimization techniques such as pilot jobs, and enable the prediction of the time to completion for these procedures on emerging exascale architectures using performance simulation.