Abstract of LCTURB Project

The proposal at hand is a result of several discussions within the EUROfusion ETASC TSVV "Towards a validated predictive capability of the L-H transition and pedestal physics" led by T. Görler (PI of this proposal). A major part of this TSVV project is based on high-dimensional ab initio plasma turbulence simulations which are further complicated by the challenging conditions – e.g., strongly shaped flux surfaces and steep gradients – near the plasma edge. The corresponding massively parallelized codes have already been highly optimized in the past but improvements are urgently needed to further alleviate the enormous computational efforts needed for this task. One such idea based on lossy compression of data structures has emerged fairly recently and could be of great benefit not only to the project at hand but for the whole community. Motivated by statistically similar results between turbulence simulations performed in single and double precision, first promising tests with (lossy) compressed distribution function arrays have been conducted with the plasma turbulence code GENE. They hint towards a great potential of reducing the memory footprint and may therefore even open opportunities for significant speed-ups, e.g., due to cache and memory-bandwidth effects.

This proposal aims at asking for specialists' support for reviewing/extending these tests by adding more observables and explore further software packages beyond the previously employed ZFP library. Based on the experience gained, the specialist may then help to introduce corresponding techniques into the TSVV and further interested codes in the EUROfusion community. If possible, this could also be done and facilitated by developing a corresponding wrapper or library.