Alleviating the scaling problem of cosmological hydrodynamic simulations with HECA

Abstract

We (the CAGE team) present a possible solution to the scaling problem that is inherent to cosmological simulations of structure formation. With an increasing number of computational nodes the resources that are lost due to communication overhead and load balancing is growing and thereby limiting the problem sizes and/or resolution level that can be computed in a reasonable amount of time. To alleviate this problem, we propose the HECA (Hierarchical Ensemble Computing Algorithm). Instead of running a full-box cosmological simulation, we perform multiple (only limited by the number of processing nodes) zoom-in simulations concurrently that are independent of each other and thereby providing a perfect scaling to large core counts. In these simulations we can reach a much higher resolution level that would be unfeasible to achieve in a full-box simulation. We show that with the help of HECA we are able to efficiently use the ressources provided by modern petascale supercomputers to simulate a statistically significant sample of galaxies.