Center for Data-Driven Computational Physics
ConFlux, hosted at the University of Michigan, is a computing cluster specifically designed to communicate seamlessly and at interactive speeds with data-intensive operations. The project will establish a hardware and software ecosystem to enable large scale data-driven modeling of multiscale physical systems. ConFlux promises to produce advances in predictive modeling in several disciplines including turbulent flows, materials physics, cosmology, climate science and cardiovascular flow modeling.
A wide range of phenomena exhibit emergent behavior that makes modeling very challenging. In this project, physics-constrained data-driven modeling approaches are pursued to account for the underlying complexity. These techniques require HPC applications (running on external clusters) to inter-connect with large data sets at run time. ConFlux provides low latency communications for in- and out- of-core data, cross-platform storage, as well as high throughput interconnects and massive memory allocations. The file-system and scheduler natively handle extreme-scale machine learning and traditional HPC modules in a tightly integrated work flow — rather than in segregated operations — leading to significantly lower latencies, fewer algorithmic barriers and less data movement.
The ConFlux cluster consists of 43 IBM Power8 CPU two-socket “Firestone” S822LC compute nodes and 17 IBM Power8 CPU two-socket “Garrison” compute nodes. Each of the Garrison nodes will also host four NVIDIA Pascal GPUs connected via NVIDIA’s NVLink technology to the Power8 system bus. Each node has a local high-speed flash memory for random access. ConFlux also has 2 large memory nodes (2TB RAM).
All compute and storage is connected via a 100 Gb/s InfiniBand fabric. The IBM and NVLink connectivity, combined with IBM CAPI Technology will provide an unprecedented data transfer throughput required for the data-driven computational physics researchers will be conducting.
80% of ConFlux resources are restricted to use of research groups within the University of Michigan. The remaining 20% will be allocated to the external community based on data-driven modeling needs.