Real-time monitoring of large scientific simulations

We present a distributed framework that enables real-time streaming and visualization of data generated by large remote simulations. We address issues arising from distributed clientserver environments and guarantee good parallel load balancing. We apply progressive computing techniques and parallel, hierarchical data streaming techniques to reduce the "distance" between the simulation hardware and the systems where the actual visualization and analysis occur. We present a simple and efficient load balancing method that scales to arbitrary simulation sizes and does not introduce additional communication cost. We demonstrate the performance of our system with a molecular dynamics code and show its ability to allow monitoring of all the time-steps of a large simulation with negligible time overhead.