Taming the beast some thoughts on exascale resiliency

The design and operation of high performance computing (HPC) infrastructures is, and always was, a huge technological challenge. Whenever the next generation of HPC system was about to be designed in the past, the community faced an ever-growing number of compute nodes and storage capacity, increasing heterogeneity of software, a new level of nonlinear computational load, questions of energy consumption and cooling, and many other non-functional issues. So far, everybody managed to deal with these issues in a exceptional and creative way. This time, it is about to become really hard.