Strategies for Fault-Tolerant Distributed Visualization

As systems grow at an increasing pace to keep up with the demand for larger data and ever more detailed simulations, failures also occur more frequently on various levels. Traditionally, fault tolerance for visualization is either disregarded due to performance reasons or handled analogously to the state of the art for high performance computing (HPC), for example rollback-recovery. In this paper we summarize the inherent challenges of distributed visualization as well as the applicability and differences of fault tolerance methods used in HPC and distributed visualization. We derive complementary methods to increase the resilience of such systems and present visualization specific methods for failure semantics and masking.