SRC08. Application of Numerical Accuracy to the Selection of Lossy Compression Error Tolerances
Student: Jon Calhoun (University of Illinois)
Supervisor: Luke Olson (University of Illinois)
Abstract: Long running HPC applications depend on checkpoint restart to recover from failures and utilize multiple time allocations. Memory bandwidth and, in particular, file system bandwidth continues to be limiters to application performance and scalability. Compression techniques can be used to reduce data size limiting its impact. Lossless compression fails to generate high compression factors for floating-point data. Lossy compression generates noticeably higher compression factors at the expense of adding a small but controllable amount of error into the simulation. In this poster, a methodology for selecting lossy compression error tolerances is presented and evaluated that interprets compression error as numerical error. This interpretation allows defining error tolerances that hide compression error in the high-order truncation error already present in the simulation. This error tolerance selection methodology is shown to add error that does not effect the result of the simulation.
Two-page extended abstract: pdf