As high-performance computing systems rapidly advance, the volumes of scientific data produced are a prominent issue. Data reduction methods, including compression and sampling, seek to alleviate these bottlenecks by significantly reducing overall size. Yet, research shows sampling yields lower overall reconstruction quality than lossy compression due to its inability to bound the error it introduces. This work presents and assesses the ability to estimate reconstruction error. We propose an absolute error estimation scheme using various metrics and evaluate its capabilities over several real-world datasets. Upon evaluation, our scheme is between 63.6% and 94.9% accurate at estimating absolute error and has an average 87.7% correlation with the error. Lastly, we implement and discuss applications of our error estimation, including a novel sampler enhancement that allows for user-specified targeted data quality and a process to achieve a user-specified storage constraint while yielding a better preservation of data than existing methods.

Progress in high-performance computing (HPC) systems has led to complex applications that stress the I/O subsystem by creating vast amounts of data. Lossy compression reduces data size considerably, but a single error renders lossy compressed data unusable. This sensitivity stems from the high information content per bit in compressed data and is a critical issue as soft errors that cause bit-flips have become increasingly commonplace in HPC systems. While many works have improved lossy compressor performance, few have sought to address this critical weakness.