High-performance computing input/output variability

Cameron et al. (2019) investigated the input/output (IO) performance variability of high-performance computing (HPC) systems under different system configurations, using the IOzone benchmark (Norcott, 2019) to measure IO performance across a range of settings. The response variable in the HPC dataset is the IO throughput variability, while the predictors include both continuous and categorical variables. We use a subset of the dataset from Cameron et al. (2019), which includes two continuous variables, CPU frequency and number of threads, and one categorical variable, the IO operation mode. The CPU frequency has 15 distinct values, and the number of threads has nine. The IO operation mode has 13 levels: Initial write, Fwrite, Random write, Pwrite, Rewrite, Randomread, Mixed workload, Stride read, Reverse read, Re-read, Fread, Pread, and Read. This dataset has also been used for statistical modeling and analysis in Zhang et al. (2021) and Xu et al. (2021).