Paper Summary

This study investigates how to use IO500 and the Ellexus Mistral tool to analyze precise performance characteristics in order to tune IO performance on Astra, an ARM-based Sandia machine with an all flash, Lustre-based storage array. This article proves, through this case study, that IO500 serves as a valid storage benchmark, even for all flash storage. This research also demonstrates the need of employing fine-grained profiling tools such as Mistral for discovering tuning demand specifics. Overall, this study highlights the utility of a wide spectrum benchmark, such as IO500, in conjunction with a fine-grained performance analysis tool, such as Mistral, for understanding particular storage system performance and making more informed tuning decisions.

Pros and Cons

Pros

This article implements IO500 on a new device (AArch64).

Cons

This article lack of the experiments with other distributed file systems.

Brain Storm

Using a comparable system metrics monitor to debug HPC programs is an excellent method, but it requires significant system and hardware support.