SMOG – An Explicitly Composable Memory Benchmark Suite for Heterogeneous Memory
Andreas Grapentin, Felix Eberhardt, Andreas Polze · 2023
In response to the increasing spatial and temporal heterogeneity of memory resources and their properties, we propose SMOG, a fully configurable and explicitly composable benchmark suite optimized for high-resolution time-series measurements. In systems with highly heterogeneous memory resources, or memory resources whose non-functional properties change dynamically, it is necessary to understand the response of different workloads and their memory access characteristics to the memory resources in order to make viable placement decisions. SMOG has been developed as a benchmarking tool that mimics the memory access behavior of real-world applications, which are typically composed of multiple concurrent workloads with varying memory access patterns, while simultaneously investigating the memory properties of the system under test by observing the response of the memory access behavior to the systems memory characteristics.