Performance Study Release v0.0.1

This preliminary release has all data for cloud executions, and the addition of analyses.

What's Changed

• container-sizes: analysis to look at size of layers by @vsoch in #34
• single-node: add preprocessing by @vsoch in #35
• compute-engine gpu: amg2023 by @vsoch in #36
• gke: gpu to add sizes 4,8,16 for amg2023 by @vsoch in #37
• google gke amg2023 runs size 32 by @vsoch in #38
• Reorganize results by @vsoch in #39
• Add amg2023 fom by @vsoch in #41
• Add lammps analysis by @vsoch in #42
• lammps: calculate hookup time by @vsoch in #44
• Add analysis osu by @vsoch in #45
• Add analysis mt gemm by @vsoch in #46
• analysis: kripke by @vsoch in #49
• analysis: kripke, add gke size 128 by @vsoch in #50
• Analysis: add minife by @vsoch in #54
• analysis: stream - preparing catalog and parsing plan by @vsoch in #55
• analysis: osu add sizes for plots by @vsoch in #56
• analysis: catalog for mixbench by @vsoch in #51
• analysis: mixbench cpu plots by @vsoch in #57
• analysis: mixbench cpu results by @vsoch in #58
• analysis: quicksilver by @vsoch in #59
• analysis: mixbench cpu by @vsoch in #60
• improve container similarity by digest by @vsoch in #62
• Update azure lammps by @vsoch in #63
• Update azure lammps by @vsoch in #64
• analysis: lammps matom steps per second by @vsoch in #66
• analysis: topology for aws by @vsoch in #67
• analysis: lammps gpu data was mixed by @vsoch in #69
• Update plot colors by @vsoch in #70
• analysis: look at scaling governor by @vsoch in #71
• analysis: fix missing amg results for gke gpu by @vsoch in #74
• analysis: magma by @vsoch in #75
• analysis: fix colors for stream, mixbench, magma by @vsoch in #76

Full Changelog: 0.0.0...0.0.1

Performance Study Release v0.0.0

This study aimed to measure performance of a select set of applications running on "bare metal" (i.e., actually bare metal or virtual machines) against containerization and Kubernetes between HPC clusters and cloud providers. We focused on several attributes of the setups, including (but not limited to) network performance and strong and weak scaling of applications. We

This first release includes raw, unparsed data for the majority of the experiments, which were completed in the last two weeks of August 2024. The study holistically involved assessing the performance and functionality of 14 HPC benchmarks and proxy apps on 8 different infrastructures: Azure AKS and CycleCloud, Google GKE and Compute Engine, AWS EKS and ParallelCluster, along with "bare metal" HPC equivalents on-premises (results not included yet).

This release will be updated when further data, intermediate data processing and scripts, and other analyses are available.