SURFace: Low-Level Server Metrics from a Large-Scale HPC Cluster

July 2020 traces from the Lisa cluster at SURFsara provide over 100 low-level, server-based metrics, including page-faults and NIC byte transfers. This archive offers the finest granularity among open-source traces, supplementing scheduler-level logs with OS-level operational data. The dataset is intended to help validate assumptions about large-scale systems and analyze full-stack datacenter performance.

Use Cases

• Validating system design assumptions based on real-world, low-level operational metrics.
• Analyzing full-stack datacenter performance using granular server data like page-faults and network transfers.
• Comparing resource utilization patterns against higher-level job-based traces from other archives.
• Investigating the environmental impact of computing systems through detailed operational data.

Strengths

• Contains over 100 low-level, server-based metrics, a specific count stated in the description.
• Offers the finest granularity among published open-source traces, according to the source.
• Cross-platform presence on paperswithcode suggests recognized importance for the research community.

Limitations

• Key metadata is missing: row count, column names, dataset size, and specific last updated date are unknown.
• The provided tags from some entries (e.g., 'Geology', 'Geography') are clearly misaligned with the dataset's actual domain.
• Temporal coverage is limited to a snapshot associated with a July 2020 publication.

Provenance

Source

Lisa cluster at SURFsara, associated with research to be published in USENIX ;login:.

Collection Method

Operational trace archive of collected metrics.

Time Range

July 2020

Freshness

Last updated date is unknown across all sources.

Geography

null