Skip to content

PSM (Pooled Server Metrics, eBay)

What is this data? Pooled Server Metrics from eBay production infrastructure: 25 aggregated CPU/memory/network channels across many server nodes, minute-level, with 13 weeks of train and 8 weeks of labeled test. A large-scale IT operations benchmark.

  • Source: https://github.com/eBay/RANSynCoders (Abdulaal et al., KDD 2021)
  • Contents: 25-dimensional server metrics; 130k+ train / 87k+ test steps, 1-minute sampling
  • License: repository is Apache-2.0
  • Download: tsad-forge download psm
  • Loader: load_psm()

Known flaws

  • Missing values in train: train.csv contains many NaNs — the loader linearly interpolates (documented default behavior).
  • Coarse label boundaries: labels are aggregated per minute, so event boundaries can be imprecise.
  • Most prior reported numbers are PA-F1 based — not comparable.

Quick EDA (from local data)

Sample test channels with labeled anomalies shaded, plus the event-length distribution. Regenerate with tsad-forge viz after downloading the data.