OS-Metrics: Performance Time Series (PTS) Dataset for Anomaly Detection and Change Point Detection

OS-Metrics is a 7-day, 15-second resolution, causally annotated performance time series dataset, used to evaluate anomaly detection and change point detection algorithms. It contains 6 metric files and 3 event log files. 1. Files Performance Metrics Data File Name Description Data Volume Invalid Data Volume /cpu_usage.csv CPU usage (percentage) 40320 0 /memory_usage.csv Memory usage (percentage) 40320 0 /memory_used.csv Used memory capacity (MB) 40320 0 /network_response_time_a.csv HTTP response time (milliseconds) observed by DCS-A 40320 1 /network_response_time_b.csv HTTP response time (milliseconds) observed by DCS-B 40320 12 /network_response_time_diff.csv The pointwise difference (in milliseconds) between network_response_time_b.csv and network_response_time_a.csv 40320 13 Invalid data is marked as -99999, which is caused by packet loss due to network fluctuations. Event Logs File Name Description Data volume /event_logs/event_logs_cpu.csv CPU Disturbance Event Log 13 /event_logs/event_logs_memory.csv Memory Disturbance Event Log 12 /event_logs/event_logs_network.csv Network Disturbance Event Log 17 2. File Structure Performance Metrics Data Files All metric files share identical structure: Column Type Description Time ISO 8601 datetime string (UTC), format YYYY-MM-DD HH:MM:SS Timestamp of observation Value numeric Observed metric value Event Logs Files All event log files share identical structure: Column Type Description id string Unique event identifier flag string Execution status start_time ISO 8601 datetime string (UTC), format YYYY-MM-DDTHH:MM:SS.SSSZ datetime string (UTC) when disturbance event began end_time ISO 8601 datetime string (UTC), format YYYY-MM-DDTHH:MM:SS.SSSZ datetime string (UTC) when disturbance event ended start_point integer Zero-based index of first affected data point in corresponding metric time series end_point integer Zero-based index of last affected data point (inclusive) in corresponding metric time series type string Disturbance category: cpu, memory, or network mode string Disturbance mode: `stable`, fluctuation, burst, or gradual intensity numeric Target disturbance strength; unit: percentage for cpu/memory, milliseconds for network duration integer Nominal duration in seconds (end_time − start_time, rounded down) ramp_up integer Duration in seconds of upward transition phase (0 if none) ramp_down integer Duration in seconds of downward transition phase (0 if none) duration_total integer Total effective duration in seconds (ramp_up + duration + ramp_down) 3. Remark OS-Metrics is evaluation-ready: all 42 disturbance events are causally annotated with start/end time, intensity, mode, and affected metrics. Metrics are sampled at 15-second intervals over 7 consecutive days. Event logs use zero-based indexing (start_point, end_point) aligned to the metric time series (first row = index 0). Remote observation bias (network_response_time_diff.csv) reflects real network path asymmetry, not artifact or error. All timestamps are in UTC; no daylight saving adjustments applied.