workload-collocation-agent

workload-collocation-agent copied to clipboard

Describe the bug Over time the value of this metric platform_rpq_read_latency_seconds is decreasing steadily and doesn't reflect reality.

To Reproduce Run WCA with uncore and derived metrics enabled: platform_rpq_occupancy, platform_rpq_inserts and platform_imc_clockticks

# requires WCA from
runner: !MeasurementRunner
  interval: 1.0
  node: !StaticNode
    tasks:
      - test1
    directory: docker
  # event_names:
  #   - task_stalled_mem_loads
  uncore_event_names:
    - [platform_pmm_bandwidth_reads, platform_pmm_bandwidth_writes, platform_cas_count_reads, platform_cas_count_writes]
    - [platform_upi_rxl_flits, platform_upi_txl_flits]
    - [platform_imc_clockticks, platform_rpq_occupancy, platform_rpq_inserts]
  enable_derived_metrics: true

Expected behavior Metric should reflect the memory system saturation and latency should increase only during high BW load situation.

Screenshots If applicable, add screenshots to help explain your problem.

Context:

• WCA master as of 2020.03.15