2019.02.18

• New - 4.2.1 (RC) will be released later today 2/18
  • Any feedback from anyone who has pulled it? (no one on call)
• Derived data:
  • In the Sampler Plugin
    • Pros: get it in the format you want it, low latency for on-node usage
    • Cons: overhead on source node
  • At the Store Plugin (Function Store):
    • Pros: store it in the format you want it
    • Cons: have to know the derivation
  • In a Downstream Store (aka database):
    • Pros: supports more complex operations
    • Cons: maybe requires more complexity and runtime processing
  • FunctionStore:
    • Raw and Simple Single and Multivariate derivations: RATE, DELTA, MIN, MAX, SUMS
    • Can also be used for subsetting the data to a hot store
• 1st order items of interest:
  • Imbalance in an application’s use of a resource
  • Variation over time in an application’s use of a resources
  • High-level aggregates:
    • Scores over an application run (e.g., high, low, variance);
    • Score variation from run to run
    • Scores of the workload
  • High-level indication of network state (e.g, % NICs with X stall/flit ratio)
  • Users: performance counters (CPI, Flops)
  • Combinations of system assessment with application assessment
• Examples shown...
• Quickstart Video from last week, still to be posted.