Skip to content

Performance Impact

In a 5 minute timeframe the user may expect to see:

IntervalEvalsImpactType of environment
3s100HighReal-time
6s50ModerateHigh-security
9s33ModerateBalanced
15s20LowHigh-traffic
30s10LowResource-constrained
45s7MinimalResource-constrained
60s5MinimalMinimal overhead needed

Each evaluation walks all enabled detection recipes.

Performance Note

Jibril's detection engine is designed to be resource-efficient and performant. While much of its logic could be parallelized, it has parts that are intentionally serialized to ensure that no more than a single CPU core is utilized at any given time, even on high-traffic systems.

Configuration Profiles

Default (Balanced)

yaml
cadences:
  env-vars: 6
  file-access: 6
  network-peers: 6
  network-flows: 9

Best for: Most production environments

  • Balanced detection speed and resource usage
  • The default configuration for most use cases
Resource-Conscious
yaml
cadences:
  file-access: 15
  network-peers: 15
  network-flows: 15
  env-vars: 15

Best for: Resource-constrained environments

  • Minimal CPU impact ands still good detection speed
  • ~4 evaluations per minute per cadence
  • Most detection recipes should not need lower cadences
Mixed Priority
yaml
cadences:
  env-vars: 15
  file-access: 3
  network-peers: 15
  network-flows: 60

Best for: Specific threat models

  • Prioritize important detection categories
  • Balance resources across different patterns
  • Customize based on your risk profile

Important Concepts

Cadences affect

  • CPU usage
    Lower intervals = higher CPU usage
  • System responsiveness
    Resource competition
  • Detection speed
    How quickly threats are identified
  • Alert latency
    Time between behavior and detection
  • Reaction time
    Time between detection and a programmed action

Cadences don't affect:

  • Policy application
    Policies are applied sync, regardless of the cadence interval
  • Detection accuracy
    Pattern matching precision is not affected
  • Data collection
    eBPF always monitors continuously, nothing is missed
  • Event loss
    Event generation will still happen, but at a different rate
  • Detection types
    Same threats detected regardless of the cadence interval