Definition
Percentage of edge workload placements where the actual P99 latency under load meets the application's latency requirement, as opposed to placements that meet average latency but violate tail latency.
The Measurement Gap
Measurement Gap and Decision
Decision Supported: Is the placement engine placing latency-sensitive workloads where they will actually meet requirements under realistic conditions, or is it optimizing for average latency while tail latency violations go unmeasured?
Measurement Gap:
MEC (Multi-access Edge Computing) orchestrators optimize workload placement based on modeled latency, typically average latency under expected load. The placement decision is made, the workload deploys, and the orchestrator reports success. But the application's actual SLA is measured against P99 latency under peak load, not average latency under expected load. A placement that looks optimal at 50% node utilization may fail catastrophically at 90% utilization when tail latency spikes. No current MEC orchestrator validates whether its placement decisions actually meet P99 requirements under real load conditions after deployment. The orchestrator assumes its model was correct and moves on to the next placement. EPLC closes this gap by measuring placement outcomes against the application's actual latency requirement at P99 during peak-hour windows: the conditions that determine whether the placement truly works.
Citations:
- ETSI MEC 003, "MEC Framework and Reference Architecture," 2024: defines workload placement requirements but does not specify post-placement validation against tail latency objectives.
- O-RAN Alliance, "O-RAN Architecture Description," 2024: describes edge placement in the O-RAN context without providing metrics for validating placement quality against P99 latency requirements.
Formula
Plain Language: For each workload placement, measure the actual P99 latency during the peak-hour window. Compare to the application's stated latency requirement. Count placements where P99 meets the requirement.
Notation:
EPLC = Placements_where_P99_meets_requirement / Total_placements_with_latency_requirements × 100
Variable Definitions:
- Placements_where_P99_meets_requirement = workload placements where actual P99 latency during peak-hour window is less than or equal to the application latency requirement (count)
- Total_placements_with_latency_requirements = all placements for workloads with stated latency SLAs (count)
- P99 = 99th percentile latency, measured during the peak traffic hour for the specific edge node, not a 24-hour average
Units: Percentage. 100% = every latency-sensitive placement meets its P99 requirement under load. Lower values indicate placement decisions are optimizing for the wrong objective.
Sign Convention:
Higher is better. 100% is perfect placement quality.
Edge Cases:
- Workloads without explicit latency requirements: exclude from EPLC. These are cost-optimized placements where latency is not the constraint.
- Peak-hour definition varies by edge node: use the per-node peak hour, not a global peak.
- Newly placed workloads that have not experienced a peak hour yet: measure at first peak-hour occurrence.
- Minimum sample: 50 latency-sensitive placements per measurement period.
Proposed formula, pending validation pilots.