Definition
Mean absolute time error between AI-predicted demand arrival and actual demand materialization for slice instantiation decisions, measuring whether slices are created at the right time or too early/too late.
The Measurement Gap
Measurement Gap and Decision
Decision Supported: Is the lifecycle AI instantiating slices with timing precision, or is it burning resources on premature instantiation or missing demand windows with late creation?
Measurement Gap:
Every network slicing platform tracks whether a slice was created and whether it eventually carried traffic. None measure when the slice should have been created relative to when the traffic actually arrived. An operator can have 100% instantiation success rate while systematically creating slices 30 minutes too early (paying for idle infrastructure the entire time) or 10 minutes too late, leaving users on degraded shared capacity during the gap. The demand forecasting models that drive lifecycle automation are evaluated on magnitude accuracy (did we predict the right volume?) but never on temporal accuracy (did we predict the right moment?). SITA closes that gap by measuring the time delta between the AI's instantiation trigger and the actual demand arrival, separating early bias (resource waste) from late bias (service degradation) through the companion Timing Bias Direction metric.
Citations:
- NGMN Alliance, "5G Network Slicing for Vertical Industries," 2024: identifies lifecycle automation as a critical enabler but provides no temporal accuracy metrics for instantiation decisions.
- 3GPP TS 28.531, "Management and Orchestration of Network Slicing," Release 18: defines slice lifecycle states but does not specify measurement of timing accuracy between instantiation trigger and demand arrival.
Formula
Plain Language: For each AI-triggered slice instantiation, record when the slice was activated and when the demand it was created to serve actually arrived (defined as demand reaching a threshold percentage of the slice's designed capacity). The difference is the timing error. Average across all instantiation events.
Notation:
SITA = (1/N) × Σ |T_instantiation_i − T_demand_arrival_i|
Companion Metric: Timing Bias Direction:
TBD = (1/N) × Σ (T_instantiation_i − T_demand_arrival_i)
Variable Definitions:
- T_instantiation_i = timestamp when slice instance i was activated by the NSMF (datetime)
- T_demand_arrival_i = timestamp when demand on slice i first reached ≥50% of designed capacity (datetime)
- N = number of AI-triggered instantiation events in the measurement period (count)
- TBD separates directional bias: positive TBD = systematically early (resource waste), negative TBD = systematically late (service degradation)
Units:
SITA in minutes. Lower is better. TBD in minutes with sign.
Sign Convention:
Lower SITA indicates better timing precision. TBD sign indicates bias direction: positive means early instantiation, negative means late.
Edge Cases:
- Slices instantiated for scheduled events (enterprise customer pre-booking): exclude. These are deterministic, not prediction-driven.
- Demand that never arrives (slice instantiated but never reaches 50% capacity): count as timing error equal to the slice's total active duration. These are the most expensive errors.
- Slices that reach capacity instantly upon activation (demand was already present): T_demand_arrival = T_instantiation, timing error = 0. This is the ideal case.
- Minimum sample: 30 AI-triggered instantiation events per slice type per measurement period.
Proposed formula, pending validation pilots.