Definition
Percentage of slice termination events followed by a re-instantiation of the same slice type within a defined window, indicating lifecycle decision oscillation around demand thresholds.
The Measurement Gap
Measurement Gap and Decision
Decision Supported: Is the lifecycle AI making stable create/terminate decisions, or is it thrashing around demand thresholds, creating service gaps and wasting orchestration overhead?
Measurement Gap:
Slice lifecycle management platforms track instantiation counts, termination counts, and average slice duration. What they do not track is whether a terminated slice gets recreated shortly after: the signature of a lifecycle model oscillating around a demand threshold. An operator can report "10,000 successful slice instantiations this month" without knowing that 2,000 of them were re-creations of slices terminated hours earlier. Each thrashing cycle carries real cost: orchestration overhead, service gaps during re-instantiation, state loss for stateful applications, and wasted signaling. The lifecycle model's termination decision looked correct at the moment it executed (demand had dropped below threshold) but the decision was wrong in context because demand was about to return. SLTR detects this oscillation pattern by tracking how often terminations are followed by same-type re-instantiations within a configurable detection window.
Citations:
- GSMA, "Network Slicing: Use Case Requirements," 2023. Describes lifecycle management requirements but provides no oscillation detection metric.
- 3GPP TS 28.533, "Management and Orchestration Architecture," Release 18: specifies lifecycle state transitions without measuring transition stability or oscillation frequency.
Formula
Plain Language: Count all AI-triggered slice terminations. Count how many were followed by a re-instantiation of the same slice type (same template, same coverage area) within N hours. Divide.
Notation:
SLTR = Terminations_with_reinstatement / Total_AI_terminations × 100
Variable Definitions:
- Terminations_with_reinstatement = slice termination events where the same slice type is re-instantiated in the same coverage area within N hours (count)
- Total_AI_terminations = all AI-triggered slice terminations in the measurement period (count)
- N = thrashing detection window; recommended starting point: 2 hours for dynamic eMBB slices, 6 hours for enterprise slices (hours)
Units:
Percentage. Lower is better.
Sign Convention:
0% = every termination was permanent (demand genuinely subsided). Rising SLTR = the lifecycle model is oscillating around demand thresholds.
Edge Cases:
- Scheduled termination/re-creation for maintenance windows: exclude from SLTR. These are deliberate, not oscillatory.
- Slice type changes (terminated eMBB, instantiated URLLC in same area): not thrashing. Different slice types serve different purposes.
- Very short-lived slices (designed to exist for less than 1 hour): use a proportionally shorter detection window (N = 30 minutes).
- Minimum sample: 50 AI-triggered terminations per measurement period.
Proposed formula, pending validation pilots.