ERA Simulator — Edge Resilience Agent

Composite Serviceability Index

Eq. (5): S_p = R(T)^α · A^β · P_norm^1-α-β

Simulation Parameters

Total Requests (N_total): 1800

Failure Count (N_fail): 46

Total Runtime T_tot (s): 1638

Avg Downtime per failure T_down (s): 2.5

α (Reliability weight): 0.40

β (Availability weight): 0.40

Performance Targets

Throughput Target (Mbps): 1000

Throughput — Normal (Mbps): 800

Throughput — Degraded (Mbps): 600

Latency — Normal (ms): 1.5

Latency — Degraded (ms): 2.5

Reliability R

97.44%

= (N_total − N_fail) / N_total

Availability A

93.44%

= MTBF / (MTBF + MTTR)

Performability P_norm

71.36%

Geometric mean of TP & iLatency

Serviceability Sp
90.03%
Rα · Aβ · Pnorm1-α-β

ERA vs K8S-HPA Comparison

Intermediate Values

MTBF—

MTTR—

T_deg—

π_up—

π_deg—

P_ss,TP—

P_ss,iLA—

Multi-RAT Seamless Handover

5G-NR ↔ Wi-Fi 6 handover interruption time across three proxy/relay mechanisms

Scenario Controls

Handover Direction

5G → Wi-Fi Wi-Fi → 5G

RSSI Threshold (dBm): -60

Simulate jitter factor: 1.0×

Paper Results (Fig. 4)

Mechanism	5G→WiFi	WiFi→5G	KPI ✓
GStreamer	21.45 s	22.28 s	✗
FFmpeg	2.15 s	2.20 s	✗
Socat	0.07 s	0.29 s	✓

KPI target: < 1 s

Handover Time Comparison

Live Handover Animation

5G-NR Wi-Fi 6 UE

Network Slicing & Dynamic Bandwidth Control

Throughput adaptation under configured rate limits — ERA communication agent (Fig. 5)

Rate Schedule

Edit the bandwidth steps (Mbps) applied over time. Paper sequence: 1000→500→100→600→900

Adaptation delay (ms): 200

Overshoot factor: 0.02

Throughput vs Configured Limit

Avg Tracking Error

—

Max Overshoot

—

Settling Time

—

AI-Driven Compute Autoscaling

ERA proactive scaling vs baseline Kubernetes HPA across 100 test groups (Alibaba Cluster Trace 2018)

ERA Scaling Policy (Algorithm 1)

Scale-up threshold (%): 80

Scale-down threshold (%): 40

Failure threshold (s): 2.5

Prediction horizon (s): 12

Traffic volatility: Medium

Paper Results

Metric	K8S-HPA	ERA
Failures (>2.5s)	789	46
Avg Response (s)	2.10	0.91
Serviceability	60.72%	90.03%

Response Time: ERA vs K8S-HPA (100 Groups)

ERA Failures

HPA Failures

789

ERA Avg RT (s)

0.91

HPA Avg RT (s)

2.10

SLA-Driven Closed-Loop Control

Three-phase ERA resilience loop: Self-Awareness → Self-Learning → Self-Adaptation (Fig. 1)

Speed: 1×

👁

Self-Awareness

Continuous monitoring of KPIs, delays, errors, and pod status. Detects SLA deviations across the network and compute planes.

RSSI—

Pod CPU—

RTT—

🧠

Self-Learning

Autoformer predictor forecasts short-horizon traffic. Downtime risk estimation refines scaling and handover policies online.

Predicted load—

Downtime risk—

Policy update—

⚡

Self-Adaptation

Executes resilience actions: pod scale-up/down, bandwidth reallocation, handover trigger, drift correction.

Action—

Pods active—

BW allocated—

Edge Resilience Agent Simulator