Average latency is fine, yet evenings collapse QoE
Question
Our network dashboards show average latency within targets and uptime is green. But every evening, users complain about buffering, page stalls, and call quality drops. Tickets bunch after 7pm, and peering graphs look busy but “acceptable.” Why does customer experience collapse when the averages look fine?
Answer ✅
Averages hide peak-time instability. Evening P95 latency, jitter bursts, micro-loss, and peering hot-spots can crater QoE even if daily means look great. SSM-Audit adds a stability band beside the KPIs you already track so you see whether service is calm and repeatable at the edges (the only time customers notice) instead of merely “good on average.”
What the bands would have shown 📊
• Evening latency P95 sliding A+ -> A0 (peaks widen despite stable mean)
• Jitter stability degrading to A- (burstiness rises around 7–10pm)
• Packet loss micro-rate tilting A0 -> A- (tiny loss spikes during buffers/freeze)
• QoE stall rate softening A+ -> A0 (video stalls cluster at prime time)
• Peering utilization stability dipping A0 -> A- (few ports saturate; imbalance grows)
• Last-mile congestion index weakening to A- / A– (segment overloads in specific geos)
What to do now 🛠️
- Band the peak window: track bands by 30-min buckets 18:00–23:00 for latency P95, jitter, loss, stall rate, and peering.
- Rebalance hot peers: when peering stability < A0, shift traffic, add capacity, or enable LAG hashing adjustments.
- Push edge relief: expand caches and enable prefetch/resume where QoE stall band slips; prioritize FQ-CoDel at bottlenecks.
- Protect DNS and control paths: if DNS latency band drops, route to closer resolvers; stagger maintenance away from peaks.
- Geo-targeted fixes: publish a last-mile band per cluster; pull forward upgrades where bands sit below A0 for 2+ weeks.
- Speak QoE, not averages: add the band panel to NOC/CS dashboards; act on worst-bucket bands, not daily means.
How SSM-Audit helps (practicalities) 🌟
• No additional infrastructure: runs beside your NMS, flow/telemetry, and QoE probes.
• Numbers unchanged: latency, jitter, loss, and QoE metrics stay the same; stability is a read-only overlay.
• Easy to use: spreadsheet/BI friendly; one lightweight nightly/weekly ritual.
• Universal language: A++ / A+ / A0 / A- / A– aligns NOC, capacity, peering, and care teams fast.
CLI 💻 — try our mini Calculator to identify the drift
(Mini CLI Download Page)
Feed your CSVs and see bands and drift at a glance (numbers unchanged).
# Evening latency P95 (prime-time buckets)
ssm_audit_mini_calc telecom.csv --kpi "Evening Latency P95" \
--out bands_lat_p95.csv --plot_kpi "Evening Latency P95" --build_id net
# Jitter stability (RTP/video)
ssm_audit_mini_calc telecom.csv --kpi "Jitter Stability" \
--out bands_jitter.csv --plot_kpi "Jitter Stability" --build_id net
# Packet loss micro-rate (0.1% bins)
ssm_audit_mini_calc telecom.csv --kpi "Packet Loss Micro-Rate" \
--out bands_loss.csv --plot_kpi "Packet Loss Micro-Rate" --build_id net
# QoE stall rate (video/audio stalls per session-hour)
ssm_audit_mini_calc telecom.csv --kpi "QoE Stall Rate" \
--out bands_stall.csv --plot_kpi "QoE Stall Rate" --build_id net
# Peering utilization stability (per port/partner)
ssm_audit_mini_calc telecom.csv --kpi "Peering Utilization Stability" \
--out bands_peering.csv --plot_kpi "Peering Utilization Stability" --build_id net
# Last-mile congestion index (geo/segment)
ssm_audit_mini_calc telecom.csv --kpi "Last-Mile Congestion Index" \
--out bands_lastmile.csv --plot_kpi "Last-Mile Congestion Index" --build_id net
Outputs you will get:
• CSVs with stability bands for each timestamp (e.g., bands_stall.csv).
• Drift charts per KPI (--plot_kpi) pinpointing prime-time breaks by geo/peer.
• Optional alerts if you enable thresholds in your setup.
Technical notes
Representation: x = (m, a) with a in (-1, +1)
Collapse parity: phi((m,a)) = m
Order-invariant pooling:
U = sum(w_i * atanh(a_i))
W = sum(w_i)
a_out = tanh( U / max(W, eps_w) )
Typical bands (example):
A++: a >= 0.75
A+: 0.50 - 0.75
A0: 0.25 - 0.50
A-: 0.10 - 0.25
A--: a < 0.10
Navigation
Back: SSM-Audit Q&A Series – IPO Readiness (Question 27)
Next: SSM-Audit Q&A Series – Semiconductor Quality (Question 29)
Page disclaimer
Illustrative scenario for research and education. Observation-only; do not use for critical decisions without independent validation.
SHUNYAYA 