← Lighthouse Briefings · Dedicated lines · UAE↔PH test · 2026-06-08

● TEST CONCLUDED · DECISION PENDING dxblink7 / sglink7 UAE ↔ Philippines Link 7, AE_PH Test start 2026-05-15 REAL DATA · 7d · per-3h

Dubai/Singapore dedicated line · UAE↔Philippines call quality test

7 days of real per-3h Kibana data (last window 2026-06-01 → 06-08). Dedicated lines lower RTT by ~17 ms vs the public route; packet loss and answer rate are on par. Keep/drop decision pending cost analysis.

Latency (dedi vs public · mean)

285 / 290 ms · vs 303 / 306 ms

Packet loss (dedi vs public)

1.20% / 0.91% · vs 1.16% / 0.95%

Answer rate (state 2 share)

88.2% / 88.1% · vs 88.9% / 88.5%

Dedicated share of volume

~7.5% AE · ~7.6% PH

01 What happened

The Dubai dedicated line (dxblink7) and Singapore dedicated line (sglink7) were provisioned to improve cross-region call quality between UAE and Philippines users. The test has been running since 2026-05-15; this briefing reads the most recent 7-day window (2026-06-01 → 06-08) of real user call data, queried directly from the Kibana ES proxy at per-3h fixed interval.

Across the three quality dimensions under evaluation: RTT shows a small, consistent improvement on the dedicated lines (Dubai dedi 285 ms vs AE public 303 ms; Singapore dedi 290 ms vs PH public 306 ms — both ~17 ms / ~5.5% lower across all 61 buckets). Packet loss is on par. Answer rate is also on par across all four routes (~88.1–88.9% state=2 share), with state=1 (failed) ~7.7–8.2% and state=3 (other) ~3.2–4.2% — the dedicated lines do not reduce connection failures.

The dedicated lines together carry ~7.5–7.6% of Link 7 AE_PH traffic on each side (transportState scope: dxblink7 102K records vs AE public 1.26M; sglink7 89K vs PH public 1.08M, last 7d). The technical conclusion is being handed off; the keep/drop decision is left to cost analysis and is awaiting input from the relevant business unit.

Note on dashboard methodology. The Kibana dashboard partitions the four routes differently per metric: the transportState panel uses a candidates.keyword wildcard on the dedicated-server IPs (128.14.7.94 for Dubai, 162.128.206.62 for Singapore), while the Count, Rtt, and Plr panels partition by targetTag and additionally require transportState=2 (so Rtt/Plr only reflect successful calls). This briefing follows the dashboard's per-panel filters faithfully — so the Count and transportState columns have slightly different denominators by design.

02 Link & topology

UAE user ──► Dubai node ═══ dedicated line ═══ Singapore node ◄── Philippines user

UAE users connect to the Dubai node; Philippines users connect to the Singapore node.
The dedicated line runs between the Dubai node and the Singapore node — that segment is the part under test.
Four routes are compared. The dashboard uses two different partitioning schemes (preserved here): Count / Rtt / Plr partition by targetTag (dxblink7 / sglink7) + region; transportState partitions by candidates.keyword wildcard on the dedicated-server IPs (Dubai 128.14.7.94 / Singapore 162.128.206.62) + region.

03 Methodology

Source: Kibana ES proxy → logs-footprint-voipstat-*, time field time. Window time ∈ [now-7d/d, now].
Common filter: name = kVoipTinyStat ∧ pp = AE_PH on every panel.
Per-panel filters (faithful to the dashboard):
- Count (targetTag-based): ∧ transportRTTAvg ∈ [0, 5000) ∧ targetTag = dxblink7|sglink7 for dedicated, ∧ baseInfo.ipRegion = AE|PH ∧ NOT targetTag = … for public.
- transportState (server-IP based, no RTT filter): ∧ baseInfo.ipRegion = AE|PH ∧ wildcard candidates.keyword = *128.14.7.94* (Dubai) / *162.128.206.62* (Singapore), negated for public.
- Rtt & Plr (success-only): same as Count + ∧ transportState = 2.
Aggregation: date_histogram on time with fixed_interval = 3h → 61 buckets (2026-06-01 00:00 → 06-08 12:00). Sub-aggs per bucket: avg(transportRTTAvg), avg(receiveLossAvg), terms(transportState), doc_count.
Metric definitions:
- transportState — transport connection state. 1 = connection failed, 2 = connection succeeded, 3 = other. State-2 share is the answer rate.
- transportRTTAvg — latency (average round-trip time, ms).
- receiveLossAvg — packet loss rate, already in percent units.
- doc_count per bucket ≈ call volume per 3h.
Acceptance: no absolute thresholds — relative comparison vs the public route only.
Data is embedded inline in this page (no live Kibana fetch at view time) so charts render offline.

04 Data — four routes, per-3h time series

Layout matches the Kibana dashboard — rows = routes, columns = metrics (Count · transportState · Rtt · Plr). line / state 1 (failed) state 2 (succeeded) state 3 (other)

Link 7, AE_PH, dxblink7, Count · Count of records

Link 7, AE_PH, dxblink7, transportState · % stacked

state : 3

state : 2

state : 1

Link 7, AE_PH, dxblink7, Rtt · avg transportRTTAvg

Link 7, AE_PH, dxblink7, Plr · avg receiveLossAvg

Link 7, AE_PH, AE public, Count · Count of records

Link 7, AE_PH, AE public, transportState · % stacked

state : 3

state : 2

state : 1

Link 7, AE_PH, AE public, Rtt · avg transportRTTAvg

Link 7, AE_PH, AE public, Plr · avg receiveLossAvg

Link 7, AE_PH, sglink7, Count · Count of records

Link 7, AE_PH, sglink7, transportState · % stacked

state : 3

state : 2

state : 1

Link 7, AE_PH, sglink7, Rtt · avg transportRTTAvg

Link 7, AE_PH, sglink7, Plr · avg receiveLossAvg

Link 7, AE_PH, PH public, Count · Count of records

Link 7, AE_PH, PH public, transportState · % stacked

state : 3

state : 2

state : 1

Link 7, AE_PH, PH public, Rtt · avg transportRTTAvg

Link 7, AE_PH, PH public, Plr · avg receiveLossAvg

How to read: each column is a metric, each row is a route. Compare top-to-bottom within a column to see dedi-vs-public. RTT column: dedi rows (1, 3) sit visibly lower than public rows (2, 4). transportState column: dedi rows are solid blue (state=2 only) — see the caveat below. Diurnal pattern is visible on the Count column for all four routes.

Summary table · 7-day window · 2026-06-01 → 06-08

Route	Records (7d total)	Peak /3h	RTT mean (ms)	Loss mean (%)	state-2 share (%)	vs public
Dubai dedicated (dxblink7)	90,770	2,294	285.2	1.200	88.18	RTT −17 ms
UAE public	1,257,913	39,911	302.6	1.162	88.90	baseline
Singapore dedicated (sglink7)	77,385	1,936	289.9	0.908	88.09	RTT −17 ms
PH public	1,070,728	33,217	306.4	0.945	88.53	baseline

Note on column denominators. The Count / Rtt / Plr columns and the transportState column have different record totals because the dashboard partitions routes differently per panel (see Methodology). transportState scope totals (last 7d): dxblink7 102,157 · AE public 1,263,526 · sglink7 88,787 · PH public 1,077,262. Within the transportState scope, all four routes show essentially the same state distribution — ~88% state-2 (succeeded), ~8% state-1 (failed), ~3-4% state-3 (other).

05 Comparative analysis · dedicated vs public

Latency · RTT

−17 ms (−5.5%)

Dubai dedi 285.2 ms vs AE public 302.6 ms · Singapore dedi 289.9 ms vs PH public 306.4 ms. Consistent improvement across the full 7-day window. Small in absolute terms but stable.

Packet loss

≈ on par

Dubai dedi 1.20% vs AE public 1.16% (slightly worse) · Singapore dedi 0.91% vs PH public 0.95% (slightly better). Differences are within bucket-to-bucket noise.

Answer rate

≈ on par

All four routes ~88% state=2. Dedicated: dxblink7 88.18% · sglink7 88.09%. Public: AE 88.90% · PH 88.53%. State-1 (failed) ~7.7–8.2% and state-3 (other) ~3.2–4.2% on all routes — dedicated lines do not reduce connection failures.

The dedicated lines deliver a small but consistent RTT win (~17 ms) and do not improve loss or answer rate. Combined with the dedicated lines reaching only ~7.5% of AE_PH traffic, the absolute quality gain across the corridor is modest.

06 Conclusion & next steps

Technical verdict

RTT improves ~17 ms on the dedicated lines; packet loss and answer rate are on par (~88% state-2 across all four routes; failed ~8%; other ~3-4%). The original hope of materially improving UAE↔PH call quality is only weakly supported: a real but small latency benefit on ~7.5% of corridor traffic; no improvement on loss or connection failures.

Decision status

Technical conclusion handed off. Keep/drop decision pending — cost analysis and management review. Inputs awaiting evaluation: monthly dedicated-line cost vs ~17 ms RTT win on ~7.5% of traffic, plus any non-quality rationale (DR, compliance, supplier contract).

Owners

VoIP Quality (data + technical verdict) · Network team (provisioning + supplier relationship) · Business / Finance (cost analysis & final keep/drop call).

Dashboard

Kibana · Link 7, AE_PH · open the Link 7 charts ↗

Query trace

Index logs-footprint-voipstat-* · common filter name=kVoipTinyStat ∧ pp=AE_PH · per-panel filters (Count/Rtt/Plr partition by targetTag with transportRTTAvg∈[0,5000) + optional transportState=2; transportState partitions by candidates.keyword wildcard on dedi server IPs) · per-3h date_histogram on time.