Environment: German ISP Telekom, GPON. Multiple customers in the same building on the same splitter. Using Zyxel PMG3000-D20B SFP sticks as ONT replacement in sfp-sfpplus1.
All units were running stable for months. After an OLT-side event at 03:27 local time, two units lost link and never recovered.
Findings:
Router
Board Revision
Factory FW
ROS
PMG3000-D20B
RB4011iGS+
none
6.45.3
6.48.3
link ok
RB4011iGS+
r2
6.45.9
6.49.7
no link
RB4011iGS+
r2
6.45.9
6.49.19
no link
RB4011iGS+
r2
7.16.2
7.16.2
no link
RB2011UiAS
—
—
6.49.19
link ok
Two different PMG3000-D20B sticks tested per affected unit – same result. ISP's own external ONT works fine on the same fiber. ROS version appears irrelevant – board revision r2 is the common factor across all failing units.
All Zyxel SFP sticks were correctly registered with the ISP (Telekom) by serial number prior to the outage and had been working for months. Serial numbers were verified on-site. Registration is not the issue.
Question: Is there a known difference in the SFP controller or driver between RB4011iGS+ (no revision) and RB4011iGS+ r2 that could affect GPON SFP compatibility? Did something change in the r2 hardware that breaks link negotiation with this stick after an OLT re-ranging event?
What kind of "OLT-side event"? It could be that redilting optical power levels are lower than before and for some modules they dropped below Rx sensitivity (which might, due to low-cost manufacturing process, vary slightly from unit to unit)?
Telekom's own RPM (Remote Performance Measurement) data shows attenuation stable around 25 dB / 24 dB for days, then a hard drop to ~3 dB at 03:00 — that's a link loss, not a sensitivity issue. The external Telekom ONT (T-Modem) on the same fiber works fine and even shows better RPM values. A 3rd customer in the same building, same splitter, with the same Zyxel PMG3000-D20B model has been continuously online throughout. Telekom diagnosis confirmed there is a maintenance window between 03:00 and 04:00. So the trigger is OLT-side maintenance, not a slow optical degradation. Question is what exactly happens during that window that the Zyxel sticks don't recover from while the T-Modem and the 3rd customer's identical Zyxel do.
link ok
no link