DWDM-Based Data Center Interconnect Architecture and Operational Optimization at Bilibili

The article introduces the Bilibili System Department Network Team’s work on data center network planning, design, construction, operation, and optimization, focusing on both the optical (光层) and electrical (电层) layers of Data Center Interconnect (DCI) solutions.

It explains the need for multi‑active data center architectures, the role of DWDM technology, and the evolution of wavelength bands (C80, C96, C120, Super L, etc.) used for high‑capacity links.

The electrical layer discussion covers the progression from 10 G to 400 G, 600 G, and 800 G coherent modules, modulation formats (PM‑QPSK, PM‑8QAM, PM‑16QAM, PM‑32QAM, PM‑64QAM), baud rates, and spectral efficiency considerations.

Three implementation phases are described:

A. Infrastructure Deployment Phase – Physical optical channel design, device optical layer (line side), and device electrical layer (branch side) with specific wavelength allocations (e.g., 4‑line‑3‑router for online rings, full‑mesh for offline rings) and vendor‑specific equipment (H‑vendor C80, G‑vendor C96).

B. Service Integration Phase – Challenges of optical fault detection and recovery, a patented solution (CN202211192482.6) that automatically triggers OTDR diagnostics and pushes alarm data to operators. The alarm payload example is shown below.

{
    "time-created":"2022-03-07T17:21:46.000+08:00",
    "is-acked":true,
    "resource-alarm-parameters":{
        "perceived-severity":"warning",
        "is-cleared":true,
        "status-change":[{"perceived-severity":"warning","alarm-text":"Alarm cleared.","time":"2022-03-07T17:22:19.000+08:00"}],
        "last-changed":"2022-03-07T17:22:19.000+08:00"
    },
    "x733-alarm-parameters":{"event-type":"equipment-alarm"},
    "operator-state-change":[{"operator":"值班人员","text":"Alarm acknowledged.","state":"ack","time":"2022-03-07T17:47:52.835+08:00"}],
    "alarm-parameters":{
        "ne-name":"1OTM",
        "location-info":"0-TO 2-2-F1DFIU-4(EIN/EOUT)-OTS:1",
        "native-probable-cause":"FIBER_BREAK_POS",
        "probable-cause":"光纤发生中断，智能光纤管理系统启动探测，定位到光纤中断位置后上报此告警",
        "alarm-text":"光纤中断位置提示",
        "other-info":"POS(断纤点距离):27.8758km;..."
    },
    "common-alarm-parameters":{
        "resource":"201e7860-d3f8-11eb-bfbb-903fea16475f",
        "layer":"LR_Optical_Transmission_Section",
        "product-type":"WDM_OTM"
    }
}

The optimization reduced average optical‑channel incident handling time by 1–2 hours.

C. Full‑Network Efficiency Phase – Adoption of ROADM and flex‑grid technologies to enable dynamic wavelength allocation, CDC architecture for cross‑plane redundancy, and flexible grid spacing down to 6.25 GHz, improving spectrum utilization by about 20 %.

References and further reading are provided, including a patent and external articles on 600 G/800 G optics.