According to data disclosed by Google, Facebook, etc., the internal traffic of these Internet giant data centers is increasing by nearly 100% every year. Currently, some Internet giants deploying 100G earlier have begun to seek higher-speed solutions, and the choice of next-generation data centers has become A topic that everyone is enthusiastic about.
The 400G Ethernet standard is preceded by the 200G Ethernet standard, which may reflect the industry’s mindset—more optimistic about 400G, or 200G is just a transition solution for 400G.
But directly from 100G to 400G is actually not very scientific.
- First of all, from the data center side, we need to rebuild the ultra-large-scale data center and define a new specification architecture. The requirements for rack power in the 400G era switch will be quite high, and the traditional air-cooling heat dissipation is more difficult.
- Furthermore, the 400G data center will use PAM4 technology, and the PAM4 technology will make the system less transparent and difficult to manage. The traditional NRZ technology together with the parallel technology can make the data center easy to manage.
In order to more flexibly adapt to the needs of the future data center and achieve a perfect transition to the 400G data center, Gigalight recently completed a low-cost data center internal parallel optical interconnection solution based on 200G NRZ transmission. This paper mainly compares 200G NRZ—Two parallel technologies in the solution, and two products as an example for simple analysis.
Fiber Parallel Solution—Is It Single- or Multi-Mode?
The traditional parallel optical module products are mainly based on optical interconnect technology of multimode fiber and have the advantages of high bandwidth, low loss, no crosstalk and matching, and electromagnetic compatibility problems. They have gradually replaced copper-based electrical interconnection products and are used in cabinets. The high-speed interconnection between the boards, the connection distance is up to 300 meters under the OM3 fiber.
At the same time, in order to apply to longer-distance transmission solutions, Parallel Single-Mode (PSM) optical modules have emerged, mainly using FP lasers to transmit 2km in single-mode fiber and DFB to transmit 10km applications, which is more difficult than multi-mode interconnection technology.
Data center cabling is a very complicated problem. The choice of multimode fiber or single-mode fiber has been the subject of heated discussion in the industry. There are also choices in major data centers. For example, in the 100G era, Facebook chooses single mode, Google chooses both multimode and single mode. At the same time, BAT (Baidu, Alibaba, Tencent) chooses multimode. From the perspective of cost, multimode fiber is expensive and multimode optical module is cheap. Single mode fiber is cheap and single mode optical module is expensive. Therefore, it is easy to combine the cost of fiber and optical module to obtain the relationship between distance and cost. Taking the 100G solution as an example, the cost advantage of a multimode solution is very obvious when the fiber distance is within 100 meters.
The parallel technology route is characterized in that each pair of multimode fibers respectively carries one optical signal. At present, IEEE’s 400G SR16 standard is a 16x 25G parallel solution, which requires 16 pairs of multimode fiber. It is far more than the 12-core MPO widely used in the 100G era, which will lead to a significant increase in cost; more importantly, multimode optical modules rely on The low-cost VCSEL optical chip solution, 2020, is likely to still require more than 12-core MPO’s 8-pair multimode fiber. The 400G SR4 that the existing 12-pin MPO can accommodate seems to be in the foreseeable future.
Therefore, in 2020, if there is no open and standardized multi-mode wavelength multiplexing technology (such as SWDM technology), low-cost VCSEL 100G technology can not achieve breakthrough, 400G multi-mode fiber solution cost advantage will no longer be obvious, single-mode fiber It may become mainstream in large-scale data centers, and short- and medium-range single-mode parallel solutions will be a cost-effective alternative to multi-mode parallel solutions.
——Yang Zhihua, “Top Ten Hotspots of Data Center Network Technology in 2020”
200G PSM8 vs. 200G SR8
Based on Gigalight’s unique PSM series product line, Gigalight recently released a new product—200G QSFP-DD PSM8, a high-speed product of single-mode parallel technology.
To achieve long-distance transmission, single-mode fiber with low dispersion loss must be used. To achieve high coupling efficiency between single-mode fiber and semiconductor, it is necessary to shape the light field emitted by the semiconductor laser to maximize the incident light field and the intrinsic optical field of the fiber.
And the 200G QSFP-DD SR8 uses an 8-channel 850nm VCSEL array that complies with the 100GBASE-SR4 protocol standard. The 200G QSFP-DD SR8 is a multimode parallel product. With the traditional VCSEL advantage platform, Gigalight uses a simple, efficient and reliable fiber coupling process technology to add a 45° prism between the laser and the fiber. The special material treatment of the fiber surface increases the coupling efficiency of the fiber to over 80%.
The two products are similar in that they belong to the optical modules in the 200G data center solution, and all use the QSFP-DD package, which can use the 16-core MTP.
The advantage of QSFP-DD is that the 1U panel can achieve a density of 36x 200G/400G, and it is forward- and backward compatible with QSFP, and is compatible with existing QSFP28 optical modules and AOC/DAC.
The main difference is that the 200G QSFP-DD PSM8 adopts an 8-way 1310nm single-mode fiber parallel solution with a transmission distance of up to 10km. The 200G QSFP-DD SR8 adopts a multi-mode fiber parallel solution and can travel over the OM4 fiber link. Up to 100m.
The multi-mode parallel solution is the core of the current data center development, and the transmission distance between the switch and the core switch is just within the scope of the multi-mode fiber.
Corning has introduced OM5 fiber in the past few years, but it has not caused the expected market reaction. The SWDM short-range wavelength division multiplexing scheme is only promoted by a few manufacturers—it is indeed lacking in the market.
In the near future, if a general enterprise data center wants to continue to use standard-certified solutions and reduce the cost of optical components, you can choose multi-mode parallel optics—after all, SMBs do not need as large a capacity as 400G.
However, if it is in the construction and deployment process of a very large-scale data center, especially considering the scalability of the system and the flexibility of the system, we should probably consider the single-mode parallel solution.
In the eyes of some people of insight, the single-mode parallel solution increases the number of fiber cores, but overall reduces the maintenance complexity, is easier to manage, and is easier to upgrade from 100G to 400G later. Without increasing fiber resources, the current 100G CWDM4 based on wavelength division multiplexing can only evolve to 200G FR4, and 100G PSM4 can be upgraded to 400G DR4).
——Li Mofei, “Review of Data Center: Cost Technology is Concise and Reconfigurable”
In general, the technology roadmap for the major switch and transceiver vendors shows a very clear and simple migration path for customers deploying parallel optics. So when optics are available and migrated from 100G to 200G or 400G, their fiber infrastructure still exists and no upgrades are required.
Reliability, product life, and maintenance costs are all interrelated. The parallel single-mode solution represented by 200G QSFP-DD PSM8 in total cost should be the cabling guide for large-scale data centers in the future.
Related Article Development of Optical Fibers, Cables and Optical Systems