The rise of 400G changes the data center pattern
Data center operators and data center supply chain adhere to the path of 400G.
Higher Ethernet speed, cloud computing, the Internet of Things and virtual data centers all put forward more requirements for data center operators. Operators of large-scale data centers are promoting the wider adoption of 100G link and module technologies. At the same time, the shape coefficient and optical module of 400G have been at the key point of comprehensive startup, which is expected to be gradually launched in 2019. This transformation within the data center industry will double the density of the tried and tested QSFP28 (quarter small shape factor pluggable 28G) module at an amazing speed, and can increase the bandwidth by up to three times. Compared with four 100G modules, the total power consumption of the 400G module is even lower.
The 56G PAM-4 ASIC (application specific integrated circuit) chip used for network switches has become increasingly powerful in the process of continuous development. Users include enterprises such as Broadcom, Innovium, Nephos and Barefoot Networks. Therefore, the demand for the next generation of optical interconnection systems and modules continues to grow.
These new ASICs can provide 12.8 Tbps bandwidth, so the next generation switch can provide 32 400 Gbps ports. Or, if the architecture of the data center needs a higher base, these ASICs can operate in the reversible transmission mode to provide 128 100 Gbps ports. Traditional OEMs, such as Cisco and Arista, as well as white box manufacturers such as Zhibang Technology, QCT and Tianhong, are competing to produce these faster switches, many of which have already been launched. With the 400G switch ready, the key point is that the optical cable and copper cable interconnection system also need to meet the specification requirements and be ready to support the actual deployment.
Main driving force of 400G
What factors are driving the new demand? According to IDC's statistics, the storage requirements of the data center increase by more than 50% every year. It is estimated that digital information will increase to 40 zebytes by 2020 and 163 zebytes by 2025. There are several key contributing factors to this growth, including a wave of transformation to cloud storage, open systems, edge computing, machine learning, deep learning and artificial intelligence.
Virtual reality has just begun to be widely welcomed, and in the foreseeable future, the reality of driverless cars will become mainstream, which will bring exponential pressure to the data center infrastructure.
In order to meet the bandwidth requirements, the overall network architecture of a large-scale data center will be upgraded every two years. Therefore, it is inevitable that the plan will be scrapped.
The supply chain of the data center has accelerated its development, and has developed unprecedented scalable solutions with more powerful functions and higher energy efficiency. At present, 100G technology can provide the fastest connection for Ethernet links. The synchronous implementation of 100G and 400G Ethernet technologies will continue to grow in the next few years, and the latter will eventually occupy a dominant position and become the speed commonly used on switching chips and network platforms.
Keep ahead on the development curve
When we look into the future, what kind of situation can we see? A series of data center infrastructure solutions have left such a deep impression on people. Their design can meet the requirements of expanding large-scale data centers for improving bandwidth and power. The next generation solution makes full use of copper and optical cables, has high signal integrity and reduces delay and insertion loss, thus achieving the highest efficiency, speed and density.
The existing copper cable (DAC) has been able to reach the speed of 400G, while the optical terminals that can realize 400G switching connection are becoming increasingly perfect and ready for full release. The independent 100G Lambda optical transceiver and 400G optical transceiver are currently in beta sampling stage and will be put into market soon. Since early users will need higher bandwidth to deploy these products, 400G will start to heat up from the middle of 2019 to the end of 2019, even if the supply chain cost has not been reduced and price erosion has not started.
Many data centers will continue to deploy 100G CWDM4 optical terminals over longer link transmission distances, while the demand for 100G PSM4 is rapidly disappearing, and suppliers are also withdrawing from the market. As the independent 100G Lambda (100G-DR or 100G-FR) optical transceiver products have been available in early 2019, due to its low expected price, it is expected that these products will occupy the market of 100G CWDM4. In addition, the independent lambda products can also achieve direct interoperability with 400G optical transceiver in the branch topology.
As bandwidth migrates to a higher level, the industry will continue to witness the phasing out of 10G and 40G technologies, ECS These technologies have been replaced by optical terminals, direct attached cables (DAC) and active optical cables (AOC), which support 100G, 200G, 400G and higher speeds in communications between and within a series of data centers. QSFP-DD (quarter small shape coefficient pluggable dual density) optical transceiver will continue to play an important role in this spiral rise process.
Development of things
QSFP-DD optical transceiver is equipped with an eight channel electrical interface, and the data transmission rate of each channel can reach 50G. The QSFP-DD module can achieve a power of up to 20 watts (according to the QSFP-DD MSA version 5.0). With the help of the innovative radiator function, it can achieve 400G performance in a series of transmission distances. This is very important. Because advanced ASICs consume more power and emit more heat, the shape coefficient of QSFP-DD can efficiently dissipate these heat through effective thermal management strategies.
OSFP (octal small shape factor pluggable) has wider and deeper shape factors, and also supports 400G. Compared with OSFP, a major advantage of QSFP-DD optical transceiver is that it is fully backward compatible with existing QSFP+and QSFP28 optical transceiver. 56G PAM-4 technology is widely considered to be the key to realizing QSFP-DD and OSFP form factor optical transceiver. Now, a platform integrating QSFP-DD and OSFP optical module shape coefficient is being introduced to support 400G Ethernet in cloud applications. These new platforms provide downward compatibility with 100G ports and can be implemented in a staggered manner in data centers or enterprises.
Without considering the shape coefficient, the 400G optical transceiver requires the use of DSP's "gearbox" to create four 100G optical channels from eight 50G channels. This will become a key component in the supply chain and play an important role in the ability of optical transceiver suppliers to provide corresponding products and output to meet the huge demand of data center consumers. With the pursuit of product differentiation by optical transceiver suppliers, the supply capacity of 7nm DSP with low energy demand in 2019 will further interfere with this supply chain.
Molex has demonstrated 100G FR QSFP28 and 400G DR4 QSFP-DD products that meet the requirements of 100G Lambda MSA. The technology ecosystem of next-generation network equipment will promote 112G PAM-4 as a foundation to support 400G solutions for high-capacity data centers. The MSA specification emphasizes the technical design challenges to achieve optical interfaces when using the PAM-4 technology of 100G per wavelength, as well as the interoperability of multiple vendors. PAM-4 technology can realize 100G optical channel with a transmission distance of 2-10km, while for 400G channel, it can achieve a transmission distance of 2km on duplex single-mode fiber. PAM-4 platform can effectively lay a preliminary foundation and complete the migration to 400G in a cost-effective way. The technology platform aggregates four 100G channels per wavelength, which can support 400G versions for branch applications, such as 400G DR4, 400G FR4, and 4X100G.
Navigating in the evolution of 400G
Modern communication networks need higher bandwidth to meet the requirements of the global data explosion. As a result, the data center switch and optical transceiver market is growing and evolving rapidly. For building 400G network equipment, high-speed optical terminals, highly flexible and scalable optical transmission products, compact connectors and optical fiber management are all critical capabilities, serving a large number of telecommunications providers, enterprises and large-scale data centers.
It is very important to evaluate the optical fiber management of the data center at 400G and higher speeds. Products such as Molex optical fiber aggregation equipment can provide efficient solutions for systems with high optical fiber density to achieve organized optical fiber management. Such products can reduce or eliminate the blocked channels, provide a passive switching location, and are extremely compact without power supply or cooling. They can also bridge the gap between the current LC duplex plug and drop and the next generation of MPO high-density connector solutions. An example of this is a data center that was originally equipped with LC duplex optical fiber equipment. CWDM4 optical terminals were used under 100G, but the center is now migrating to DR4 under 400G, requiring parallel optical infrastructure.
Exceeding 400G
The ASIC supplier of the data center switch has announced that the 56G PAM-4 12.8Tbps ASIC can now be widely available, and the 112G PAM-4 25.6Tbps ASIC is currently being developed, which can promote the development of the 32 port switch. Each port can support the speed of 800 Gbps. Then, this function of ASIC will generate a series of challenges related to signal integrity, heat, power and loss. Here are a few examples. At the same time, the question will be raised whether the interconnection system can or should continue to be modular, and if so, Which shape factor can actually support this.
As data center operators are making plans to rapidly expand in a cost-effective manner, the design of 100G and 400G infrastructure can be optimized through close cooperation with suppliers who have the capabilities, expert experience and scalability required by today's data centers. The implementation process needs good arrangements to coordinate the data transmission between hundreds or thousands of components, achieve the optimal data center structure, reduce the overall risk and meet various dynamic requirements in the future.
