Embracing 5.5G
to Advance the Intelligent World

The Middle East has always played a unique role in history as a place of connections, bridging great geographic, economic, and cultural divides.

During the Middle Ages, Arab merchants used camels and ships to traverse the Silk Road, bringing silk and porcelain from China to India and the Malay Islands, where they would pick up spices, minerals, and cane sugar to trade for gemstones in Central Asia and ivory and gold sand from East Africa. Their powerful networks then went on to connect Northern Europe with honey, fur, and wood. This mercantile influence spanned the world and they played an important role in the Arab Empire's prosperity, in turn influencing the world's economy, science, technology, culture, and art.

Today, the six nations of the Gulf Cooperation Council (GCC) are returning to this model through economic restructuring. Digital and intelligent transformation are presenting the perfect opportunity to reduce their economic reliance on oil. The United Arab Emirates and Saudi Arabia, for example, have set in motion plans to become digital hubs for their surrounding regions. Multinational carriers in the region, like stc, e&, and Zain have jumped at the chance to become the new merchants of the digital era by supporting digital upgrades across the Middle East and Africa, and developing new strategies for cloud, DCs, and toB integration.

All have turned their attention towards building world-class ICT infrastructure.

The first year of commercial 5.5G in the Middle East

During the 2023 Global Mobile Broadband Forum (MBBF), six Middle Eastern carriers kicked off the region's commercial launch of 5.5G . Shortly after, those same carriers successfully completed 10-Gbps testing and more than 20 scenario-specific pilots of 5.5G technologies like RedCap, Passive IoT, and glasses-free 3D.

Exploring new 5.5G services for sustained 5G business success

So far, these carriers have been pragmatic in their approach to 5.5G. Their existing success in 5G has been the core driver of their continued investment.

By the end of 2023, 5G had a user penetration rate of over 40%, and over 40% of mobile data traffic came from 5G connections. Around 2018, many carriers had begun experiencing a decline in revenue as 4G reached maturity, but 5G has reversed this trend. Leading carriers have realized 20% to 45% revenue growth thanks to 5G, and FWA services are rapidly transforming the industry landscape in the Middle East.

Middle Eastern carriers are therefore eager to capitalize on 5.5G. Many have already identified a range of high-value 5.5G use cases that will help improve user experience and foster innovative new applications for individuals, homes, and industries.

Individual services: Immersive glasses-free 3D experience stimulates 10-fold traffic

Glasses-free 3D started making waves back in 2009 with the release of the sci-fi movie Avatar. Now, more than 10 years later, the core technologies behind glasses-free 3D are beginning to see large-scale commercialization as related technologies become increasingly reliable, with better experiences delivered, 3D content easier to produce, and more affordable devices available.

Shopping malls in the UAE and Oman are already using glasses-free 3D screens for advertising, searches for "glasses-free 3D" have surged over the past half year according to Google Trends, with over half of those searches being made in countries like Saudi Arabia, the UAE, and Oman. This is a prime example of how carriers are exploring new approaches to 5G monetization. One carrier found more success with 3D tablets than with glasses-free 3D phone screen protectors thanks to the better visual experience they deliver, deciding to bundle 3D tablets with FWA and phone screen protectors with toC services. These sales plans are set to launch in the near future. A second carrier plans to introduce glasses-free 3D use cases to their 5.5G Innovation Center and work with OSN (a local OTT service provider) and YouTube to build a 3D content ecosystem. A third carrier is planning to launch its own 3D livestreaming services and working with OTT service providers to establish a dedicated 3D service section and also plans to bundle 3D services into 5G subscription packages. And a fourth carrier has decided to first explore 3D screen protector services before turning to 3D content.

Many carriers are already designing premium service packages that will increase monetization and are directly working on developing a 3D content ecosystem. Many of these glasses-free 3D services are expected to launch in the first half of 2024.

Home services: Fiber-like to air-fiber FWA experiences

Another top area of focus for carriers is sustainable FWA growth.

The household penetration of FWA services in the Middle East already exceeds 21%, and the service is growing rapidly, even in markets with extensive optical deployment, such as the UAE, Qatar, and Bahrain. However, this rapid growth has created two pressing challenges for carriers. First, per-site user volume growth is increasing network congestion, which in turn harms user experience and increases complaints. This has made FWA user satisfaction a key area of concern. Second, fiber downlink rates are growing by over 40% every year. Mainstream fiber-based packages already deliver 300 Mbps to 500 Mbps rates, so, how can FWA maintain a competitive edge?

Over the past year, multiple carriers have worked with Huawei to explore these issues. What we've found is that there is still tremendous untapped potential in FWA if we upgrade offerings from unguaranteed 100-Mbps experience packages to reliable 300-Mbps experience packages. Our surveys have found that 70% of user packages can be upgraded, and 67% of users are willing to pay over 20% more for better experiences. This means that a reliable user experience is required throughout the user journey. Carriers will need to assess available network resources, coverage, and interference before high-speed FWA services can be provisioned. Provisioning will also often need higher resource scheduling priorities and higher performing CPEs. In some cases, onsite assisted installation will be needed to ensure the optimal installation position. Once services go live, carriers will need to be able to monitor user experience KPIs in real time for preventive optimization, and provide dedicated VIP service channels for consumers.

The UAE's du is an FWA pioneer. During MWC Barcelona 2024, du announced that it was ready to launch high-speed, low-latency packages for gamers and bundle more TV privileges with FWA packages ^[2].

FWA's potential remains huge as optical technologies evolve.

Industry services: 10 Gigabit City infrastructure facilitates industry digitalization

In 2022, Huawei signed an MoU with Saudi Arabia's Ministry of Communications and Information Technology on building a 10-Gbps Society. The country's 10 gigabit vision is now being implemented as the nation plans new urban areas. The enhanced network capabilities brought by 5G and 5.5G will enable carriers to provide ubiquitous basic connectivity while also helping them build global leadership in integrated national ICT solutions.

• A low-carbon "cognitive city": Saudi Arabia's NEOM aims to use 5G to connect people and things to build a "cognitive city". The tourism developer Red Sea Global then worked with Zain to build zero-carbon 5G networks for a massive future city project that would deploy 5G-powered IoT solutions to monitor important ecological sites, like coral reefs and sea turtle nesting sites^[4]. Carriers are now verifying multi-scenario IoT applications using 5G smart lamp poles. Some of these applications include intelligent security, electronic screens, EV charging piles, environmental monitoring, and autonomous driving.
• Digital and intelligent industries: Industries are now increasingly receptive of 5G. According to a survey by the UAE government, about 40% of enterprises plan to use some form of 5G application within the next year. For example, stc Group and the Saudi Ports Authority launched the Smart Ports Initiative to automate port operations through different scenario-based applications including remote crane operation, 3D visualization of cranes, license plate recognition, and facial recognition^[5]. etisalat by e& also released a 5G-Advanced white paper on multiple 5G products, including 5G mobile edge, 5G campus, and 5G networks for cars to promote the application of 5G and 5.5G in the government, transportation, healthcare, aviation, and oil and gas sectors^[6].
• Digital twins: Carriers in the Middle East are also actively testing and applying 5.5G technologies such as passive IoT and RedCap. In the UAE, passive IoT's long standby times and low costs are enabling a number of smart retail applications, connecting a wide range of goods to digital twins. This has increased tracking and stock allocation efficiency. In Saudi Arabia, low-cost 5G RedCap modules are also being used to explore new CCTV and smart security applications.

Building 10G networks using leading technologies

Leading carriers in the region have already completed 5.5G target network planning.

User-perceived speeds are expected to double in the next three years, with the average 5G downlink rate expected to jump from 100 Mbps to 200 Mbps by 2026, and the user-perceived FWA rate spiking from 165 Mbps to 500 Mbps.

Once this is achieved, carriers are likely to shift their focus towards pursuing ubiquitous 10G connectivity. By 2026, peak user rates will already be at 5 Gbps with under 20 ms latency, so the next milestone will be achieving 10-Gbps user-perceived downlink and sub-10-ms latency by 2030.

A consensus has already been reached by the region's carriers that TDD 3CC aggregation (300 MHz band) will be one of the fundamental differences between 5.5G and 5G. Because of this, they are working with Huawei on joint innovation in this area. For example, the region has already taken the lead in fostering dual-band META (2.6/2.3 GHz + C-band, dual-band 3CC) products that support simplified 5.5G deployment, as well as 3CC small cell products, to improve indoor user experience. The UAE's du is also already proceeding with large-scale TDD 3CC deployment. The extremely large antenna array (ELAA) technology has increased the number of dipoles from 192 to 384, enabling a peak rate of 5 Gbps, 30% higher user-perceived rates and 30% lower energy consumption^[8].

Saudi Arabia's stc is similarly deploying 4CC, which includes C-band and low and medium bands, to improve network performance and user experience, and speed up 5.5G evolution^[9].

At MWC Barcelona 2024, Huawei signed three MoUs with:

• du to transform the UAE into a 5.5G country
• Kuwait's Communication and Information Technology Regulatory Authority (CITRA) to build 5.5G smart cities
• Zain KSA on "All in 5.5G".

Many carriers are working with vendors like Huawei to seek smooth 5.5G network evolution, 5G super packages (which use new services like New Calling, 3D, and MR to build the 5.5G brand), and rate-based FWA pricing evolution. These demands have become an important factor driving innovation at Huawei.

High-quality 5G construction is creating a digital oasis

More carriers and governments are pursuing digital, intelligent, and low-carbon transformation.

The GCC nations have all launched national visions that will significantly increase the prominence of their digital economies and help them become global leaders, such as Saudi Arabia's Vision 2030 and the UAE's "We the UAE 2031". These governments are encouraging 5G network deployment through policy and funding support, which is successfully driving improvements in mobile network quality. Currently, four out of the world's 10 leading countries according to Speedtest are from the Middle East. Bahrain, the UAE, Saudi Arabia, and Oman have made significant commitments to carbon neutrality. Digital and intelligent transformation are also being clearly seen in many cities and industries across the region.

At the carrier level, Middle Eastern carriers are continuing to build high-quality communications networks as they digitalize their own operations. Success in 5G has helped many of these carriers improve brand value. From 2018 to 2023, stc's brand value increased by 85%, helping them move up 19 places in global rankings and making them the 11th top global carrier. e& also moved up to 14th place. Middle Eastern carriers are pursuing industry leadership with their 5.5G evolution. 2024 will be the first year of large-scale 5.5G construction in the Middle East, with leading countries like the UAE expected to achieve nationwide 5.5G deployment.

This has made the Gulf region significantly more economically dynamic. In 2023, Dubai Chambers saw a 43% increase in the number of registered new enterprises, with the growth rate of Abu Dhabi's startup ecosystem ranking sixth globally. The UAE plans to incubate 20 unicorn companies by 2031 and Saudi Arabia saw a 33% increase in its number of startups in 2023. Saudi Arabia also led the Middle East and North Africa with a record US$1.38 billion venture capital invested last year. This has made the Gulf region one of the prime destinations for global startups^[11].

5.5G will likely continue to drive growth in the Middle East's economy and society. This will trigger a chain reaction, supporting national visions like Saudi Vision 2030 and We the UAE 2031 and stimulating new digital vitality by attracting global talent and encouraging more innovation. This growing digital oasis will undoubtably create a new benchmark for the world in digital transformation.

Optical networks form infrastructure that deliver ultra-broadband, large-capacity, and low-latency connectivity for the digital world. From voice-based communications in the 1980s to modern broadband communications that typically support video applications, fixed networks have evolved through five generations. 2024 is the first year of F5G-Advanced (F5G-A), and AI-driven applications and the 3D display industry are now on the rise. Breakthroughs in AI foundation model technology, represented by OpenAI, are leading to the construction of numerous AI clusters and the adoption of many new applications. Furthermore, products based on immersive 3D display technology, represented by Apple Vision Pro, are entering the market on a large scale. The enormous computing and transport power that accompanies these developments urgently needs all-optical infrastructure. Optical communications networks have been presented with unprecedented opportunities and challenges, which can only be addressed through breakthroughs in the architecture and key technologies in the following areas:

• Backbone networks: Improvements in single-channel rates and system capacity to support increasing traffic between and within data centers – such as that driven by China's "East Data, West Computing" project – remain a major force driving evolution.
• Metro networks: Changing service traffic directions are driving transition from the traditional north-south network architecture to a cloud-centric, low-latency, high-quality, and T-shaped network architecture.
• Access networks: As 3D applications make their way into homes and AI foundation models into devices, the focus of evolution will shift from bandwidth improvement to high-bandwidth and high-quality experiences.

Backbone network architecture: Evolution of single-channel rates and system capacity

Capacity improvement is a key driving force behind the generational evolution of backbone networks. The acceleration of AI adoption is further increasing requirements for the transmission capacity of backbone networks. Backbone connectivity capacity can be improved in three directions: improving single-channel rates, broadening frequency bands, and evolving towards space-division multiplexing (SDM) systems.

Figure 1: 2030 all-optical network architecture

Direction 1: Improving single-channel rates

Improving single-channel rates is a major method of increasing backbone capacity. The evolution from coherent 100 Gbps with 50 GHz channel spacing to 400 Gbps with 150 GHz channel spacing has improved both spectral efficiency and capacity by 33%. Requirements for higher single-channel rates are often accompanied by expectations for lower per-bit costs and power consumption. To meet such requirements, single-channel rates will evolve towards coherent 1.6 Tbps with 500 GHz channel spacing. As we approach the Shannon limit, spectrum improvement is becoming increasingly difficult, requiring breakthroughs in both algorithm and component innovation. The current technical challenges are as follows:

• Improving spectral efficiency with higher-order modulation formats and optimized algorithms: The spectral efficiency of 1.6 Tbps with 400 GHz channel spacing (4 bit/Hz) is 50% higher than that of 400 Gbps (2.67 bit/Hz). In addition, the modulation format is changed from QPSK to CS16QAM, and long-distance backbone transmission must be covered. These changes require better modulation format designs and optimized algorithms.
• Components with 400G baud rates: Achieving 1.6 Tbps coherent transmission on a single transceiver requires components that support close-to-400G baud rates, such as AD/DA converters, modulators, and photodiodes. The latest industry developments can achieve modulation and demodulation at baud rates of about 250G, but there is still a significant gap to 400G baud rates.
• High-integration superchannel technology: Limited component techniques mean baud rate increases are becoming increasingly difficult. An alternative way to achieve 1.6 Tbps rates per channel is multi-channel integration, which means integrating two 800-Gbps channels or four 400-Gbps channels. A multi-channel solution uses lower-bandwidth components, but requires better photonic integration technology that is capable of solving issues like relative frequency control, multi-channel component yield improvement, reduced raw-material usage, and the warping of boxes and substrates.

Direction 2: Broadening the spectrum

System capacity is equal to the single-channel rate multiplied by the number of channels in the system. Increasing the number of channels by broadening the spectrum is another key direction for the evolution of optical system capacity. A C120+L120 system delivers two-times wider spectrum than a C80 system, as well as higher single-channel rates, increasing system capacity from 8 Tbps, which was delivered in the early days of coherent systems, to today's 32 Tbps. Beyond the C-band and L-band, the S-band and U-band have the potential to help broaden the spectrum. However, the optoelectronic components for the S-band and U-band, such as tunable lasers, photodiodes, and optical amplifiers, are not yet mature. Moreover, modulators and receive-end mixers are wavelength-sensitive and must be re-designed. Adding these two bands will increase stimulated Raman scattering (SRS), a non-linear effect in the system that causes the transmission performance of the C-band and L-band to deteriorate, increasing system-design and O&M complexity. These challenges require further research and investment by both academia and the industry. Even if the challenges are successfully overcome, system capacity can still only be doubled, meaning a low return on investment and difficulties for long-term evolution.

Direction 3: Evolution towards SDM systems

The growing complexity of technical challenges in relation to the two aforementioned solutions means that backbone network evolution towards SDM systems is becoming increasingly important and urgent. SDM was proposed 60 years ago, and there has been a large amount of academic research into key SDM technologies, with solid progress being made over the past dozen years. However, the following issues regarding SDM productization remain unresolved:

• Academic research has proposed multiple technical routes for SDM, including uncoupled multi-core fiber, coupled multi-core fiber, multi-fiber, few-mode fiber, multi-mode fiber, and orbital angular momentum (OAM) solutions. These routes each have their own advantages and disadvantages, and it is too early to say which of them will ultimately become dominant.
• SDM can be carried by different types of optical fibers, such as G.652, G.654, and hollow-core.
• It is not yet clear whether the system architecture would be WDM first followed by SDM or the other way around.
• SDM can have two, four, or more channels.

According to Huawei's analysis, the multi-fiber solution and uncoupled fiber solution are currently the two most mature. The former is a mature design based on parallel single-mode fibers, while the latter has lower crosstalk than other solutions and is made with ground-breaking drawing techniques. Huawei believes that breakthroughs and consensus are needed in the following four areas for SDM be commercialized across the industry:

The first is compatible system architecture. The WDM system was the key to the past success of optical backbone networks, meaning compatibility with the WDM system is necessary to commercialize the SDM system. SDM and WDM multiplex different resources, so the design through which these two systems are made compatible must be carefully considered. A solution with WDM followed by SDM is recommended because the compatibility in question can be realized through the use of simple-structure components, including WDM/SDM multiplexers, optical amplifiers, optical cross-connect (OXC) devices, and fan-in/fan-out (FI/FO) devices. This solution must support both multi-fiber and uncoupled multi-core fiber technologies. The two technologies are suitable for different types of networks, with the former designed for the upgrade of legacy fiber networks, and the latter designed for newly-deployed fiber networks. It is also possible to use the two technologies together by adding new multi-core fibers to existing single-mode optical fibers in order to form a hybrid SDM network. In terms of fiber types, SDM systems must support both single-mode fibers and hollow-core fibers. Hollow-core fibers have attracted a lot of attention in recent years due to their low nonlinearity, low dispersion, and low latency, and have the potential to become a disruptive fiber technology.

The second is simplified management system. Another key issue to consider when productizing SDM is adding one more dimension of multiplexing without increasing the complexity of network management. Huawei believes that WDM-like network management on an SDM system, despite the added space dimension, is the main focus and area of breakthrough for management and control technologies. Traditional WDM systems, multi-core SDM systems, and multi-fiber SDM systems all use the same network management system (NMS) to centrally manage OTU ports, wavelengths, and fiber cores. This involves a range of technologies including network resource pooling (establishing an OTU port resource pool, a wavelength resource pool, and a fiber core resource pool to share resources among different systems) and network as a service (NaaS) application programming interfaces (APIs).

The third is highly-integrated system architecture. Limited equipment room spaces require more closely integrated network equipment, meaning that increasing system capacity without requiring larger equipment is critical to productizing SDM. From a system composition perspective, a more integrated SDM system requires optimized component sizes including optical modules, optical amplifiers, OXC, and FI/FO devices. Optical module improvement in baud rates will inevitably encounter a bottleneck. Superchannel optical modules supporting higher port rates may necessitate a change in technologies, which will present the same challenges as mentioned earlier, including relative frequency control, multi-channel component yield improvement, reduced raw-material usage, and the warping of boxes and substrates. Challenges to improving optical amplifier integration include: multi-component rare earth doping used for integrated wide-spectrum amplification, high-power and low-cost pumps used to support multi-channel amplification, and multi-channel passive components with high integration and low loss. The main challenge in relation to OXC is that the additional space dimension requires high-degree OXC, posing serious challenges to 128-degree wavelength selective switches (WSS) that can only be addressed with OXC architecture innovation. In terms of optical connections for an SDM system, another technical challenge is direct connection between multi-core fibers and single-mode fibers, and also between multi-core fibers and optical modules.

The fourth is converged applications such as communications, sensing, and security. SDM can support more application scenarios than large-capacity communications. For example, technologies like distributed optical fiber sensing and quantum key distribution (QKD) that have emerged in recent years are significantly expanding the application scenarios of optical fibers, enabling converged applications of integrated sensing, communications, and encryption. With multiple spatial channels, SDM can be used for applications that integrate communication capacity expansion, integrated communications and sensing, and security. For example, in a four-core system, one core is used for traditional optical communications, another two cores are used for QKD-based communication and key negotiation, and the fourth core is used for distributed optical fiber sensing.

Metro network architecture: Cloud-centric low-latency and high-quality networks

AI's massive requirements for computing power will transform metropolitan area network (MAN) traffic from a north-south model to a T-shaped model that goes both north-south and east-west, and cause traffic to shift from user-to-user to mainly user-to-cloud. In addition, services like AI, VR, and smart manufacturing require higher bandwidth and lower latency. It is essential that the metro network architecture supports one-hop access to the cloud and ultra-low latency between AZs to meet user requirements for high-quality experiences. Legacy networks with an architecture that features hop-by-hop forwarding and multi-ring stacking can no longer meet the latest requirements. Key technologies like all-optical interconnection, fine-grain OTN (fgOTN), and optical-layer digitalization are required to ensure high bandwidth and low latency for the optical metro network architecture.

• All-optical interconnection Centrally managing the wavelength resources of metro network access rings through OXC, with the resources in optical fibers shared among the rings, will enable the resources of each ring to be flexibly adjusted in order to effectively cope with traffic imbalance between rings. In addition, OXC eliminates the need for optical-to-electrical conversion, ensuring one-hop access to the cloud with ultra-low latency.
• fgOTN OTN technology is deployed at the network edge and customer edge to provide hard pipes for services. The hard pipes feature physical isolation and high security, while supporting the continuous evolution of single-channel rates as well as hitless and fast bandwidth adjustment, thus meeting service requirements for increased and more flexible bandwidth. Furthermore, fgOTN hard pipes can provide deterministic latency, enabling deterministic experience for premium industry customers.
• Optical-layer digitalization Optical performance visualization can be extended to the edge of optical networks. Traffic flows are changed to be based on a deterministic cloud-centric model that facilitates optical network planning and O&M. Digital optical-layer technology can be used to accurately model the optical layer, thereby visualizing the optical network. This supports accurate service provisioning in the planning phase, intelligent fault location in the O&M phase, and quick service recovery on highly-reliable automatically switched optical networks (ASONs). When used alongside an operation app, this will help monetize network O&M.

Access network architecture: High bandwidth and high-quality experience

Optical access network technology has evolved along the following path: PON > GPON > 10G PON > 50G PON > Beyond 50G PON, providing high-bandwidth networks for access users. Emerging new services (e.g., AI, AR, VR, and holographic display) and applications (e.g., smart manufacturing) require access networks that deliver not only higher bandwidth, for example, 2 Gbps for 8K VR with high-quality experience, but lower latency, lower jitter, and security isolation. For example, manufacturing requires deterministic μs-level latency and jitter. In addition, vertical industries require multiple service networks to be carried on a single physical network and be strictly and logically isolated to uphold the service level agreements (SLAs) of each service network. Therefore, future optical access networks will not only increase bandwidth towards beyond 50G PON, but transition from a best-effort model to one that ensures differentiated services with guaranteed quality. Deterministic bandwidth, latency, and jitter, as well as high reliability, will be provided for diamond services (e.g., for high-end villas), while deterministic bandwidth and sub-ms-level latency and jitter will be provided for silver services (e.g., for apartments). On-demand bandwidth and ms-level latency and jitter will then be allocated for copper services (e.g., for remote rural areas) at affordable prices.

The high-bandwidth and low-latency requirements of beyond 50G PON can potentially be met by the following three technical solutions:

• Direct modulation and detection This solution is based on the continued evolution of current-generation technologies. It provides a time division multiple access (TDMA) mechanism and supports low-cost, point-to-multipoint (P2MP) access, but cannot provide deterministic latency. The solution relies on higher-bandwidth optoelectronic components. Furthermore, direct detection means a 3 dB decrease in receiver sensitivity each time bandwidth doubles. Bandwidth improvements will also bring higher dispersion penalty. Reusing the optical distribution network (ODN) will require much higher transmit power than that provided by a 10G PON network, posing a great challenge to high-power lasers.
• WDM/FDM & direct detection Frequency-division multiple access (FDMA) or wavelength-division multiple access (WDMA) is implemented through multiple wavelengths, and independent high-quality P2P access is implemented through a single frequency or wavelength, providing users with μs-level latency and jitter. This solution does not require a high power budget, but does require optical network units (ONUs) equipped with tunable lasers. Access networks are highly cost-sensitive, meaning low-cost tunable lasers are a key technical challenge for this solution.
• Coherent solution Through the application of digital subcarrier technology, coherent networks can also deliver high-quality P2P access and ensure deterministic latency with physical isolation. Coherent systems can achieve high bandwidth with low-bandwidth components and higher-order modulation, which both realizes higher receiver sensitivity and meets the power budget requirements of ODNs. However, this requires ONUs, which are cost-sensitive, to be equipped with intrinsic lasers, creating a challenge regarding laser costs.

Looking ahead to 2030, ground-breaking transformations in fields like AI foundation models and immersive 3D interaction will require an all-optical network architecture with higher bandwidth and lower latency. Developing an all-optical network architecture system will require breakthroughs in key technologies related to backbone networks, metro networks, and access networks to support the connectivity required by the massive and constantly surging number of digital services. At Huawei, we believe that all-optical network architectures will help make people's lives more digitalized and intelligent and serve as the connectivity foundation for an intelligent world.

In the digital age, every innovation in network technology has the potential to initiate a major industry transformation. Application-aware IPv6 Networking (APN6) uses the programmable space in IPv6 packets to bring application information (parameters such as identifiers and/or network performance requirements) into a network. This makes the network aware of applications and their requirements, and enables refined network services and accurate network O&M, facilitating innovation in cloud-network-edge-device convergence.

As differentiated application requirements and diverse network technologies and services emerge, a wide range of applications designed for these requirements are being developed. Specific application scenarios include:

• Mobile interconnection applications oriented towards enhanced Mobile Broadband (eMBB) such as HD video, virtual reality, cloud storage, high-speed mobile Internet access, and AI
• Device interconnection applications oriented towards massive Machine-Type Communications (mMTC) such as environmental monitoring, smart meter reading, and smart agriculture
• Special applications oriented towards Ultra-Reliable and Low Latency Communications (URLLC) such as Internet of Vehicles (IoV), industrial control, smart manufacturing, and remote surgery
• As industries migrate to the cloud, cloud applications are used over the Internet in industries such as finance, manufacturing, education, and healthcare, as well as by individual users and households. They are reshaping industries and people's social activities, entertainment, and everyday lives. Major applications include smart cities, private networks for financial cloud, cloud-based healthcare, online education, remote offices, cloud private lines for e-commerce, and cloud gaming.

Diversified application scenarios present new challenges for network O&M. Effectively realizing refined network services and accurate network O&M is key to meeting differentiated application requirements, ensuring SLAs and promoting sustainable network development and evolution. In-depth application-network integration is a feasible path to achieving this goal and has catalyzed the emergence of the APN6 technology.

APN6 technology: Two solutions

How does application-aware IPv6 networking realize intelligent conversations between networks and applications? Key components of the APN6 framework include applications, network edge nodes, and head nodes, intermediate nodes, and tail nodes that provide network services based on APN information (Figure 1).

Figure 1: APN6 framework

Depending on where APN information is added, there are two main types of APN6 solutions, application-side and network-side, each with its own advantages and disadvantages.

In the network-side APN6 solution, application and user information is added to packets by network border equipment. The advantage of this solution is that network border equipment and network equipment providing services based on APN information are managed and controlled by the same carrier or enterprise and belong to the same trusted domain. Therefore, there are no privacy or cybersecurity issues. Its disadvantage is that the application and user information added to packets may be inaccurate or incomplete. This is because the information is added by network border equipment, instead of applications, but the equipment cannot acquire the information of certain applications.

In the application-side APN6 solution, application and user information is directly added to packets by applications. Its advantage is that it ensures the application and user information added to packets is accurate and complete. The disadvantage is that it creates privacy and cybersecurity challenges, as information must be transmitted between different trusted domains such as terminal devices, network infrastructure, and cloud services.

The IETF draft (draft-li-apn-framework) defines the application information (APN attributes) carried in APN6 packets, including application identifier information (APN IDs) and application requirement parameter information (APN parameters). An APN ID provides information that helps the network distinguish between different application flows and different users (user groups) of a certain application or certain type of applications. Such information may include Application Group IDs and User Group IDs. APN parameters optionally carry information that may include parameters related to application requirements for network performance, such as bandwidth, latency, jitter, and packet loss rate.

APN6 promotes innovation in the next-generation Internet protocol

APN6 generates significant changes to Internet architecture and promotes innovation in the next-generation Internet protocol.

Network architecture consists of three key aspects: identification, forwarding, and control. These aspects have different impacts on the network architecture. Identification can cause fundamental architecture changes as, when it changes, both forwarding and control must also change, which alters the entire architecture. However, identification changes are also extremely difficult. For example, changes in address identification from IPv4 to IPv6 require upgrade in the entire Internet infrastructure.

The US launched five Internet architecture research projects through Internet2 around 2010. These projects covered Internet architecture transformation, with the more well-known ones including Mobility First and NDN. These projects provided a wealth of valuable ideas about Internet architecture, but were ultimately unsuccessful for three reasons:

1. The Internet's influence is too large, meaning it was incredibly difficult to push for transformation.
2. Transformation was incompatible with the evolution of existing Internet technologies, and fully upgrading network infrastructure was too costly and time-consuming.
3. Existing network software and hardware capabilities could not fully support new network technologies.

APN6 brings important changes to the network identification system. APN6 can be understood as introducing APN IDs, in addition to IPv6 addresses, so that packets contain both IP addresses and identity-like IDs. APN6 means more than just enhancement and innovation in IPv6. It provides a foundation for the next-generation Internet protocol, creating new space for Internet development. As a result, the IP technology architecture will shift from network services based on IP addresses to network services based on APN IDs or on both IP addresses and APN IDs.

The network architecture upgrade enabled by APN6 differs from the upgrade from IPv4 to IPv6. A key issue encountered during the IPv4-to-IPv6 upgrade was compatibility. Incompatibility between IPv6 and IPv4 addresses means that Internet infrastructure must be upgraded to support IPv6. This is partly why IPv6 deployment has been slow. The APN6-enabled network architecture upgrade is based on an identification system that consists of IPv6 address and APN ID, thanks to two factors:

First, the IP address system and functions are already well defined, and the address space issue is resolved by the upgrade from IPv4 to IPv6. Therefore, we should shift our efforts from an IP-address-based identification system to an identity-like APN-ID-based identification system, which involves allocation, use, and management of APN IDs, and define various protocol extensions on that basis.

Second, by learning lessons from the IPv4-to-IPv6 upgrade, incremental deployment and upgrade can be performed based on IPv6's backward compatibility mechanism. With an APN ID carried in the IPv6 extension header, an IPv6 node on the network provides services based on the APN ID if the node identifies the APN information in the packet. If the IPv6 node cannot identify the APN information in the packet, it considers the packet to be a common IPv6 packet, and forwards it according to the destination IPv6 address. This allows networks to evolve incrementally.

The combination of IPv6 addresses and APN IDs can be regarded as another expansion of the IP address space. The identifier, which was previously based on a 128-bit IPv6 address, has been lengthened to include more bits — there are now three types of APN IDs: 32-bit, 64-bit, and 128-bit. This means an APN ID used in combination with a 128-bit IPv6 address is equivalent to a 160-bit, 192-bit, or 256-bit identifier.

The network architecture upgrade in question is made possible with a new identifier–APN ID–in addition to the IPv6 address.

APN6 deployment use cases and technical value

An APN6-based video conferencing experience assurance solution has been deployed on an e-Government extranet (Figure 2). This means APN6-based solutions are already commercially viable for live networks.

Figure 2: APN6-based video conferencing experience assurance solution

Traditional video conferencing assurance is difficult as it is highly labor-intensive and networks cannot support targeted assurance services due to a lack of integration between the video system and the network, meaning faults during video conferencing are difficult to locate.

With the APN6-based video conferencing assurance solution, IPv6 packets sent by the video conferencing application carry the application ID corresponding to the video conference that needs assurance. Equipment at the border of the IP transport network can then steer the traffic into a network slice dedicated to the video service in accordance with the application ID. With resource isolation supported by IP network slices, other services will not affect traffic in the video slice, thus guaranteeing good experience for video conferencing users. Furthermore, the IP transport network performs in-situ flow information telemetry (IFIT) for applications based on application IDs to provide application-level, service-quality visualization. This makes it possible to both quickly locate issues and optimize traffic when video conferencing experience is poor.

APN6 can also be applied to scenarios such as computing-network convergence (Figure 3). Latency is critical to good XR service experience. Latency assurance depends not only on network load, but on computing power load on the edge cloud. Therefore, traffic optimization must be performed with consideration for both network load and computing power load.

Figure 3: XR experience assurance through optimization based on both network and computing power loads

A computing-aware traffic steering solution implements comprehensive scheduling based on the network and computing power loads (Figure 4). With this solution, the network must be notified of the location and application identifiers and load information of a computing service. The head node of the network forms multiple path-forwarding rules based on received information like different computing power locations and loads of the same application, points to different computing service locations, and identifies different comprehensive load information relating to network load and computing power load. When receiving new service flow packets for accessing an application, the head node can select a path with a light load, and direct the flow to the corresponding computing service location, thus guaranteeing optimal application access experience.

Figure 4: Computing-aware traffic steering solution

The aforementioned two sample APN6 applications also show the benefit of separating locators from IDs. Traditional Internet packets have only an IP address as the identifier, meaning the IP address serves both as a locator and an ID. In mobile settings, application addresses constantly change, and an IP-address-based identifier system means the address and ID will both change, making it necessary to adjust the IP-address-based traffic steering policy. APN6 introduces an identity-like ID (APN ID) beyond IPv6 addresses, so that the ID remains unchanged when the application address changes. This means an ID-based network policy can still work without adjustment. This mechanism greatly simplifies network O&M.

More agreements reached on standardization

Since early 2019, over a dozen IETF drafts on APN6 technology have been submitted, including the problem statement, framework, IPv6 encapsulation, YANG model, and BGP extensions. APN6 poses challenges to traditional Internet design principles such as the end-to-end principle and the application-network separation principle.

Early on, APN6 caused many debates in the standardization community. But over time, a greater consensus has been reached. APN6 can be preferentially applied to limited domains where both the application and network are controlled, but further development is required for Internet applications in which more third parties are involved.

Regardless, APN6 has made huge progress in terms of both theory and practical applications. The advancement of emerging services like AI computing is leading to a clearer trend in cloud-network-edge-device convergence, which will pave the way for the further development of APN6.

Since the release of ChatGPT in November 2022, large AI model technology has been on a rapid upward trajectory, with model training emerging as the driving force behind technological progress. However, the data volume and diversity of data types involved in model training are increasing exponentially, meaning that data silos pose a major barrier to progress.

Many organizations are struggling to see the way forward.

Huawei's AI data lake solution, however, does not just create bridges between data silos, it lights the way to a more intelligent world and a new era of AI innovation and development.

Trends and challenges of large AI model services

In 2023, Google released the Gemini multimodal model, which is able to understand, operate, and integrate different types of information, including text, code, audio, images, and video. In February 2024, OpenAI released a video model called Sora, which can create realistic scenes from text instructions. By combining a diffusion model with the large language model, Sora exhibits amazing 3D consistency when learning from the physical world.

The speed with which large AI models have developed has far exceeded people's expectations. Witnessing the progression from ChatGPT to Gemini and now to Sora, we have identified two key trends:

• Trend 1: The shift from pure NLP models to multimodal models has brought a corresponding shift in training data. Training datasets now encompass a mix of text, images, audio, and video, rather than text alone. This means that the volume of data required to train today's most advanced models is a staggering 10,000 times greater than was required for earlier models (as shown in Figure 1).
• Trend 2: Computing power, algorithms, and data are the three core elements of large AI models. Through the stacking of computing power, aggregation of data, and parameter scaling (from hundreds of billions to trillions, and even up to 10 trillion), complex behaviors emerge within the framework of deep learning algorithms. For example, in a video released by Sora, we witness a stylish woman strolling down a street. As she moves, the perspective on the street scenes behind her seamlessly adjusts. As she walks by, objects and other pedestrians are momentarily obscured, and then reemerge, maintaining impeccable 3D consistency and object permanence, faithfully replicating how the world is perceived through human eyes.

Figure 1: Multimodal models for exponential growth in the volume of training data

The transition from unimodal to multimodal models will create a deluge of data, and a solution is urgently required that can integrate scattered data resources. As a way of presenting the real world, data serves as the cornerstone of training large AI models. With deep learning algorithms, the quantity and quality of data determines how far AI training will go. However, most data owners prioritize the efficient access of data by service applications, often overlooking where data is stored. Similarly, most data administrators focus on whether data is effectively stored, and may neglect data ownership and types. As a result, data ends up siloed in different data centers. One carrier we spoke to has seen the total amount of data they need to store soar to hundreds of PBs in recent years, and hundreds of TBs of new data is generated each day. However, this data is scattered across multiple data centers.

To provide as much data as possible for model training, the carrier's technical department has to migrate or replicate the data from scattered data silos across regions. As a result, the data preparation process is prolonged, accounting for over 50% of the entire model training duration.

Therefore, the biggest challenges and primary considerations for constructing large AI model infrastructure revolve around eliminating data silos and effectively gathering data in one place, swiftly transforming collected datasets into model training materials, and ensuring that AI computing power has efficient access to training materials.

New requirements for data storage and management

The ideal AI data infrastructure should be able to support the key phases of large AI model training, such as data ingestion, data preprocessing, and model training, to provide high-quality data services. To achieve this goal, at least two layers of data infrastructure should be considered: the storage device layer and the data management layer.

Storage device layer

When dealing with a large amount of heterogeneous data from multiple sources, especially in multimodal AI training scenarios, the ideal storage device layer should feature multi-protocol interworking, high I/O performance, and easy scalability to address various challenges and support the following pivotal phases of training large AI models:

• Data ingestion: In this phase, data is often scattered across different silos, stored in various formats, and accessed through different protocols. To efficiently centralize heterogeneous data from different sources, data storage devices need to support various data formats and access protocols, and deliver high write bandwidth. Additionally, storage solutions need to be both flexible and scalable, while also remaining cost-effective, in order to handle new data sources that may be incorporated into training at any time. While data formats and access protocols may vary during the data ingestion phase, servers only access data through the file interfaces during the training phase. Therefore, ideal storage hardware should support multi-protocol interworking to ensure that underlying data can be accessed by different protocols or interfaces, thereby eliminating the data replication caused by protocol conversion.
• Data preprocessing: Data preprocessinginvolves the cleansing, transformation, augmentation, and standardization of diverse data to yield high-quality training materials from massive volumes of raw data. In this phase, a large amount of temporary data is generated due to the diversity of preprocessing tools, resulting in data expansion. Therefore, storage devices should offer extensive shared storage capacity while also delivering high read/write bandwidth and supporting random access to expedite preprocessing.
• Large AI model training: In the model training phase, the performance of storage devices in tasks such as training data loading and checkpointing directly affects training efficiency. Although the volume of pre-processed training data is small, high file access performance (OPS and IOPS) and low latency are required to ensure quick data loading and avoid wasting GPU/NPU computing power. Saving checkpoints during the training process is critical to ensure that, in the event of an interruption, training can be resumed rather than started from scratch. Therefore, it requires high write bandwidth for fast and frequent archiving and more stable training.

Data management layer

Based on the flexible capacity expansion and high hybrid-load performance provided by the storage device layer, the data management layer further provides advanced data management capabilities for AI training. It helps data owners and administrators maximize data value in a more efficient manner from three dimensions: visibility, manageability, and availability.

• Visibility: Data asset owners and administrators should have a comprehensive understanding of their data, including data storage location, data volume, and data types, akin to having a data map. Considered from this dimension, owners and administrators can quickly identify which data needs to be collected based on this map.
• Manageability: After determining the data to be collected, a mechanism is required to implement policy-based data flow. For example, a policy may be used to define the source and target of data flow, start and end time window, maximum rate limitation, and minimum rate guarantee, facilitating the management of data.
• Availability: Raw data needs to be preprocessed and converted into training data. While there are numerous tools for data preprocessing, it is still important to have a data preprocessing framework that works in synergy with the storage device layer. Such a framework should help users streamline data preprocessing and enhance processing speed to improve data availability.

Core capabilities of AI data infrastructure

The ideal AI data infrastructure should have the core capabilities outlined in Figure 2.

Figure 2: Core capabilities of AI data infrastructure

To sum up, three key features are indispensable:

• High performance

  High-performance data infrastructure underpins and accelerates every phase of large AI model training, from data ingestion and preprocessing in the early stage to training data loading and periodic checkpointing in the model training phase. High performance in this context is comprehensive and includes both high OPS and low latency for random access and high read/write bandwidth for sequential access.

• High capacity

  The data ingestion and preprocessing phases involve the intensive reading, writing, and generation of temporary data, which creates storage challenges. Data infrastructure therefore needs to offer flexible and non-disruptive scale-out and tiering capabilities. This ensures a balance between cost and capacity, while delivering the necessary bandwidth, IOPS, and OPS to handle sequential and random data access patterns.

• Exceptional ease-of-use

  Data infrastructure should feature global data management, efficient data mobility, AI platform and toolset collaboration, and optimization and enhancement tailored for large AI model training. This would accelerate AI training and learning by ensuring that data is visible, manageable, and available.

AI data lake solution

Huawei has been actively engaging in large AI model training with customers in diverse industries, including carriers. Over the years, we have gained extensive expertise in AI data infrastructure. Recently, we have launched an AI data lake solution to help customers solve the problems encountered in the deployment of data infrastructure for large AI model training. This enables our customers to focus on model development and training. The architecture of the Huawei AI data lake solution, as shown in Figure 3, consists of three layers: a data storage layer, a data fabric layer, and a data service layer.

Data storage layer

In this layer, data is stored in different data centers.

In each data center, data is intelligently stored in the hot and warm tiers. The hot tier is powered by the OceanStor A series, which is Huawei's high-performance storage specially designed for large AI model training. It can scale out to thousands of nodes. The warm tier uses Huawei's OceanStor Pacific scale-out storage to handle mass unstructured data. Intelligent tiering can be implemented between the OceanStor A series and the OceanStor Pacific series. That is, in the same storage cluster, the OceanStor A series nodes form a high-performance storage tier, and the OceanStor Pacific series nodes form a high-capacity storage tier. The two tiers are combined to externally present a unified file system or object bucket that supports multi-protocol interworking (one copy of data can be accessed through multiple protocols). Internally, data is intelligently and automatically tiered to achieve an optimal balance between capacity, performance, and cost.

A data replication relationship can be created between storage clusters across data centers to facilitate reliable and on-demand data mobility between data centers. This ensures the data device layer is fully prepared to support data ingestion for large AI model training.

Data fabric layer

The data fabric layer creates a seamlessly interconnected network to facilitate data mobility. It makes data visible, manageable, and available, helping large AI model training derive value from data.

Huawei uses a software layer powered by Omni-Dataverse to achieve data visibility, manageability, and availability. Omni-Dataverse is an important component of Huawei's Data Management Engine (DME). It provides the unified management of metadata on Huawei storage devices across different data centers to form a global data asset view. Omni-Dataverse also invokes interfaces on storage devices to control data movement. Omni-Dataverse executes actions based on user-defined policies, and can implement GPU/NPU-storage passthrough and intelligent file prefetching on demand to ensure training data is readily available, minimizing computing power wait times.

This makes data ingestion and model training much more efficient, improving cluster utilization.

Data service layer

The Huawei AI data lake solution provides commonly used service frameworks at the data service layer, including data preprocessing, model development, and application development frameworks.

The data preprocessing framework performs data cleansing, conversion, augmentation, standardization, and other preprocessing operations. Large AI model customers can integrate their algorithms and functions into this framework for simplified preprocessing management. Furthermore, customers can choose an alternative framework.

Similar to data preprocessing, model development and application development are the other two frameworks provided by Huawei for customers' convenience, which they can select on demand.

The Huawei AI data lake solution is built on years of experience and expertise in large AI model training. It helps enterprises eliminate data silos, achieve smooth data mobility, and implement a data fabric between data applications and storage devices, making data visible, manageable, and available. As large AI models continue to evolve from unimodal to multimodal, the ever-increasing volume and diversity of data will inevitably lead to a non-linear increase in management complexity and performance requirements. An AI data lake solution running on a three-layer architecture can effortlessly address these challenges to facilitate the development of large AI models, accelerate the emergence of intelligence in model training, and push AI innovation and development to new heights.

Figure 3: Architecture of Huawei AI data lake solution