FlexE拥抱5G承载（英文版）.pdf

Independent market research and competitive analysis of next-generation business and technology solutions for service providers and vendors Meeting 5G Transport Requirements With FlexE A Heavy Reading white paper produced for ZTE AUTHOR: STERLING PERRIN, PRINCIPAL ANALYST, HEAVY READINGHEAVY READING | ZTE | MEETING 5G TRANSPORT REQUIREMENTS WITH FLEXE 2 INTRODUCTION Work to standardize 5G New Radio continues, and leading-edge operators are in a race to be first movers. Scale and performance requirements dictated by the 5G radio network are forcing a radical rethink of access and aggregation transport networks, and many operators report that transport networks must be built and upgraded one to two years before wide-scale launches. Under tremendous pressure to be first to market, the stakes are high for transport teams that are tasked to produce future-proof architectures. Meeting ultra-reliability and ultra-low latency requirements, scaling capacity and economically addressing the diverse use-case requirements are the biggest challenges for 5G transport. Under the banner of xhaul, there are many technology options available, and combined with the functional split variations defined in 5G, the situation is complex and largely undecided. This white paper focuses on Flex Ethernet (FlexE), an emerging transport technology choice that shows promise for 5G. The paper provides an overview of FlexE technology and describes how its attributes address the high-reliability, low-latency, capacity and economics challenges posed by 5G. Key to the value proposition are extensions to FlexE that evolve it from a point- to-point to a fully networked technology. 5G NETWORK OVERVIEW The 5G New Radio draft specifications for the first phase were approved in December 2017, and work on the second phase (officially, Release 16) is expected to be finalized in December 2019. Even in the absence of full standardization, network operators and governments around the world are clamoring to attain rights to leadership status in 5G. Among the most ambitious countries are South Korea, Japan, China and the U.S. Early commercial launches are set to begin as soon as this year, though the bulk of commercial activity must wait until after full standardization, with 2020 looking to be a key year for big launches if the industry stays on its current plan. Like other mobile generations, 5G promises greater capacity to end devices. It also promises a tenfold increase in capacity, up to 1 Gbit/s to end devices, as a practical (not theoretical) data rate. High capacity is a hallmark of the enhanced mobile broadband (eMBB) use case, which is just one of three primary use cases around which 5G is being defined and built. Mas- sive machine-type communications (mMTC) describes Internet of Things (IoT) applications in which data rates to individual sensors can be very low (measured in kbit/s) but connected devices number in the billions. The third major use case, ultra-reliable low-latency communi- cations (URLLC), describes mission-critical and extreme-precision applications in which end- to-end latency may be 1 ms, jitter less than 1 s and reliability measured to six nines. The diversity of use cases for 5G combined with the stringent and extreme attributes across the use cases, as described above, has spurred the development of new radio access net- work (RAN) architectures. While 4G was dominated by a distributed RAN architecture, new fronthaul and midhaul RANs are added to the mix for 5G. We define them briefly below: Distributed (or classical) RAN: All cellular traffic is backhauled to the mobile switching center and switched out to baseband units (BBUs)/cell towers as needed. There is some separation of the remote radio head (RRH) and the BBU functionality, but only along the length of the cell tower (top to bottom).HEAVY READING | ZTE | MEETING 5G TRANSPORT REQUIREMENTS WITH FLEXE 3 Centralized RAN: BBU functionality is separated from the RRH units and pool at central hub sites, or BBU pools. As a result, the BBU process can be shared among many RRUs for greater coordination and bandwidth efficiency. In its fully centralized form, separation of RRH and BBU is limited to 15 km, due to restrictions imposed by coordination between radios and BBU processing. However, hybrid architectures are emerging in which a portion of BBU functionality resides at intermediate points between radio and the central hub site, thus easing some of the distance and timing burden. (We discuss functional splits and variations in more detail below.) Cloud or virtualized RAN: Cloud RAN describes the introduction of NFV in the RAN network, such that compute and storage can move closer and further from individual users across the network dynamically, based on specific cell site requirements that change over time. Greater agility and adaptability are the key benefits that cloud/ virtualization bring to the RAN. Defining cloud RAN cannot rely solely on the 3GPP, however, as virtualization in telecom is a broad trend, of which 5G is just a subset. 5G Functional Segmentation Historically in the RAN, Layer 1-3 processing functions resided within a distributed BBU. While 4G architectures provided some separation between the RRH and the BBU, that separation was limited to the height of a cell tower, at the base of which sat the BBU. RAN architectures become more flexible and complex with 5G. Known as NG-RAN, the 3GPP 5G RAN archi- tecture introduces new functional modules that place new demands on the transport network, which is tasked with providing scalable and economical connectivity between them. In 5G, radio base station (gNB) functionality is split into three functional modules: the central- ized unit (CU), the distributed unit (DU) and the radio unit (RU), which can be deployed in multiple combinations. By centralizing BBU functions, operators can share network resources and tightly coordinate radio activity to improve network performance. On the other hand, by distributing Layer 2 functionality (either to the cell tower or to an intermediate location in the RAN) aggregation and statistical multiplexing can be used to reduce (potentially greatly) transport network capacity and costs. Figure 1 provides an overview of the different func- tional options currently in play for 5G. Figure 1: 5G RAN Functional Unit Variations Source: NGMN, 2018HEAVY READING | ZTE | MEETING 5G TRANSPORT REQUIREMENTS WITH FLEXE 4 The separation of BBU functionality across distances has created two new RAN transport segments: the fronthaul and the midhaul network. The fronthaul network connects the RU to the DU across a distance (i.e., when the two are not colocated at the cell site). The Common Public Radio Interface (CPRI) Consortium built a new protocol specifically to ad- dress new fronthaul transport demands. Called eCPRI, the new specification introduces flexibility of functional splits to consistent with NG-RAN and adapts CPRI for packet-based transport to reduce transport capacity and costs. One challenge with fronthaul is that RU to DU connections are highly sensitive to latency. The midhaul network connects the DU to the CU across a distance, using a higher-layer functional split that is more tolerant to delay. Connecting at Layer 2, it always allows for statistical multiplexing and aggregation to reduce costs. As shown in Figure 1, the DU may be located at the cell site or at an intermediate RAN location. These two new segments complement backhaul, which remains in the 5G architecture and defines connectivity after the CU. TRANSPORT REQUIREMENTS FOR 5G Early on, operators realized that the transport network will play an essential role in delivering many of the attributes promised by 5G, including high bandwidth, low latency, greater den- sity, IoT connectivity, reliability, coverage and costs. Significant challenges exist particularly in the access network, including fronthaul, midhaul and backhaul. Through operator surveys and one-on-one discussions, Heavy Reading has identified the following as the top transport challenges for operators building for 5G. High Bandwidth Requirements While 5G is not solely about greater capacity, it clearly plays an important role in the migra- tion from 4G networks to 5G, and Heavy Reading operator research shows that greater capacity will likely be the primary initial driver for this migration. 5G promises an order of magnitude greater capacity delivered to end devices (up to 1 Gbit/s to devices, as noted earlier). To be clear, the order of magnitude capacity mandate applies specifically to the radio network, but the increase in radio network capacity has ramifications throughout the entire wireline network that supports it. In fact, many operators are planning to increase their RAN capacities at least tenfold to accommodate the coming 5G traffic. Whereas current 4G cell sites are typically served by 1 Gbit/s backhaul, operators see a move to 10-100 Gbit/s xhaul, with variations in between, depending on the specific protocols used and xhaul segment whether its fronthaul, midhaul or backhaul connectivity. A metro core which converges mobile, residential broadband and enterprise traffic may require 400+ Gbit/s capacity in the not-too-distant future. Densification the expansion of wireline connectivity to new cell sites is also one of the mandates of 5G, which calls for 10 to 100 times the number of connected devices compared to 4G, as well as up to 1,000 times the bandwidth per unit area. Densification, which comes primarily in the form of new small cells, will be another major driver of wireline bandwidth. Aggregation points must be able to accommodate ten times the number of users.HEAVY READING | ZTE | MEETING 5G TRANSPORT REQUIREMENTS WITH FLEXE 5 Ultra-Low Latency Requirements In the 5G context, ultra-low latency has two aspects that must be addressed through the transport network. One is a user/application-driven latency requirement; the other is a RAN- imposed latency requirement. Well address both here. Application-Imposed Ultra-Low Latency: Ultra-low latency is one of the three key use-case pillars on which 5G is being defined (as part of the ultra-reliability and low-latency use case). Ultra-low latency applications are those in which the end-to- end latency demands imposed by the applications themselves are far more stringent than previous mobile applications. Examples include self-driving cars that rely on network connectivity, virtual/augmented reality and industrial automation. For some industrial automation applications, for example, end-to-end latency may be limited to 1 ms and jitter held to just 1 s. Network-Imposed Ultra-Low Latency: The network-imposed latency challenge is determined by communications requirements between the RUs and the higher-layer processing functions within the BBU. This was a non-issue in classical distributed RAN architectures, because all BBU processing occurred at the cell tower itself. However, with 5Gs proposed functional separation (Figure 1), limitations of latency and dis- tance become key factors. RU/DU separation must be held to 125 s round trip time (and 15 km maximum distance, based on the physical limits of light through fiber), imposing a severe limitation, regardless of the application traffic riding on top. Diverse Application Requirements The requirement to serve three diverse use cases poses a unique challenge to 5G compared to all previous mobile technology generations. Building separate transport networks that address each distinct use case is not economically viable. The challenge for operators is how to address the diversity of use cases reliably and efficiently meeting the specifications for each while sharing a single transport network. The proposed solution for this 5G problem is network slicing. Network slicing allows multiple logical networks to run on top of a shared physical network infrastructure. Each logical net- work, or slice, carries its own performance attributes, quality of service (QoS), service-level agreements (SLAs) and share of network resources end to end, thus enabling an operator to deliver an IoT sensor connectivity, gigabit broadband and vehicular communications with the same physical network. Reliability Requirements While building and partitioning to address all the requirements above, reliability is paramount. In general terms, operators must provide reliable services in what will be a highly competi- tive market environment, and many operators are positioning five nines reliability as the network benchmark. In addition, ultra-reliability is specifically called out in one of the three main 5G use cases, URLLC, in which ultra-reliability may or may not be combined with ultra-low latency. Ultra- reliable applications can include industrial automation applications, in which a short downtime could cost an enterprise millions of dollars, or vehicular communications, in which eliminating downtime can be a matter of life and death. Ultra-reliable applications will require six-nines (99.9999 percent) uptime.HEAVY READING | ZTE | MEETING 5G TRANSPORT REQUIREMENTS WITH FLEXE 6 FLEX ETHERNET (FLEXE)-BASED SOLUTION Defined by the OIF, Flex Ethernet is an interface that supports various Ethernet media access controller (MAC) rates that dont correspond to existing Ethernet physical layer (PHY) rates. Supported FlexE client rates can be higher than the Ethernet PHY rate (through bonding) or lower than the PHY rate (through sub-rate and channelization). The Flex Ethernet Implemen- tation Agreement 1.0 was published in March 2016, with a focus on 100 GE bonding. Version 2.0 extends bonding up to 200GE rates and 400GE rates, among other added features, and was published in June 2018. From a technical perspective, FlexE introduces three features to standardized IEEE Ethernet: Bonding: Bonding is a common technique in networking whereby two or more links are combined greater throughout on the link. In FlexE, a high 200 Gbit/s Ethernet client transmission can be supported by combining (bonding) two 100 Gbit/s Ethernet PHYs. The initial implementation agreement (IA) specified bonding for nx100 Gbit/s Ethernet PHYs; the 2.0 version extends bonding to 200 Gbit/s and 400 Gbit/s PHYs. The IA specifies a maximum of 254 FlexE groups for 100GE PHY, 128 for 200GE PHY and 64 for 400GE PHY. Sub-rate: With sub-rating, FlexE maps lower-rate Ethernet MACs (including non- standard rates) into higher-rate PHYs. For example, a 50 Gbit/s MAC could be trans- ported over a 100 Gbit/s PHY. Bonding and sub-rate can be combined for hybrid client rates, such as a 250 Gbit/s MAC over 3x100 Gbit/s bonded PHYs. Channelization: Channelization is another common networking technique in which the available bandwidth in a link is shared among multiple clients. In FlexE, channel- ization can be defined within a single PHY or across multiple bonded PHYs. Channeliza- tion granularities are specified by the FlexE client definitions