SkyOctopus: Enabling Low-Latency Mobile Satellite Network through Multiple Anchors

Abstract

The rapid deployment of low earth orbit (LEO) satellite constellations has drawn attention to the potential of non-terrestrial networks (NTN) in providing global communication services. Telecom operators are attempting to collaborate with satellite network providers to develop mobile satellite networks, which serve as an effective supplement to terrestrial networks. However, current mobile satellite network architectures still employ the single-anchor design of terrestrial mobile networks, leading to severely circuitous routing for users and significantly impacting their service experience. To reduce unnecessary latency caused by circuitous routing and provide users with low-latency global internet services, this paper presents SkyOctopus, an advanced multi-anchor mobile satellite network architecture. SkyOctopus innovatively deploys traffic classifiers on satellites to enable connections between users and multiple anchor points distributed globally. It guarantees optimal anchor point selection for each user’s target server by monitoring multiple end-toend paths. We build a prototype of SkyOctopus using enhanced Open5GS and UERANSIM, which is driven by actual LEO satellite constellations such as Starlink, Kuiper, and OneWeb. We conduct extensive experiments, and the results demonstrate that, compared to standard 5G NTN and two other existing schemes, SkyOctopus can reduce end-to-end latency by up to 53%.

Index Terms—Mobile Satellite Network, UPF, 6G, LEO

低地球轨道（LEO）卫星星座的快速部署已引起人们对非地面网络（NTN）在全球通信服务中潜力的关注。电信运营商正尝试与卫星网络供应商合作，发展作为地面网络有效补充的移动卫星网络。然而， 当前的移动卫星网络架构仍沿用地面移动网络中的单锚点（single-anchor）设计，这导致了用户流量的路由严重迂回 ，并显著影响了其服务体验。为减少因路由迂回造成的非必要延迟，并为用户提供低延迟的全球互联网服务，本文提出了一种名为 SkyOctopus 的先进多锚点（multi-anchor）移动卫星网络架构。SkyOctopus 创新性地在卫星上部署流量分类器，以实现用户与全球分布的多个锚点之间的连接。 它通过监测多条端到端路径，保证为每个用户的目标服务器选择最优的锚点。我们利用增强的 Open5GS 和 UERANSIM 构建了 SkyOctopus 的原型系统，并由 Starlink、Kuiper 和 OneWeb 等真实 LEO 卫星星座数据驱动。我们进行了大量的实验，结果表明，与标准的 5G NTN 及其他两种现有方案相比，SkyOctopus 能够将端到端延迟最高降低 53%

Introduction

Nowadays, we are witnessing the rapid development of low earth orbit (LEO) satellite constellations, such as SpaceX’s Starlink [1], Amazon’s Kuiper [2], and OneWeb [3]. These satellite constellations, with their dense satellite distribution and inter-satellite links (ISLs), provide global internet and communication services while complementing terrestrial mobile networks in underserved areas in a cost-effective manner.

如今，我们正见证着低地球轨道（LEO）卫星星座的迅猛发展，例如SpaceX的Starlink [1]、亚马逊的Kuiper [2]和OneWeb [3]。这些卫星星座凭借其密集的卫星分布和星间链路（ISL），在为服务欠缺地区提供具有成本效益的地面移动网络补充的同时，也提供了全球性的互联网与通信服务。

Meanwhile, as the developer of 5G, 3GPP has explicitly stated that non-terrestrial networks (NTN) will be an essential component of future 5G-Advanced and 6G networks [4][6]. In practice, it has become a trend for telecom operators and satellite network providers to collaborate to build mobile satellite networks, as exemplified by partnerships such as TMobile with SpaceX [7] and AT&T with AST [8].

与此同时，作为5G的制定者，3GPP已明确指出非地面网络（NTN）将是未来5G-Advanced和6G网络的重要组成部分[4]-[6]。在实践中，电信运营商与卫星网络供应商合作共建移动卫星网络已成为一种趋势，T-Mobile与SpaceX [7]以及AT&T与AST [8]的合作即是例证。

To ensure compatibility with existing terrestrial mobile networks, mobile satellite networks largely adopt the design of terrestrial mobile networks, moving only the access network to the satellite [5]. Since the core network remains deployed on the ground and unchanged, each protocol data unit (PDU) session corresponds to a specific anchor point. Given the global random access patterns of users, this single-anchor design poses significant challenges for the user data plane in satellite mobile networks. In such cases, user traffic transmission needs to traverse a fixed ground anchor point, resulting in circuitous routing and increased latency.

为确保与现有地面移动网络的兼容性，移动卫星网络在很大程度上沿用了地面网络的设计，仅将接入网部分移至卫星上[5]。由于核心网依旧部署于地面且保持不变，每个协议数据单元（PDU）会话都对应一个特定的锚点。考虑到用户全球随机接入的模式，这种单锚点设计为卫星移动网络中的用户数据平面带来了巨大挑战。在此情况下，用户流量的传输需经过一个固定的地面锚点，从而导致路由迂回并增加延迟。

解释一下

1. "由于核心网依旧部署于地面且保持不变，每个协议数据单元（PDU）会话都对应一个特定的锚点" 什么意思
2. "用户流量的传输需经过一个固定的地面锚点，从而导致路由迂回并增加延迟" 什么意思

(1) PDU:

PDU会话 (Protocol Data Unit Session)：在移动网络系统中, 手机就会向网络发起一个建立PDU会话的请求。一旦会话建立成功，您的手机就获得了一个IP地址，可以开始访问互联网了

(2) CoreNet and UPF:

• 核心网是移动通信网络的大脑和中枢。
• 在5G核心网中，有一个非常重要的网元叫做UPF: UPF是“专属数据通道”通向外部互联网的总出口或网关
• UPF 即为所谓的 锚点 (Anchor Point)

(3) 为什么固定的地面锚点会导致“路由迂回”？

一个身在新加坡的用户，打开了他的星链手机。手机通过头顶的卫星连接到了5G网络。根据网络当时的策略，系统为他分配了一个位于日本东京数据中心的UPF作为他这次PDU会话的锚点。从这一刻起，他在本次上网期间的所有数据都必须经过东京的这个锚点

现在，这个新加坡用户想要访问一个也位于新加坡的CDN Server

• 用户的实际访问数据: 新加坡的手机出发 -> 上行到头顶的LEO卫星 -> 卫星通过星间链路把数据传到离东京最近的地面站 -> 数据通过光纤进入东京的UPF（锚点） -> 从东京的UPF进入公共互联网 -> 数据再从互联网绕一大圈回到位于新加坡的 CDN Server
• 理想路径: 新加坡的手机 -> 卫星 -> 新加坡的地面站 -> 新加坡的CDN Server

A natural solution to this issue is to deploy the anchor point onto the satellite nearest to the base station. By deploying anchor points on satellites and minimizing the distance between mobile network infrastructures, this approach aims to mitigate the long end-to-end latency caused by circuitous routing. However, considering the changes in network topology caused by the high-speed movement of satellites, users would face severe anchor point reselection issue, significantly impacting service continuity and incurring substantial reselection costs. Therefore, this solution is difficult to widely apply in realworld scenarios.

一个针对此问题的自然解决方案是将锚点部署到离基站最近的卫星上。通过在卫星上部署锚点并最小化移动网络基础设施间的距离，该方法旨在减轻由路由迂回所引发的长端到端延迟。然而，考虑到卫星高速移动所带来的网络拓扑变化，用户将面临严峻的锚点重选问题，这会严重影响服务连续性并产生高昂的重选成本。因此，该方案难以在真实世界场景中被广泛应用。

In this paper, we present SkyOctopus, an advanced multi-anchor mobile satellite network architecture. SkyOctopus supports the simultaneous existence of multiple anchor points within a single PDU session by using traffic classifiers deployed on satellites. It also employs a fine-grained selection strategy, which uses location-based criteria for the initial selection of anchor points and updates anchor point choices based on network conditions through continuous monitoring. Additionally, based on the correspondence between base stations and traffic classifiers, and parallelized signaling transmission, we design a new PDU session establishment process for SkyOctopus. Users can quickly establish PDU sessions without concern for the number of anchor points.

We construct a prototype of SkyOctopus using enhanced Open5GS [9] and UERANSIM [10], driven by real LEO satellite ephemerides, including Starlink, Kuiper and OneWeb. Based on this prototype, we conducted extensive experiments, and the results indicate that SkyOctopus significantly reduces end-to-end latency by up to 53% compared to the other three schemes and reduces session establishment time by 86%.

在本文中，我们提出了 SkyOctopus，一个先进的多锚点移动卫星网络架构:

SkyOctopus 通过部署在卫星上的流量分类器，支持在单个PDU会话中同时存在多个锚点 。它还采用了一种细粒度的选择策略，该策略利用基于位置的标准进行锚点的初始选择，并通过持续监控根据网络状况更新锚点的选择。此外，基于基站与流量分类器之间的对应关系以及并行化的信令传输，我们为 SkyOctopus 设计了全新的PDU会话建立流程。用户可以快速建立PDU会话，而无需关心锚点的数量

我们利用增强的 Open5GS [9] 和 UERANSIM [10] 构建了 SkyOctopus 的原型系统，并由包括 Starlink、Kuiper 和 OneWeb 在内的真实LEO卫星星历数据驱动。基于此原型，我们进行了大量的实验，结果表明，与其他三种方案相比，SkyOctopus 将端到端延迟最高降低了 53%，并将会话建立时间减少了 86%

Contributions of this paper can be summarized as follows:

• We for the first time expose the issue of high end-to-end latency caused by circuitous routing in emerging mobile satellite networks, which is essentially due to the design of single-anchor PDU session.

• We propose SkyOctopus, an advanced mobile satellite network architecture that achieves low-latency global internet services through multiple anchor points and fine-grained anchor point selection strategy.

• We construct a prototype of SkyOctopus and conduct comprehensive experiments to demonstrate its effectiveness in reducing end-to-end latency and its efficiency in terms of session establishment time.

本文的贡献可总结如下:

• 我们首次揭示了新兴移动卫星网络中，由单锚点PDU会话设计本质上导致的路由迂回所引发的高端到端延迟问题
• 我们提出了 SkyOctopus，一个通过多锚点和细粒度锚点选择策略实现低延迟全球互联网服务的先进移动卫星网络架构
• 我们构建了 SkyOctopus 的原型系统，并进行了全面的实验，以证明其在降低端到端延迟方面的有效性以及在会话建立时间上的高效性

The rest of this paper is structured as follows. Section II introduces the background of the problem and our motivation. Section III presents an overview of the proposed SkyOctopus architecture. Section IV provides detailed explanations of the three aspects of SkyOctopus. We conduct extensive experiments and analyze the results in section V. Section VI presents a review of related work in the field. Section VII discusses additional considerations and issues related to our work. Finally, Section VIII briefly concludes this work.

本文的其余部分结构如下。第二节介绍问题背景与我们的动机。第三节概述所提出的 SkyOctopus 架构。第四节对 SkyOctopus 的三个方面进行详细阐述。我们在第五节中进行大量实验并分析结果。第六节回顾该领域的相关工作。第七节讨论与我们工作相关的额外考量和问题。最后，第八节对本文进行简要总结。

Background

A. PDU Session in Mobile Networks

In the 5G network, a PDU session refers to the logical channel between a user and a data network (e.g. Internet) through a specific base station and user plane function (UPF). The UPF manages the user session context and fixes the transmission path of user traffic, thus it is referred to as the anchor point of the PDU session in mobile networks. As the sole network function in the core network that handles user traffic, the UPF is responsible for executing all user plane policies according to packet detection rules (PDRs).

在5G网络中，PDU会话指的是用户通过特定的基站和用户平面功能（UPF）与数据网络（如互联网）之间建立的逻辑信道。UPF负责管理用户会话的上下文信息并固定用户流量的传输路径，因此它在移动网络中被称为PDU会话的锚点（anchor point）。作为核心网中唯一处理用户流量的网络功能，UPF负责根据分组检测规则（PDRs）来执行所有的用户平面策略。

Specifically, Fig. 1 illustrates the process by which the UPF handles incoming user traffic [11]. Taking the uplink direction as an example, upon receiving the user packet from the base station (BS), the UPF identifies the specified packet forwarding control protocol (PFCP) session to which the packet corresponds. Then, it selects the highest precedence PDR within the matching PDR of the PFCP session. Next, the UPF processes the data packet based on the associated rules specified by the selected PDR, including forwarding action rules (FARs), buffering action rules (BARs), QoS enforcement rules (QERs), and usage reporting rules (URRs). Finally, the packet is forwarded, with its direction determined by the matching FAR of the selected PDR.

具体而言，图1展示了UPF处理上行用户流量的过程[11]。以上行方向为例，当从基站（BS）接收到用户数据包后，UPF首先识别出该数据包对应的特定分组转发控制协议（PFCP）会话。然后，它在该PFCP会话匹配的PDR中选择一个优先级最高的PDR。接下来，UPF根据所选PDR中指定的关联规则来处理该数据包，这些规则包括转发行为规则（FARs）、缓冲行为规则（BARs）、QoS执行规则（QERs）和使用报告规则（URRs）。最后，根据所选PDR匹配的FAR来决定数据包的转发方向并将其发送出去。

3GPP has defined an intermediate UPF (I-UPF) that does not serve as an anchor point but is deployed as a traffic classifier between the base station and multiple UPFs [12]. It achieves traffic classification by using different PDRs to match packets with various target IP addresses or data network names. These PDRs are often associated with different FARs, which forward the packets to different UPFs. In terrestrial networks, the I-UPF is primarily used in private networks requiring high reliability (such as vehicular networks and smart factories) or multi-access edge computing (MEC) services to classify private and public network traffic.

3GPP定义了一种中间UPF（I-UPF），它不作为锚点，而是作为流量分类器部署在基站和多个UPF之间[12]。它通过使用不同的PDR来匹配具有不同目标IP地址或数据网络名称的数据包，从而实现流量分类。这些PDR通常与不同的FAR相关联，而这些FAR会将数据包转发至不同的UPF。在地面网络中，I-UPF主要用于需要高可靠性的专用网络（如车联网和智慧工厂）或多路访问边缘计算（MEC）服务中，以对私有网络和公共网络的流量进行分类。

PDU session with an I-UPF is established using an insertion-based process [13]. For example, when a user moves into the service range of a specific MEC service, the core network inserts an I-UPF into the user’s current PDU session and instructs the I-UPF to establish a connection with the additional UPF associated with the MEC service. When the user leaves the MEC service range, the I-UPF and the additional UPF are removed.

带有I-UPF的PDU会话是使用一种基于插入的流程来建立的[13]。例如，当一个用户移动到某个特定MEC服务的服务范围内时，核心网会将一个I-UPF插入到该用户当前的PDU会话中，并指示该I-UPF与和此MEC服务关联的附加UPF建立连接。当用户离开该MEC服务范围时，这个I-UPF和附加的UPF则会被移除。

关于 I-UPF 的背景知识

I-UPF 这个“基于插入的流程” 的精髓在于它的灵活性和无感化

1. 不会中断原有的PDU会话，而是在需要时，动态地、像“插件”一样把一个智能分流点（I-UPF）“插入”到现有的数据路径中
2. 只有当您进入特定服务区域时，这个“分流员”才会上岗
3. 可以在不改变主要上网通道的前提下，为访问那些近在咫尺的边缘服务（MEC）提供一条宝贵的“捷径”，从而实现超低延迟

如图:

B. Mobile Satellite Networks

3GPP has defined two architectures for mobile satellite networks: the transparent mode (also known as bent-pipe) and the regenerative mode. In the transparent mode, the satellite acts as a transparent relay node between the user and the ground base station, whereas in the regenerative mode, base stations are deployed on the satellite, and user traffic can be forwarded through ISLs.

3GPP为移动卫星网络定义了两种架构：透明转发模式（也称作“弯管”模式，bent-pipe）和再生模式（regenerative mode）。在透明转发模式中，卫星扮演用户与地面基站之间的透明中继节点；而在再生模式中，基站被部署在卫星上，用户流量可以通过星间链路（ISL）进行转发。

However, regardless of whether base stations are deployed on satellites, user traffic must be sent to the specific ground-based anchor point before being forwarded to the target server. Considering that users’ access targets are randomly distributed globally, the anchor point often deviates from the path to the user’s access server, causing significant detours.

然而， 无论基站是否部署在卫星上，用户流量都必须被发送至一个特定的、位于地面的锚点，然后才能被转发至目标服务器 。考虑到用户的访问目标在全球范围内随机分布，该锚点的位置常常偏离用户访问服务器的最优路径，从而导致显著的路由绕行。

Fig. 2 plots a typical example of circuitous routing in a mobile satellite network. For the current PDU session of UE U, the UPF at point F serves as its anchor point. User traffic must first pass through the satellite network to reach the anchor point, and then proceed through the ground network to reach the server. Consider a scenario where the user accesses server S, then the user traffic follows the path U − A − B − C D − F − S. It is evident that this path is not the fastest path in the network. In fact, U −A−B −E −S is a more optimal path, which could significantly reduce latency.

图2描绘了一个移动卫星网络中路由迂回的典型例子。对于用户设备U当前的PDU会话，位于F点的UPF是其锚点。用户流量必须首先经过卫星网络到达该锚点，然后再通过地面网络到达服务器。假设用户访问服务器S，那么用户流量遵循的路径是 U-A-B-C-D-F-S。显而易见，这条路径并非网络中的最快路径。实际上，U-A-B-E-S 是一条更优的路径，它能够显著降低延迟。

As a specific example, consider the scenario of a user located in the Atlantic Ocean (42.2 ◦ N, 60.0 ◦ W) accessing a server in Paris through different paths within the Starlink constellation, as shown in Table I. When establishing a PDU session, the user selects the nearest ground station (GS) located in Ashburn, USA. Consequently, the user’s traffic is first transmitted via satellite to Ashburn and then through the terrestrial network to the server in Paris, with a total latency of 50.3ms. If the user selects the ground station in London as the anchor point, the total latency can be reduced to 26.8ms, a reduction of 44%.

举一个具体的例子，考虑一个位于大西洋（42.2° N, 60.0° W）的用户通过Starlink星座内的不同路径访问位于巴黎的服务器，如表I所示。当建立PDU会话时，用户选择了位于美国阿什本（Ashburn）的最近地面站（GS）作为锚点。因此，用户的流量首先通过卫星传输至阿什本，然后再经由地面网络到达巴黎的服务器，总延迟为50.3ms。如果用户选择位于伦敦的地面站作为锚点，总延迟则可以降低至26.8ms，延迟降幅达44%。

To address the issue of circuitous routing in mobile satellite networks, a straightforward method is to place the anchor point at the user satellite access point. This involves deploying a fully functional UPF on each satellite in regenerative mode. Users are provided with user plane services by the anchor point on the connected satellite, thereby avoiding detours caused by anchor points deviating from the shortest path.

为解决移动卫星网络中的路由迂回问题， 一个直接的方法是将锚点设置在用户的卫星接入点上。这涉及到在再生模式下的每颗卫星上都部署一个全功能的UPF 。用户由其所连接的卫星上的锚点来提供用户平面服务，从而避免因锚点偏离最短路径而导致的绕行。

However, this design faces frequent anchor point reselection. When a base station handover occurs due to user or satellite movement, the anchor point is also reselected. Since the anchor point remains unchanged throughout the PDU session, this reselection means that users need to release the current PDU session and establish a new one. During the session reestablishment period, users experience an average service interruption of several hundred milliseconds, in addition to the interruption caused by the base station handover. More critically, session reestablishment can lead to the reassignment of the user’s IP address [12], causing interruptions in services that rely on connections. Considering the high-speed movement of LEO satellites, the reselection of anchor point occurs every 25 minutes [14], significantly impacting the service continuity for users. Therefore, this method is not a reasonable solution to the circuitous routing problem of mobile satellite networks.

然而，这种设计面临着频繁的锚点重选问题。当由于用户或卫星的移动而发生基站切换时，锚点也随之被重选。由于在整个PDU会话期间锚点是保持不变的，这种重选意味着用户需要释放当前的PDU会话并建立一个新的会话。 在会话重建期间，除了由基站切换本身引起的中断外，用户还会经历平均数百毫秒的服务中断。更关键的是，会话重建可能导致用户的IP地址被重新分配，从而中断依赖于连接的服务。考虑到LEO卫星的高速移动，锚点的重选大约每2-5分钟就会发生一次，这严重影响了用户的服务连续性。因此，该方法 并非解决移动卫星网络路由迂回问题的合理方案。

Design Overview

Based on the above discussion, the essence of the circuitous routing problem in mobile satellite networks lies in the reliance on a fixed anchor point per PDU session and the difficulty of deploying the anchor point on satellites. This reliance directly leads to the issue where the anchor point is often not on the fastest path from the base station to the user’s target server, resulting in additional latency. To address this, a natural approach is to expand the number of available anchor points in a single session and select different anchor points based on the actual target of the user’s traffic to avoid detours.

根据以上讨论，移动卫星网络中路由迂回问题的本质在于每个PDU会话对固定锚点的依赖以及将锚点部署在卫星上的困难。这种依赖性直接导致了锚点常常不位于从基站到用户目标服务器的最快路径上，从而产生额外延迟。为解决此问题，一个自然思路是 在单次会话中扩展可用锚点的数量，并根据用户流量的实际目标来选择不同的锚点以避免绕行。

In this paper, we propose SkyOctopus, a multi-anchor mobile satellite network architecture that enables users to have multiple available anchor points distributed globally within a single PDU session. In SkyOctopus, multiple UPFs are deployed as anchor points at ground stations. On one hand, by introducing the Satellite UPF (S-UPF), user traffic can be forwarded to different anchor points based on its target IP addresses. This design keeps the anchor points on the ground while moving the traffic classifiers to the satellite, thereby avoiding the circuitous routing problem and the frequent anchor point reselection issue caused by satellite mobility. On the other hand, by redesigning the session establishment process, the S-UPF can establish connections with multiple anchor points simultaneously, avoiding the issue of prolonged session establishment times caused by the insertion-based process.

在本文中，我们提出了 SkyOctopus，一个多锚点移动卫星网络架构，它能让用户在单个PDU会话中拥有多个全球分布的可用锚点。在SkyOctopus中，多个UPF作为锚点被部署在地面站。一方面，通过引入卫星UPF（S-UPF），用户流量可以根据其目标IP地址被转发至不同的锚点。这种设计将锚点保留在地面，同时将流量分类器移至卫星，从而既避免了路由迂回问题，也规避了由卫星移动性引发的频繁锚点重选问题。另一方面，通过重新设计会话建立流程，S-UPF可以同时与多个锚点建立连接，避免了“基于插入的流程”所导致的会话建立时间过长的问题。

However, there are two main challenges to applying this architecture. The first challenge is the anchor point selection problem. Although SkyOctopus allows for proactive routing selection, it is difficult to ensure reasonable anchor point selection to minimize end-to-end latency, considering the diversity of user targets and the mobility of satellites. The second challenge is the anchor point distribution problem, which involves determining the optimal locations for anchor points. Given the deployment and connection costs of UPFs, operators can hardly deploy anchor points without limitations. Therefore, it is necessary to strategically select their deployment locations given a fixed number of anchor points.

然而，应用此架构存在两个主要挑战。第一个挑战是锚点选择问题。尽管SkyOctopus允许主动的路由选择，但考虑到用户目标的多样性和卫星的移动性，如何确保合理的锚点选择以最小化端到端延迟是一个难题。第二个挑战是锚点部署问题，即决定锚点的最优部署位置。考虑到UPF的部署和连接成本，运营商几乎不可能无限制地部署锚点。因此，在锚点数量固定的情况下，有必要策略性地选择其部署位置。

To address the first challenge, we propose a fine-grained anchor point selection strategy. The S-UPF uses PDRs based on the mapping of IP addresses to geographical locations to determine the initially chosen anchor point for users. Additionally, the path update mechanism ensures that the S-UPF can always select the optimal anchor point for users, even when network conditions change. The mechanism evaluates both intra-network and inter-network conditions to ensure users experience end-to-end low-latency access.

To tackle the second challenge, we analyze the anchor point deployment problem and prove it is an NP-hard problem. Based on this, we propose a greedy algorithm for selecting deployment locations for a fixed number of anchor points.

为应对第一个挑战，我们提出了一种细粒度的锚点选择策略。S-UPF利用基于IP地址到地理位置映射的PDR来为用户决定初始选择的锚点。此外，路径更新机制确保了即使在网络状况变化时，S-UPF也总能为用户选择最优的锚点。该机制会评估网络内部和网络之间的状况，以保证用户体验到端到端的低延迟接入。

为解决第二个挑战，我们分析了锚点部署问题，并证明了它是一个NP难问题。基于此，我们提出了一种贪心算法，用于在锚点数量固定的情况下选择部署位置。

TL; DR

多锚点架构, 设计很值得借鉴 (inspired by I-UPF)

(1) 架构: "决策上天、处理落地"

S-UPF是核心机制, 这个天上的调度中心本身不处理包裹，但它会查看每一个包裹的目的地，然后决定这个包裹应该交由哪个地面的“区域分仓”（多个作为锚点的UPF）来处理是最高效的

1. S-UPF:
   • 天上的调度中心, 全局视野, 避免 路径弯曲
2. Multi-UPF: (geo distributed)
   • 区域分仓, 都在地面上, 位置固定，规避了 频繁会话重建 的问题

(2) 联系: 高效并行式建立连接

在任务开始时，天上的 S-UPF 会向地面的所有“区域分仓” 并行 + 一次性 地发送请求，建立好连接