RELATED WORK¶

We briefly discuss the related works in this section.

Replica placement and request assignment in CDNs. A considerable number of previous research has studied the replica placement or request assignment problem in cloudbased CDNs [27], [40], [50], [53], [54]. GRP [27] is a replica placement scheme in Cloud CDNs that minimizes the cost incurred on the CDN providers, while satisfying the QoS requirements. CostTLO [53] exploits the use of issuing redundant requests to fetch contents from different cloud storage nodes. Similarly, TailCutter [28] explores the idea of sending multiple requests in parallel to optimize the content access latency while meeting the cost constraint. NetSession [57] is a peer-assisted CDN system that leverages both dedicated, centrally managed infrastructures and clients to distribute contents. NetSession can deliver several of the key benefits of both infrastructure-based and peer-to-peer CDNs. More recently, authors of [50] proposed algorithms that provision CDN traffic to cache servers and minimize the midgress which is the cache miss traffic between the CDN’s server and the content provider origins. Authors in [39] and [55] proposed mechanisms to leverage on-satellite caches to place contents and serve terrestrial users globally. All these existing efforts use either cloud caches or satellite caches alone for content distribution.

The fundamental difference between STAR FRONT and the above existing works is that: to our best knowledge, STAR FRONT is the first content distribution framework cooperatively exploiting network and storage resources on clouds (static) and LEO satellites (dynamic) to optimize the content access latency globally. Different from prior efforts (e.g., NetSession), STAR FRONT characterizes the high mobility of LEO satellites, and leverages their predicted trajectory to assist the calculation of content placement and request assignment decisions. As shown in the evaluation (§VI), through exploiting LEO satellites equipped with storage and high-datarate ISLs, STAR FRONT can improve the content access latency of global CDNs in a cost-effective manner, especially for the large population in remote areas.

Exploring “NewSpace” satellite networks. Our work is inspired by a number of recent studies on characterizing and understanding the network performance of “NewSpace” constellations [22], [29], [30], [33], [34], [35], [36], [41], [42], [45], [46], [52]. Multiple works have studied the routing algorithm in satellite networks. For example, to understand the latency properties of LEO satellite networks, the author in [34] builds a simulator to evaluate how to use the laser links to enable low-latency communication over satellites. Similarly, a ground-relay based routing strategy is proposed in [35]. Motifs [23] is proposed to dynamically build network connectivities in space, tackling the high dynamism in LEO satellites. Authors in [56] studied the impact of space-ground topology design on the attainable network performance (e.g., latency and jitter) of integrated satellite and terrestrial networks. These efforts complement our work.

In-orbit computing. Recent works also have explored the feasibility of leveraging the improved computation capability in emerging nanosatellite constellations to build new computer systems [22], [31], [43] in space. Authors in [22] qualitatively examine the opportunities and challenges of in-orbit computing. An orbital edge computing (OEC) [31] architecture is proposed to address the limitation of existing “bent-pipe” architecture and optimize the edge processing latency. Instead of focusing on nanosatellites only, our work proposes a cooperatively architecture that jointly exploits the resource on both terrestrial cloud data centers and LEO satellites to judiciously improve the performance of content distribution on a global scale.

Emerging satellite applications. OrbitCast [48] is a hybrid and low-latency data delivery architecture for various earth observation (EO) tasks. SpaceRTC [47] extends existing realtime communication (RTC) applications, which are conventionally built upon cloud platforms, to mega-constellation to achieve low-latency wide-area communication. Different from these existing efforts above, STAR FRONT targets at a different application scenario with new problem formulations. We believe all these works can jointly facilitate the development of futuristic satellite Internet ecosystem.

CDN 中的副本放置和请求分配。大量先前的研究已经研究了基于云的 CDN 中的副本放置或请求分配问题。GRP 是一种云 CDN 中的副本放置方案，可在满足 QoS 要求的同时最大限度地降低 CDN 提供商产生的成本。CostTLO 利用发出冗余请求从不同的云存储节点获取内容。同样，TailCutter 探索了并行发送多个请求以优化内容访问延迟，同时满足成本约束的想法。NetSession 是一种对等辅助 CDN 系统，它利用专用、集中管理的设施和客户端来分发内容。NetSession 可以提供基于设施和对等 CDN 的一些关键优势。最近，有人提出了一种算法，该算法将 CDN 流量配置到缓存服务器，并最大限度地减少中间流量，即 CDN 的服务器和内容提供商源站之间的缓存未命中流量。和的作者提出了利用卫星缓存来放置内容并为全球地面用户提供服务的机制。所有这些现有工作都仅使用云缓存或卫星缓存进行内容分发。

STAR FRONT 与上述现有工作之间的根本区别在于：据我们所知， STAR FRONT 是第一个协同利用云（静态）和 LEO 卫星（动态）上的网络和存储资源来优化全球内容访问延迟的内容分发框架。与先前的努力（例如，NetSession）不同，STAR FRONT 描述了 LEO 卫星的高移动性，并利用其预测的轨迹来辅助内容放置和请求分配决策的计算。如评估（§VI）所示，通过利用配备存储和高速率 ISL 的 LEO 卫星，STAR FRONT 可以经济高效地提高全球 CDN 的内容访问延迟，特别是对于偏远地区的大量人口。

TL;DR

现有工作都仅使用云缓存或卫星缓存进行内容分发

STAR FRONT 协同利用云（静态）和 LEO 卫星（动态）上的网络和存储资源来优化全球内容访问延迟

基于LEO的高移动性
利用其预测的轨迹来辅助内容放置和请求分配决策的计算

探索“新太空”卫星网络。我们的工作受到最近许多关于描述和理解“新太空”星座网络性能的研究的启发。多项工作研究了卫星网络中的路由算法。例如，为了了解 LEO 卫星网络的延迟特性，有人构建了一个模拟器来评估如何使用激光链路来实现卫星上的低延迟通信。类似地，也有人提出了一种基于地面中继的路由策略。Motifs 被提出来动态地构建空间中的网络连接，以应对 LEO 卫星中的高动态性。也有作者研究了空间-地面拓扑设计对集成卫星和地面网络的可实现网络性能（例如，延迟和抖动）的影响。这些努力补充了我们的工作。

在轨计算。最近的工作还探索了利用新兴的纳米卫星星座中改进的计算能力来构建空间中的新计算机系统的可行性。有作者定性地考察了在轨计算的机遇和挑战。提出了一种轨道边缘计算（OEC）架构，以解决现有“弯管”架构的局限性并优化边缘处理延迟。我们的工作不仅关注纳米卫星，还提出了一种协同架构，该架构共同利用地面云数据中心和 LEO 卫星上的资源，以明智地提高全球范围内内容分发的性能。

新兴卫星应用。OrbitCast 是一种用于各种地球观测（EO）任务的混合且低延迟的数据传输架构。SpaceRTC 将现有的实时通信（RTC）应用程序（通常建立在云平台上）扩展到大型星座，以实现低延迟的广域通信。与上述这些现有工作不同，STAR FRONT 针对具有新问题公式的不同应用场景。我们相信所有这些工作都可以共同促进未来卫星互联网生态系统的发展。