Other Related Work

In addition to prior solutions introduced in §II-C, we discussed other efforts related to our study in this paper.

除了在第二节C部分介绍的先前解决方案外，我们在本节讨论与本研究相关的其他工作

Multi-path routing. Plenty of research focuses on multipath routing and load balancing. In data center networks, ECMP [19] is commonly deployed to achieve good load balancing by distributing traffic equally over multiple paths with the same cost using a simple round-robin pattern. But the overall throughput is limited by the path with minimum capacity. WCMP [52] was proposed to distribute load flexibly to alleviate the problem of ECMP. The paths used by one source are not guaranteed to be disjoint, so the competition of bandwidth within one source may happen especially in such a meshlike satellite network. In mobile ad-hoc networks, a number of multipath routing protocols are proposed in order to increase the reliability while transmitting data (e.g.,, fault tolerance) or load balancing. Two presentative protocols are Split Multipath Routing (SMR) [27], Ad hoc On-demand Multipath Distance Vector (AOMDV) [30]. SMR is designed to find maximally disjoint paths between one source and destination pair by exploiting dynamic source routing. AOMDV is an extension to the AODV [33] protocol for computing loop-free and linkdisjoint paths. A recent work [43] designed a similar multipath routing protocol but combined it with network coding to achieve better performance of multipath EO data transmission in satellite networks. However, none of them considers the path relation among different sources. Path contention undermines the advantages of multipath transmission. Our routing scheme improves the multipath transmission performance by considering the path correlation among different sources.

大量的研究聚焦于多路径路由和负载均衡。在数据中心网络中，等价多路径路由（ECMP）[19]被广泛部署，通过简单的轮询模式将流量均等地分配到多条成本相同的路径上，以实现良好的负载均衡。但其总吞吐量受限于容量最小的路径。为缓解ECMP的问题，加权多路径路由（WCMP）[52]被提出以更灵活地分配负载。然而，这些方法不保证同一源所使用的路径是无交集的（disjoint），因此源内的带宽竞争可能会发生，尤其是在类似我们研究的网格状卫星网络中。

在移动自组织网络（mobile ad-hoc networks）中，研究人员提出了多种多路径路由协议，旨在提高数据传输的可靠性（例如，容错性）或实现负载均衡。两个代表性协议是分裂多路径路由（SMR）[27]和Ad hoc按需多路径距离矢量路由（AOMDV）[30]。SMR旨在通过利用动态源路由，在单个源和目的节点对之间找到最大程度无交集的路径。AOMDV是AODV [33]协议的一个扩展，用于计算无环和链路无交集的路径。近期的一项工作[43]设计了一个类似的多路径路由协议，但将其与网络编码相结合，以在卫星网络中实现更好的多路径EO数据传输性能。

然而，上述所有工作均未考虑不同源之间的路径关联。路径竞争会削弱多路径传输的优势。我们的路由方案通过考虑不同源之间的路径相关性，提升了多路径传输的性能。

Co-flow scheduling. Many coflow scheduling works that aim to minimize the coflow completion time are proposed in data center networks. [51] proposes a coflow-aware network optimization framework that seamlessly integrates routing and scheduling for better application performance. [29] studies the routing and scheduling of multiple coflows to minimize the average coflow completion time. [38] also proposes algorithms to deal with single coflow scheduling and multiple coflow scheduling problems. However, there are two main differences between the coflow scheduling problem in data center networks and our CEOMD problem. Firstly, unlike data center networks, the topology of LEO satellite networks changes with time since the high mobility of satellites. Second, we exploit multipath for each flow. All of these contribute to the difficulty to solve the CEOMD problem.

在数据中心网络中，许多旨在最小化协流完成时间（Coflow Completion Time）的协流调度工作被提出。文献[51]提出了一个感知协流的网络优化框架，该框架无缝集成了路由和调度，以获得更好的应用性能。文献[29]研究了多个协流的路由和调度问题，以最小化平均协流完成时间。文献[38]也提出了处理单个协流调度和多个协流调度问题的算法。

然而，数据中心网络中的协流调度问题与我们的CEOMD问题存在两个主要区别。

首先，与数据中心网络不同， 由于卫星的高机动性，LEO卫星网络的拓扑结构是随时间动态变化的。

其次，在我们的问题中，我们 为每个流都利用了多路径传输。 所有这些因素都增加了求解CEOMD问题的难度。