Concluding Discussion

Delivering data from multiple constellations: Recent years have seen emergence of the Ground-station-as-a-service (GSaaS) model by many commercial entities [3, 31, 36]. These companies allow constellation operators to rent ground station time by the minute to schedule data download. We expect withhold scheduling to be effective in such contexts. However, the measurement of network queue size needs to be indirect, as the queue size at the ground station may not be visible to satellite constellation operators. We note that UQE will get worse if the backhaul bandwidth of ground stations decrease, as shown in our evaluation. Therefore, we expect Umbra to be more efficient under the GSaaS scenario.

处理来自多个星座的数据 🛰️

近年来，许多商业实体推出了地面站即服务 (Ground-station-as-a-service, GSaaS) 的模式 [3, 31, 36]。这些公司允许星座运营商按分钟租用地面站时间来安排数据下载。我们预期暂缓调度 (withhold scheduling) 在此类场景下同样有效。然而， 网络队列长度的测量需要通过间接方式进行，因为地面站的队列状态可能对卫星星座运营商并不可见。 我们注意到，如我们的评估所示， 如果地面站的回程带宽 (backhaul bandwidth) 下降，队列的无序演化 (UQE) 问题会变得更糟。 因此，我们预期Umbra在GSaaS场景下会更高效。

Inter-station and inter-satellite links: Our evaluation assumed the absence of these links because they are not common in today’s deployments. However, both these kinds of links can be added to our graph and our TEN-based solution (Section 3) would still generate a solution. For instance, a satellite could route data through another satellite using an inter-satellite link, especially when the latter satellite is connected to a low-queue station.

站间与星间链路 🔗

我们的评估假设不存在这些链路，因为它们在当今的部署中尚不普遍。然而，这两种链路都可以被添加到我们的图中，而我们基于时间扩展网络 (TEN) 的解决方案（第3节）仍然能生成一个有效的方案。例如，一颗卫星可以通过星间链路 (inter-satellite link) 将数据路由到另一颗卫星，特别是当后者连接到一个队列负载较低的地面站时。

What did not work? To make withholding decisions, we experimented with iterating between: (a) identifying the best matching between a satellite and ground station at a given time instance, and (b) computing the max flow in the time expanded network. In principle, this was reasonable because if a satellite withholds data from a ground station, a different satellite may want to use this ground station (this decision can be made in the next iteration). However, we noticed that the scheduling objective (e.g. throughput) showed little improvement beyond more than one iteration, and only increased computation cost. We believe that this is because most Dove satellites follow each other in one of two orbits and come in contact with the same ground station sequences. So, withholding decisions are similar for different satellites in proximity with a crowded ground station.

未能奏效的尝试 💡

为了制定暂缓决策，我们曾尝试在以下两步之间进行迭代：(a) 在给定时间点，识别卫星与地面站之间的最佳匹配；(b) 在时间扩展网络中计算最大流。原则上，这种方法是合理的，因为如果一颗卫星决定暂缓向某个地面站传输数据，那么另一颗卫星或许可以利用这个空闲的地面站（这个决策可以在下一次迭代中做出）。然而，我们注意到，经过一次以上的迭代后，调度目标（如吞吐量）几乎没有改善，反而只增加了计算开销。我们认为，这是因为大多数“鸽群”(Dove)卫星在两条轨道上彼此跟随，并以相同的顺序接触地面站序列。因此，对于邻近某个拥挤地面站的不同卫星而言，它们的暂缓决策是相似的。

Ground Station Backhaul Bandwidths: Our work explores the ground station-cloud backhaul link as the bottleneck in satellite data transfers. Over the next few years, we expect three factors to increase the demand for backhaul bandwidth even more: (i) satellite-ground bandwidths continue to improve to 5 Gbps and beyond ([41, 49]), (ii) operators will amortize the cost of site acquisition and licensing by deploying more antennas at the same ground station site, and (iii) increased computation demands at the ground stations (for pre-processing), will cause queuing delays towards the cloud to continue increasing, even if backhaul bandwidths keep improving.

地面站回程带宽的未来趋势 📈

我们的工作探讨了 "地面站-云端" 回程链路作为卫星数据传输中的瓶颈。我们预计，未来几年有三个因素将进一步增加对回程带宽的需求：

(i) 星地通信带宽持续提升至5 Gbps甚至更高 ([41, 49])

(ii) 运营商将通过在同一地面站部署更多天线来分摊场地购置与许可的成本

(iii) 地面站的计算需求增加（用于数据预处理），这将导致朝向云端的排队延迟持续增加，即便回程带宽本身在不断提升

Umbra and Resource Utilization Efficiency: While provisioning higher backhaul bandwidth can always reduce queues on the ground stations and reduce the latency, it will not solve the problem of the low resources utilization. We note that the constellation sizes continue to grow over time. Umbra can support more satellite with the current infrastructure, as well as use less bandwidth to support emerging constellations. Moreover, in the GSaaS setting, the ground station capacity and bandwidth needs to be rented from the provider, and increasing its efficiency has economic benefits.

Umbra 与资源利用效率 💰

虽然配置更高的回程带宽总能减少地面站的队列并降低延迟，但这并不能解决资源利用率低下的问题。我们注意到，卫星星座的规模正随着时间持续增长。Umbra能够用现有基础设施支持更多卫星，或使用更少的带宽来支持新兴星座。此外，在GSaaS的场景下，地面站的容量和带宽需要向供应商租用，因此提升其使用效率将直接带来经济效益。