Discussion && Conclusion

The premise of in-orbit computing as a service may certainly seem outlandish. Our goal is not to push a strongly affirmative case for it; we only suggest that it might be worth not dismissing casually, as it has unusual and interesting properties that some applications may benefit from.

In-orbit computing could offer low-latency access to compute from anywhere on Earth, and enable a new abstraction: GEO-like stationarity without the GEO latency penalty. These properties open up potentially new use cases, including for edge computing, meetup servers for multi-user interaction, and opportunistic data analytics for space-native data.

However, in-orbit compute would be many times more expensive than terrestrial data center compute, and would only be useful in settings where the latter is limited due to either access latency, or bottlenecks in downloading satellite-generated data. Terrestrial infrastructure could potentially also expand to address the latency issue. For instance, an ambitious effort is currently exploring the use of off-shore container-based data centers [38]; coincidentally, a key barrier for such data centers is also power provisioning.

The types of applications that could benefit from in-orbit LEO compute are also limited on another front: for some settings where terrestrial data center infrastructure is limiting, GEO satellites are perfectly acceptable, because latency is not an issue. One such example is video broadcast, a common application of GEO satellites. It is unlikely that serving video through LEO satellites would be worthwhile.

Our analysis of the issues of weight, volume, space hardening, and life-cycle, indicates that these are not prohibitive problems. Power is perhaps the biggest impediment, and it is unclear if the ∼20% overhead from a beefy server would be acceptable, given the substantial modifications it could require to a satellite’s power provisioning.

Weather, which we did not analyze yet, also poses limitations on availability: LEO network interruptions due to weather attenuation on the ground-satellite links would make in-orbit compute temporarily unavailable from the affected locations.

Lastly, while we attempted to inform ourselves broadly of the challenges and downsides, there may be “unknown unknowns” that could potentially make in-orbit compute entirely infeasible. We hope that our readers will either raise such objections, or suggest use cases that we did not foresee.

在轨计算即服务的构想当然可能显得异想天开。我们的目标并非为其强力地推崇；我们仅是建议，这个想法或许不应被轻易地忽视，因为它具有一些非同寻常且有趣的特性，某些应用可能会从中受益。

在轨计算可以提供从地球任何地方到计算资源的低延迟访问，并能实现一种新的抽象：类似GEO的静止特性，且没有GEO的高延迟代价。这些特性开启了潜在的新用例，包括边缘计算、用于多用户交互的汇合点服务器，以及针对空间原生数据的机会性数据分析。

然而，在轨计算的成本将比地面数据中心高出数倍，并且仅在后者的使用受限于访问延迟或下载卫星生成数据的瓶颈时才有用。地面基础设施也可能通过扩展来解决延迟问题。例如，一项宏大的计划目前正在探索使用离岸集装箱式数据中心[38]；巧合的是，这类数据中心的一个关键障碍也是电力供应。

可能受益于在轨LEO计算的应用类型在另一方面也受到限制：对于某些地面数据中心基础设施受限的场景，GEO卫星是完全可以接受的，因为延迟不是问题。视频广播就是一个这样的例子，它是GEO卫星的常见应用。通过LEO卫星提供视频服务不太可能具有价值。

我们对重量、体积、空间加固和生命周期等问题的分析表明，这些并非无法克服的难题。电力或许是最大的障碍，目前尚不清楚一台高性能服务器带来的约 \(20\)% 的额外开销是否可以接受，因为它可能需要对卫星的电力供应系统进行重大修改。

我们尚未分析的天气因素，也对可用性构成了限制：由于地-星链路上的天气衰减导致的LEO网络中断，将使得受影响地区的在轨计算暂时不可用。

最后，尽管我们试图广泛了解其中的挑战与弊端，但可能存在一些“未知的未知因素”，它们或将使得在轨计算完全不可行。我们希望我们的读者能提出此类反对意见，或建议我们未曾预见到的应用场景。