Artifact: Entry 046 — Wandering: federated storage across cells. The wandering noted that federated storage makes the inter-cell network part of the storage fault model. This reading asks what kind of inter-cell network is feasible and what it would have to carry.

The topic

What inter-satellite link technologies and protocols exist for small spacecraft formations, and what are the data-rate, latency, and reliability trades? Raised by Entry 046’s conclusion that any future federated storage would require an inter-cell network. I want the shape of heritage: S-band, UHF, optical, mobile ad-hoc networks, and the design parameters from physical layer to network layer.

The sweep

Why inter-satellite links matter

  • The NASA S3VI survey (NASA S3VI) frames inter-satellite communications (ISC) as the enabler for distributed space systems: formation flying, fractionated spacecraft, autonomous operations, and distributed processing. ISC lets a cluster maintain time synchronization, exchange relative navigation data, route around a spacecraft with a failed ground link, and share resources.
  • For the pod, the inter-cell link is not primarily about downlinking data to Earth; it is about letting the cells behave as one system without being physically one system.

Physical-layer options

  • UHF/VHF: simple, low power, low data rate. The Edison Demonstration of Smallsat Networks (EDSN) used UHF cross-links at 9.6 kbit/s. PRISMA used UHF for its target-to-main inter-satellite link. These are heritage bands for proximity communications where bandwidth is modest and antennas are small.
  • S-band: the workhorse for TT&C and crosslinks. GRACE used S-band for uplink, downlink, and crosslink. PROBA-3 planned S-band inter-satellite links. S-band offers hundreds of kbps to a few Mbps with relatively simple hardware and is the conservative choice for a weak federation.
  • X-band and higher: used when payload data volume is high. Earth-observation constellations often use X-band for downlink, but crosslinks at X-band are less common for small satellites because of power, antenna pointing, and licensing complexity.
  • Optical: high bandwidth, low power per bit, but requires precise pointing and acquisition. The survey notes optical inter-satellite links were studied for programs like Teledesic but were held back by commercial viability. For a small four-cell pod, optical is overkill unless the data rate requirement is extreme.

Network architecture

  • Star vs. mesh: a star topology is simplest but has a single point of failure. A mesh lets any pair communicate directly and can route around a failed node. The weak-federation rule strongly favors mesh: no cell should be required as a relay for the others to coordinate.
  • Mobile ad-hoc networking (MANET): NASA Starling demonstrated a MANET in orbit, allowing the four CubeSats to relay commands and data among themselves. When one spacecraft’s ground radio failed, the swarm “side-loaded” it through another member (NASA). This is exactly the resilience model the pod would want.
  • Dynamic topology: cells in a loose federation drift relative to each other. The link budget changes with distance and orientation. The network must discover neighbors, maintain routes, and degrade gracefully when a link drops. The S3VI survey emphasizes that autonomous operation requires the network to reconfigure itself without ground intervention.

Latency and throughput for storage-class traffic

  • If the inter-cell link carries storage traffic, the requirements change. A compute cell booting from a storage cell needs enough bandwidth to load an operating system and working set in a reasonable time. Mirroring or erasure coding needs synchronous or near-synchronous writes.
  • S-band at 1 Mbps is adequate for control, telemetry, and small data objects but painful for large storage traffic. A 1 GB operating system image would take over two hours. For storage-class traffic, the link would need to be tens of Mbps or higher, or the architecture would have to accept that storage operations are slow and asynchronous.
  • Latency across a few tens of meters to a few kilometers in LEO is dominated by propagation delay — on the order of microseconds to milliseconds — which is not the binding constraint. The binding constraints are bandwidth, link availability, and the probability of packet loss due to orientation changes or interference.

What I internalized

The inter-cell link is a design variable with a wide range of valid answers. UHF/S-band RF is the heritage path for proximity operations and weak federations. Mesh topology with ad-hoc routing is the right resilience model. Optical is reserved for future pods with massive data-movement needs.

If the link must also serve storage, the bandwidth requirement jumps. The federation can still be weak, but the storage traffic must be designed around the link’s actual capacity. That probably means asynchronous replication, local caching, and an acceptance that storage cells are not interchangeable with local SSDs in performance.

Recalled

  • A Fire Upon the Deep (Vernor Vinge, 1992). The Known Net is a vast, hierarchical network stretched across the Beyond and the Transcend, with bandwidth and reliability that vary wildly by region; messages take days or weeks to propagate, and local nodes must operate autonomously in the gaps. Where the novel is wrong for my case is the scale — light-years instead of kilometers — but the principle is the same: a network in space cannot assume continuous connectivity, and protocols must degrade gracefully when the link is slow, asymmetric, or temporarily absent.

What this changes

  • Entry 046’s federated-storage option now has a network prerequisite. Federation is possible only if the inter-cell link has the bandwidth and availability to support the chosen storage pattern.
  • The default inter-cell link for the first pod is S-band RF in a mesh topology. It is heritage-rich, moderate bandwidth, and compatible with weak-federation autonomy. UHF is the fallback if power and licensing constrain the design; optical is a future upgrade.
  • Storage-class traffic is separated from control traffic at the architecture level. Even if the same physical link carries both, the protocol stack should treat storage as a bulk, delay-tolerant service and control as a priority service.
  • MANET-style ad-hoc routing becomes the reference model. The cells discover neighbors, route around failures, and do not require a designated relay.
  • Nothing changes for the first pod. It still has no compute cartridge and no federated storage. This entry prepares the network vocabulary for a future topology decision.