Downlink-First or Orbit-First: The Buffer Horizon That Decides What Survives
An Earth observation satellite that cannot reach the ground keeps imaging, and everything it captures during that gap goes to a recorder that will eventually fill up or fail. You have two choices for that recorder. It can stay a private buffer on each satellite until the next pass, or it can become one shard of a distributed object store spread across the optical crosslink mesh. Both designs work on a good day, and they split on the day a ground station goes dark.
Two ways to hold data nobody has received yet
Downlink-first treats each spacecraft as an isolated store-and-forward node. Payload data lands on the solid-state recorder, gets packaged under space data systems standards such as CCSDS file delivery and the delay-tolerant bundle protocol, and leaves the satellite only when a ground contact opens. The ground segment is the system of record, so durability becomes a ground problem, handled with ordinary multi-region replication. Orbit-first moves the durability boundary up into orbit: each acquisition is erasure-coded and its fragments are pushed over intersatellite links to neighbours, so the object survives the loss of the satellite that captured it before any byte reaches Earth. The price is crosslink capacity, power for laser terminals that would otherwise sit idle, and a metadata plane in orbit that has to know where every fragment lives.
The failure each design actually fears
In downlink-first, the dangerous failure is a recorder or power fault on a satellite holding hours of undelivered imagery. No second copy exists, so the data is gone, and nobody notices until the expected pass returns nothing. Orbit-first removes that single point of failure but adds a subtler one: when a crosslink partition hits mid-stripe, some objects are left with fewer fragments than the code needs to rebuild them, and the system cannot tell whether a missing fragment is lost or just unreachable until the partition heals. If the fragment catalog sits on a few coordinator nodes, losing those nodes freezes repair across the whole shell, so orbit-first couples failures that downlink-first keeps independent. A well-built orbit-first system limits this in two ways. It keeps all fragments of any one object inside a single orbital plane, where link geometry is stable, and it replicates the catalog with the data instead of hosting it somewhere else.
Where the decision flips
The deciding parameter is the buffer horizon, in hours: usable recorder capacity divided by net ingest rate on the most heavily tasked satellite, where net means acquisition minus whatever downlink that satellite gets in a normal orbit. Compare it with the worst credible ground outage, which is rarely a single station dropping and usually something correlated: rain fade across a regional cluster of Ka-band or optical sites, a backhaul cut, or a software push that takes a whole ground network offline at once. My estimate is that most commercial ground networks should plan for correlated outages lasting several hours, not minutes. If the buffer horizon comfortably exceeds that outage, downlink-first wins, because the satellite simply waits, failure domains stay small and independent, and you never pay for an orbital metadata plane you would rarely use. If the outage can outlast the horizon, orbit-first wins outright, because spreading data across neighbours extends effective capacity and turns a single recorder fault from data loss into a repair job. Where the two numbers are close, pick orbit-first only if the crosslinks already exist for routing, since adding a laser mesh purely for durability rarely pays for itself.
The trade you still carry
Even when orbit-first wins, you take on a repair policy decision that downlink-first never forced on you: how long to wait for a partitioned fragment before spending crosslink bandwidth to rebuild it, a timer that trades wasted capacity after short outages against exposure to a second fault after long ones.
Technical FAQ: Can a hybrid avoid making this choice before freeze?
Partly. You can run downlink-first by default and switch a satellite to erasure-coded offload when its recorder crosses a fill threshold, but that only works if the orbital fragment catalog and repair logic already exist and are exercised in normal operations, because a durability path that runs for the first time during an outage is the one that fails.