AI
Why Standard Kubernetes Breaks Beyond Earth Orbit
A Kubernetes cluster spanning Earth and Mars faces a 13-minute one-way signal delay. TCP’s three-way handshake alone costs 39 minutes, and the connection times out before it completes. Image pulls require a stable internet connection that deep-space links cannot provide. Disruptions are frequent, bandwidth is low, and there is no persistent end-to-end connectivity. Standard Kubernetes was built for data centers with stable, low-latency networks, not for satellites or rovers operating 13 light-minutes away.
How Delay-Tolerant Networking Fills the Gap
DTN replaces permanent end-to-end connections with hop-by-hop transfers. Nodes buffer data in memory or on persistent storage for an arbitrary amount of time, surviving power outages and communication blackouts. A bundle layer sits as an overlay network above heterogeneous underlying protocols, using convergence layers to translate between them. If node A can never reach node C directly, it sends the bundle to node B, which physically travels closer to node C and delivers it later.
The Dark Soul Project: DTN Meets Kubernetes
D3TN, Dresden University of Technology, and Cyrus Technology formed the Dark Soul project. The name stands for DTN Architectures for Space Operations and Space Cloud Life Cycle Management. The project targets three goals: a delay- and disruption-tolerant cloud infrastructure, secure multi-tenant separation for shared space hardware, and disruption-tolerant application operations. Each Kubernetes cluster stays local to one set of nodes with low-latency internal links. MicroDTN bridges the gaps between clusters across space links.
Helm-DTN: Shipping Applications to Mars
The Helm-DTN plugin intercepts a normal helm dtn install push call. It scans charts and values files, identifies all container images and other dependencies, downloads them to the local machine, rewrites registry references to point at a local in-cluster registry, and packs everything into a tar archive. MicroDTN instances on both sides bridge the space link using the bundle protocol. On the remote cluster, an application lifecycle management engine unpacks the archive, pushes images to the local registry, and calls helm install internally.
DGate: A Custom CRD for DTN Ingress and Egress
The Kubernetes Gateway API expects to be the first point where data enters a cluster, but MicroDTN receives bundles at the edge first. Mapping DTN-specific operations, bundle processing, store-carry-forward persistence, custody transfer, and opportunistic routing, into Gateway API listener and route models has no clean abstraction. D3TN built a custom resource definition called DGate (Dark Soul Gateway) instead. DGate registers agents at MicroDTN, forwards received data to the correct backend service, and lets applications reuse the same agent connection to send data back out.
What Works Now and What Comes Next
Application deployment and lifecycle management over DTN links work today. Clusters still cannot span multiple planets because Kubernetes components expect heartbeats far sooner than 13 minutes. Remaining work includes autonomous Kubernetes environment upgrades, tenant separation, and opportunistic cluster discovery for spacecraft on deep-space missions. A follow-up project plans to test the full stack on satellites deployed and operated by ESA to validate the approach in real space-terrestrial internetworking scenarios.
Notable Quotes
just for now, it’s basically impossible to have a Kubernetes cluster span Earth and Mars. Tobias Nöthlich · ▶ 02:28
if we try to do this over TCP, the average single trip latency is 13 minutes. TCP has a three-way handshake. So even if it weren’t working um this will be 39 minutes just for establishing a connection which would time out anyways Tobias Nöthlich · ▶ 12:11
MicroDtn is doing far more than just forwarding bytes. Uh it’s performing bundle processing, right? We need to unpack all the data structures from from the bundle protocol. Tobias Nöthlich · ▶ 18:05
Key Takeaways
- A 13-minute one-way delay makes TCP and standard Kubernetes image pulls unworkable over Mars links.
- Helm-DTN pre-bundles all container images locally before transmitting a single tar archive over MicroDTN.
- DGate, a custom CRD, handles DTN ingress and egress where the Kubernetes Gateway API cannot.
About the Speaker(s)
Tobias Nöthlich is a Research and Development Engineer at D3TN GmbH with a Master’s degree in Computer Science from Dresden University of Technology. He specializes in Delay- and Disruption-Tolerant Networks and has contributed to multiple DTN projects, including the work presented here on space cloud lifecycle management.