Databelt is a stateful serverless framework for the Edge-Cloud-Space 3D Compute Continuum. It introduces a Service Level Objective (SLO)-aware function state propagation mechanism and a function state fusion strategy to reduce latency and increase workflow efficiency in dynamic, satellite-powered environments.

• SLO-Aware State Propagation: Proactively migrates function states to optimal nodes based on network latency, node availability, and execution constraints.
• Function State Fusion: Co-locates related states within a shared sandbox to reduce redundant network and storage operations.
• Optimized Scheduling: Incorporates a distributed, pluggable scheduler that considers satellite availability, power, compute limits, and temperature.
• Edge + Space Ready: Lightweight implementation in Rust using WasmEdge, designed for constrained LEO satellites and edge nodes.

• 🚀 Up to 66% reduction in end-to-end workflow latency.
• ⚡ Up to 50% higher throughput compared to stateless cloud-native models.
• 🌐 79% local state availability, reducing inter-node hops and latency.
• ✅ 0% SLO violations under tested dynamic conditions.

Databelt has a 3-Phase Architecture: Identify, Compute, and Offload phases ensure efficient data placement across space, edge, and cloud nodes.

Its main components include:

• Databelt Service: Central service managing topology, SLOs, and state placement decisions.
• Databelt Middleware: Middleware library embedded in functions for transparent state access and propagation.
• Databelt Scheduler: Distributed scheduler with network-aware heuristics and resource filters, compatible with Kubernetes/Knative.