Skip to main content

​
Orbit Scheduler

Coming Q2 2026 — This is a design preview. Request early access to be notified when available.

​
Overview

The Orbit Scheduler orchestrates workloads across heterogeneous compute nodes spanning Earth datacenters and orbital infrastructure. It understands orbital mechanics, energy availability, and network topology to make optimal placement decisions.

​
Key Capabilities

  • Orbit-aware scheduling — Accounts for orbital position, eclipse periods, ground contacts
  • Energy optimization — Routes work based on power availability
  • Latency-aware routing — Minimizes round-trip time based on geometry
  • Fault tolerance — Automatic failover between nodes
  • Workload splitting — Distribute work across Earth + orbit

​
Architecture

                    ┌─────────────────────┐
                    │   Orbit Scheduler   │
                    │                     │
                    │  ┌───────────────┐  │
                    │  │ Placement     │  │
                    │  │ Engine        │  │
                    │  └───────────────┘  │
                    │  ┌───────────────┐  │
                    │  │ Orbit Model   │  │
                    │  └───────────────┘  │
                    │  ┌───────────────┐  │
                    │  │ Energy Model  │  │
                    │  └───────────────┘  │
                    └─────────┬───────────┘
                              │
        ┌─────────────────────┼─────────────────────┐
        ▼                     ▼                     ▼
  ┌──────────┐         ┌──────────┐         ┌──────────┐
  │ Earth DC │         │ LEO-1    │         │ LEO-2    │
  │ us-west  │         │ 550km    │         │ 550km    │
  └──────────┘         └──────────┘         └──────────┘

​
API Preview

​
Submit Job with Placement Hints

from rotastellar import RotaStellar

client = RotaStellar(api_key="rs_...")

job = client.runtime.submit(
    model="llama-70b",
    prompt="...",
    placement={
        "prefer": "orbital",           # Prefer orbital nodes
        "fallback": "earth",           # Fall back to Earth if needed
        "region_affinity": "europe",   # Prefer nodes with Europe visibility
        "max_hops": 2                  # Max ISL hops
    },
    constraints={
        "latency_sla_ms": 100,
        "energy_budget_wh": 0.5
    }
)

print(f"Placed on: {job.node}")
print(f"Reason: {job.placement_reason}")

​
Query Node Status

# Get current node availability
nodes = client.runtime.nodes()

for node in nodes:
    print(f"{node.id}: {node.type}")
    print(f"  Location: {node.location}")
    print(f"  Status: {node.status}")
    print(f"  Available power: {node.available_power_w}W")
    print(f"  Queue depth: {node.queue_depth}")

    if node.type == "orbital":
        print(f"  In eclipse: {node.in_eclipse}")
        print(f"  Ground contact: {node.has_ground_contact}")

​
Schedule Future Work

Schedule jobs to run at optimal times: 
# Schedule for optimal conditions
scheduled_job = client.runtime.schedule(
    model="llama-70b",
    prompt="...",
    schedule={
        "window_start": "2026-01-22T00:00:00Z",
        "window_end": "2026-01-22T12:00:00Z",
        "optimize_for": "energy"  # or "latency", "cost"
    }
)

print(f"Scheduled for: {scheduled_job.scheduled_time}")
print(f"Expected node: {scheduled_job.expected_node}")
print(f"Energy savings: {scheduled_job.energy_savings_percent}%")

​
Placement Strategies

​
Orbital-First

Prefer orbital nodes, fall back to Earth: 
job = client.runtime.submit(
    model="...",
    prompt="...",
    placement={"prefer": "orbital", "fallback": "earth"}
)

​
Earth-First

Prefer Earth, use orbital for overflow: 
job = client.runtime.submit(
    model="...",
    prompt="...",
    placement={"prefer": "earth", "fallback": "orbital"}
)

​
Latency-Optimized

Route to minimize latency to specific region: 
job = client.runtime.submit(
    model="...",
    prompt="...",
    placement={
        "optimize_for": "latency",
        "user_location": {"lat": 51.5, "lon": -0.1}  # London
    }
)

​
Energy-Optimized

Route to nodes with best energy availability: 
job = client.runtime.submit(
    model="...",
    prompt="...",
    placement={"optimize_for": "energy"}
)

​
Scheduling Factors

The scheduler considers:
FactorWeightDescription
Energy availabilityHighCurrent and predicted power
LatencyHighNetwork path to user
Queue depthMediumCurrent load on node
Thermal headroomMediumTemperature margin
Eclipse statusMediumUpcoming power constraints
Ground contactLowCommunication availability

​
Node Types

TypeLocationCharacteristics
earthTerrestrial DCUnlimited power, stable network
leoLow Earth OrbitVariable power, intermittent ground contact
meoMedium Earth OrbitStable power, higher latency
geoGeostationaryContinuous visibility, 240ms+ latency

​
Next Steps

Adaptive Runtime

Learn about energy-aware inference

Resilient Compute

Learn about fault tolerance